Home

URL - of /pub - Freie Universität Berlin

1. 1400 140 J 1200 120 2 5 1000 100 5 5 5 800 Bis is Z g 2 600 60 1 3 400 40 5 gt 3 200 20 0 0 10 20 50 100 150 200 220 data packet payload size Figure 5 Ghost performance for program image update In fact these metrics are not only Ghost dependent but also re flect some characteristics of the applied communication protocol the node s software architecture and hardware components Regarding the ROM size our current Ghost implementation takes lt 880 B without and lt 1 kB with clone functionality depending on the compiler s optimization level The RAM requirements of the high level module are about 126 B fixed plus the buffer size for the data packet payload As figure 5 shows the latter has significant impact on the data dissemination time Thus we commonly chose 100 B for trade off between speed and memory consumption When updating the software within our test bed we found that SmartNet managed to deliver the data very reliable though the nodes main application also made intensive use of wireless com munication during the remote maintenance process Finally repro gramming all 45 nodes by an image groupcast commonly took the same time as reprogramming a single one approx 22 sec In comparison manual flashing of the same WSN system via JTAG successively node by node took approximately 1 hour and was sim ply too much time
2. s Body sensor Environment sensor A Pressure sensor Coordinator Figure 2 An health care scenario where some nodes are attached to the patient body to mea sure different biological parameters such as heart and pulse rate blood pressure blood oxygen sat uration electrocardiogram ECG and electroen cephalograph EEG values Other nodes are de ployed inside the room to measure the physical pa rameters such as temperature air pressure humid ity and the light values The pressure sensors de ployed under the patient bed timely alert the medical staff for an unexpected downfall of the pa tient The coordinator is responsible to collect and forward the traffic generated by these nodes odic traffic which is generated on the basis of some unex pected event occurred with the patients or within the room is very critical and needs guaranteed access to the channel and bandwidth as they must report before a worse case sit uation happens The examples for such traffic are dramat ically increase decrease in the blood pressure or an heart attack to the patient and unexpected temperature change in the room and the GTS services are used for such traffic It looks straight forward to implement IEEE 802 15 4 in such scenarios but following limitations of the protocol must be figured out before attempting this The first and foremost problem with
3. Figure 1 Overview of AirFlash For experimental WSNs a few more features are needed which will be listed below 1 Several images on a node ID of image names and versions Check AirFlash compatibility of an image Golden Image Easy setup of nodes for AirFlash oh ot ge gt Support heterogeneous networks 7 Background Execution The first feature describes the storage of multiple firmware images on a node where each image may contain a different application This allows selecting which application to run by a simple wireless command The second feature is necessary to trace the versions of images stored on a node This is especially useful during development when a node contains different versions of the same image The features three and four avoid malfunctions while using AirFlash Before a node is programmed with an image AirFlash must check the image for compliance with the current AirFlash version such that no incompatible images will be loaded A Golden Image 1 is an application that resides in a predefined part of the node s memory and provides basic A rFlash functionality If something goes wrong during wireless reprogramming the Golden Image is automatically executed on the node Thus the node will stay accessible The fifth feature is the one step preparation of the node for AirFlash This includes formatting of the nodes memory and installation of the Golden Image Also as feature
4. and J Schiller The ScatterWeb MSB 430 Platform for Wireless Sensor Networks Poster and Abstract Contiki Hands On Workshop 2007 Kista Sweden 03 2006 2 Texas Instruments MSP430F1612 datasheet slas368e user s guide slau049f Texas Instruments Inc Dallas US TX 2006 3 J Corbet A Rubini Kroah Hartman Linux Device Drivers Third Edition O Reilly Media Inc Sebastopol USA 02 2005 4 V Handziski J Polastre J H Hauer C Sharp A Wolisz D Culler D Gay TinyOS TEP 2 Hardware Abstraction Architecture Draft 1 6 2007 02 28 for TinyOS 2 0 published on TinyOS Website http www tinyos com 5 K Klues K P Levis D Gay D Culler and V Handziski TinyOS TEP 108 Resource Arbitration final version for TinyOS 2 x TinyOS published on TinyOS Website at http www tinyos com 6 D Gay P Levis D Culler E Brewer nesC 1 1 Language Reference Manual published on nesC Website 05 2003 http nescc sourceforge net 7 K Klues V Handziski C Lu A Wolisz D Culler D Gay P Levis Integrating Concurrency Control and Energy Management in Device Drivers in Proceedings of the 21st ACM Symposium on Operating Systems Principles SOSP 2007 Washington USA 10 2007 8 Dunkels Gr nvall and Voigt Contiki a lightweight and flexible operating system for tiny networked sensors In Proceedings of the First IEEE Works
5. last update Y N Y__ 22 regular operation sufficient energy N application Ghost HL available haunt Ghost LL reset Figure 4 Ghost low level operation during system runtime The SNOWS 2 5 nodes integrate aMSP430F 1611 10 controller with 10 kB RAM and 48 kB ROM running at 8 MHz An exter nal 2 MB flash memory allows simultaneous buffering of various Ghost program and configuration images Communication is do ne via a CC1100 9 radio transmitter supporting data rates up to 500 kbps SmartOS 3 was developed for real time operation of low power and low performance devices like sensor nodes The preemptive scheduling of prioritized tasks allows prompt processing of inter nal and external events Altogether with the sophisticated resour ce management policy it allows the easy composition of arbitrary tasks to still efficient applications Thus SmartOS based software can easily be extended by the Ghost subsystem Finally SmartNet was developed for both CSMA CA or slotted wireless communication It allows broadcasts groupcasts the in tegration of routing protocols and self organizing communication via e g HashSlot 1 The SmartNet high priority task processes incoming data packets and redirects them to the appropriate recei ver tasks Similar to TCP ports in IP networks this is done by each packet s port ID and destination address In turn the receiver task is signalled and reactivated by Smar
6. 0 0 2 5 5 6 0 6 4 5 0 4 Long Term Link Quality Failure History a Influence of recent transmission success rate on short term link quality b Influence of recent transmission failure rate on short term link quality A label of k h stands for k successes during the last h transmissions and A label of k h stands for k failures during the last h transmissions and h is is a shorthand for h h a shorthand for h h Figure 2 Influence of success and failure of recent transmissions events on short term link quality ing node as next hop alternatives when they 1 identify the link from the forwarding node as short term reliable and 2 conclude that they offer a better routing choice for the ongo ing flow As a result the forwarding node has an increased number of choices for routing Apart from the number of ex pected transmissions to a destination as used in BVR other routing metrics such as link load queue length or battery lev els can be integrated similarly 3 3 Use Case for STLE Packet overhearing technique employed in STLE can ben efit from the bursty traffic patterns observed in WSN Typical applications 8 11 18 19 of WSN involve monitoring envi ronment for events that are of interest to the users Although these events occur rarely but their occurrence results in large bursts of packets that represent major fraction of the overall network traffic In such situations STLE after
7. is obtained using Eq 1 Ea i Estart nal Escanlc na 1 Er Ea 2 where Estart denotes the energy required for the transition from Power Off to Receive state In our simulation nodes always transmit at maximum power we therefore overestimate Escan and Ea by multiplying tscan and ta re spectively by U and Irx gt Irx 8 The energy consumed during the period where the node waits for a retry to as sociate is obtained by adding the energy consumptions for transitions from and to Power Save state to the energy consumptions in Power Save for the remaining time of a 4 SIMULATION RESULTS For analyze the performance the 802 15 4 association pro cedure under varying conditions we simulate it in ns 2 33 To reduce side effects we concentrate on a very simple topol ogy 9 sensor nodes and one sink node taking the role of the PAN coordinator are situated on a line separated by either 4 5 m We use node IDs from 0 to 9 for the sink node and the sensor nodes in increasing distance from the sink The node with ID 9 has thus the greatest distance to the PAN coordinator Using the default loss less ns 2 two ray ground channel model with the radio characteristics of CC2420 and considering varying transmission output pow ers each node can communicate with 1 to 4 neighbors per direction Furthermore we chose A 5 min for 1 sec To obtain repeatable results we use the seeds from 1
8. T T T T T 30000 T T T T T DSR DSR 10000 25000 20000 1000 Delay ms Route setup delay ms 15000 100 1 1 1 10000 1 1 1 0 10 20 30 40 50 60 0 10 20 30 40 50 60 Route neighbor cache TTL 5 Route neighbor cache TTL s a Average delay of DSR and Greedy b Average route setup time of DSR 400 350 300 250 gt N 200 8 9 150 100 50 0 10 20 30 40 50 60 0 10 20 30 40 50 60 Route neighbor cache TTL 5 Route neighbor cache TTL s c Average delay of DSR after route discovery and Greedy d Success rate of DSR and Greedy 20 6 DSR Greedy x 5 57 15 oN 5 2 F 8 10 5 ee g 2 A 4 I 5 3 5 0 3 0 10 20 30 40 50 0 10 20 30 40 50 Route neighbor cache TTL s Route neighbor cache TTL s e Average data rate of DSR and Greedy f Average Hop count of DSR and Greedy Figure 1 First results showing the performance of dynamic source routing and greedy routing under different route cache and neighbor cache invalidation intervals 76 How to Take Advantage from Correlated Node Movement in Mobile Sensor Networks Alexander Klein Innovation Works EADS Deutschland GmbH Munich Germany alexander klein eads net ABSTRACT Many applications are designed for mobile scenarios which have high demands on the routing protocol Frequent topol ogy changes are
9. 3 V Turau and C Weyer Randomized Self stabilizing Algorithms for Wireless Sensor In Proceedings of the International Workshop on Self Organizing Systems IWSOS 06 Passau Germany Sept 2006 4 A Woo and D Culler Evaluation of Efficient Link Reliability Estimators for Low Power Wireless Networks Technical Report UCB CSD 03 1270 U C Berkeley Computer Science Division Sept 2003 5 A Woo T Tong and D Culler Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks In Proceedings of the First ACM Conference on Embedded Networked Sensor Systems SenSys 03 Los Angeles CA USA Nov 2003 Challenges in Short term Wireless Link Quality Estimation Muhammad Hamad Alizai Olaf Landsiedel Klaus Wehrle Distributed Systems Group RWTH Aachen University Germany firstname lastname rwth aachen de Abstract Identifying reliable low power wireless links for packet forwarding in sensor networks is a challenging task Cur rently link estimators mainly focus on identifying high quality stable links leaving out a potentially large set of intermediate quality links capable of enhancing routing progress in a multihop network In this paper we present our ongoing work on short term link estimation that captures link dynamics at high resolu tion in time A short term link quality is calculated based on the recent transmission characteristics of a link This short term quality of a link combi
10. 40 Anzahl empfangener Nachrichten ARNA NN N UN Ex AN a 0 3 0 5 0 6 Wahrscheinlichkeit c Zahl der empfangenen Pakete 0 7 Anzahl gesendeter Nachrichten Knoten erreicht Vergleich gesendeter Nachrichten 0 2 0 3 0 4 0 5 0 6 Wahrscheinlichkeit b Zahl der versendeten Pakete 0 7 0 8 Vergleich erreichte Knoten Top1 Glomo Top1 Tossim Top1 Prism Top1 KSN Top3 Glomo Top5 Tossim Top5 Prism m Tops KSN 0 3 0 4 0 5 0 6 Wahrscheinlichkeit d Anzahl der erreichten Knoten 0 7 0 8 0 9 1 Figure 3 Evaluationsergebnisse bisher genannten Ursachen auch auf starke Varianzen der Ergebnisse bei geringem p zur ckzuf hren ist Die Ergebnisse zeigen insgesamt dass eine grunds tzli che Vergleichbarkeit schon jetzt gegeben ist Deckungs gleiche Ergebnisse werden jedoch durch die stark unter schiedliche Flexibilit t und Konfigurierbarkeit der ver wendeten Werkzeuge sowie das stark vereinfachte Ener giemodell verhindert Um realistische Werte des Energieverbrauchs ermitteln zu k nnen ist die Energieabsch tzung zur Laufzeit und Energiemessungen der Knoten in einem Testbett 7 in teressant Das wird erm glicht durch eine Erweiterung des v
11. Sauff L Torres Dembowski Wagner MULTI CLIENT SYSTEMS IN WIRELESS SENSOR NETWORKS ueesssssseeessssnnnnnsssnnnnnnnnesnnnnnnnnennnnnnnnnnn 31 L Thiem K Scholl M Schuster Th Luckenbach IMPLICIT SLEEP MODE DETERMINATION IN POWER MANAGEMENT OF EVENT DRIVEN DEEPLY EMBEDDED SYSTEMS une sea 37 A Sieber K Walther R Karnapke A Lagemann J Nolte TAB WONS CALIBRATION APPROACH FOR WSN BASED ULTRASOUND LOA ON YS A see an ern 41 C M hlberger M Baunach IMPROVING RESPONSE TIME OF SENSOR NETWORKS BY SCHEDULING LONGT ETEY TIR T 45 N Gollan J B Schmitt DESIGN CONCEPTS OF A PERSISTENT WIRELESS SENSOR TESTBED ccessscsssscesssscsssecesssceseeeens 49 B Blywis F Juraschek M G ne J Schiller KLASSIFIKATION VON SICHERHEITSRELEVANTEN EINSATZSZENARIOS F R DRAHTLOSE SENSORNETZ versenden ee zer 53 Ch Wieschebrink M Ullmann REAL TIME BANDWIDTH AND ENERGY EFFICIENT IEEE 802 15 4 FOR MEDICALAFPPLICATION S ee ee 57 Kumar G nes A Al Basset Al Mamou J Schiller A CLOSER LOOK AT THE ASSOCIATION PROCEDURE IN LOW POWER 802 15 4 MULTIHOP SENSOR tt tent tnne or neee P n E EPEA LEPEP ALEE PAAE PEEP EE Enne aeneae enne 61 B Staehle A NOVEL APPROACH FOR A LIGHTWEIGHT CRYPTO FREE MESSAGE AUTHENTICATION IN WIRELESS SENSOR NETWORKS uuu cceeessccccceccceessssnccccecccessssssecececccesssssaeeeeeees 65 I Martinov
12. To create a virtual device class the virtual device entity is extended to a virtual device class with its own class specific set of methods callbacks configuration and state Callbacks will usually be triggered by interrupts and are directly for warded to a consumer State is dynamic information while configuration is static A virtual device class has a well known set of methods which is implemented platform dependently From outside driver code virtual devices are strictly to be accessed only by the API 21 2 Using Virtual Devices The structures we discussed so far can be kept slim to be implemented on a resource limited device without much im pact on usability However defining an adequate interface specification for virtual device classes that fulfils our require ments uses only a small amount of memory and still can be implemented in the C programming language is a challenge For each device class a minimum and maximum set of op erations can be found The minimum set contains all atomic operations which are absolutely necessary and found on every hardware device The maximum set will also contain more depends on Device Entity Control Interface Hardware Device multiplexes corresponds to Platform Specific Method Implementation Figure 2 Simplified entity relationship diagram of the compositional architecture multiplicity in Chen notation omitted for 1 1 relations 16 high level
13. In this paper we present our on going work on short term link estimation STLE 2 that takes fine grained link dy namics in the order of milliseconds into account and in creases the prediction quality for successful packet transmis sions especially for highly dynamic links STLE integrates into routing protocols by adapting neighbor tables to accu Alexander Becher sd amp m AG Dusseldorf Germany alexander becher sdm de rately reflect the current situation of a dynamic link Overall short term link estimation has three key contributions 1 to predict the probability of successful packet transmission of any link type by taking short term dynamics into account 2 to suggest links of low to intermediate quality for rout ing when they have become temporarily reliable and 3 to integrate easily with today s long term link estimators and routing protocols 2 Related Work The identification of reliable links in WSN has received much attention in the recent years However to the best of our knowledge there is no thorough analysis of short term dynamics in link quality This paper aims to fill this gap by quantifying their extent and characteristics While investigating several approaches for online link estimation Woo et al 20 identified window mean es timator with an exponentially weighted moving average WMEWMA to be an optimal choice to aggregate packet reception rate as an indicator of link quality Four bit l
14. Protocol HL rn LL Functions SmartOS Device Drivers Een Communication External Flash Program Config Module Memory Memory Figure 2 Ghost integration with WSN applications lt Energy Harvester Ghost 2 1 Ghost Functionality Always keep in mind that software based remote maintenance sub systems stay passive for most of the entire system uptime whereas the fraction of their active operation time is relatively small Thus the complexity and permanent resource requirements CPU RAM ROM must be adjusted carefully to the available performance of the nodes and a well balanced cost benefit ratio is required This is why the high level module runs as regular task and requires no further adaptation of the remaining software This task accepts five basic commands from virtually any communication channel 1 New initiates a new image transmission by sending the pre conditions required to run or use the image e g CPU type This is required to avoid accidental installation of incompati ble software or configurations which would render the node inoperable 2 Data successively transmits data fragments until the entire image is transferred The fragments are limited by the pay load size of the underlying network protocol 3 Reset is used for explicit restarting of nodes 4 Query retrieves application hardware and runtime related information about the node This always includes applicati on type version
15. U 0 x T where 0 lt z lt 1 3 choose x where x is an arbitrary constant Above U a 6 stands for a random variable uniformly dis tributed in the interval a b Those simple choices already result in a vast parameter space we postpone the investiga tion of advanced options like deriving from the fixed duty cycling schedule therfore to future works 62 3 2 Evaluation To compare the benefits and trade offs of the different solutions we will use the following metrics for all nodes n of a WSN topology Numerical values for the metrics are obtained as averages from p simulation runs sa n 0 1 indicates if n was able to associate dur ing a target time A Averaging sa n over runs clearly leads to 0 lt sa n lt 1 ta n gives the time when n is associated E n denotes the energy n consumes until it is asso ciated Obviously these metrics vary heavily between nodes of the same topology But the influence of the duty cycle the starting order of the nodes the used transmission output power the network density and more factors make it hard to obtain a general closed form analytical expression Before we evaluate the different association strategies in topologies with varying characteristics in a simulation we have a closer look at the different metrics 3 2 1 Association Success sa n The most straightforward metric of our model is the asso ciation success or percentage Giv
16. cation for our concrete real world test bed 2 CONCEPTS AND OPERATION Figure 1 shows wireless remote maintenance as one possible Ghost infrastructure In this case data is transferred from a workstation computer to a dedicated gateway node The gateway creates data packets and transmits them to the destination nodes by any net work protocol available in the WSN Depending on this protocol unicasts and broadcasts including routing are possible If suppor ted groupcasts offer the special possibility to modify software in a role specific manner simultaneously for certain subsets of nodes Ghost is implemented as add on for integration into any WSN ap plication To be always available for remote commands it runs in parallel to each node s individual software and performs three basic operations 1 reception and integrity check of Ghost command and data packets image fragments from any of the node s communi cation interfaces radio infrared ethernet CAN etc 2 direct execution of Ghost commands and buffering of images in an external flash memory and finally 3 reprogramming of affected memory blocks haunting Therefore the Ghost subsystem is organized in two modules Fig 2 Steps 1 and 2 are executed during regular node operation high level task During step 3 the operating system along with the entire application is stopped low level functions After haunting the device is reset Sl Application Code
17. r ln Kam Figure 2 Graphical User Interface for rule edition The MCS user interface in the upcoming approach will have several modes of usage reflecting the different roles a user can have while interacting with the home network environment in an intuitive way via the MCS Therefore the following user modes are defined Q gt d 10 09 AutHoNe Mainmenu Connected to MyAutHoNe Logged in as ADMIN Options Logout Figure 3 Main menu of the user interface In the General Mode users are allowed to do their registration at the gateway and to get an overview of the whole system The handling of the own profile is also permitted In the second mode called Standard Mode easy as autonomous as possible less impact through the user users are allowed to get a report of all existing networks and network elements such as devices sensor and actuators Furthermore the configuration of rules thresholds and events can also be done in the Standard Mode Finally the Profi Mode complex less autonomous strong impact of the user is suitable for the advanced users like network administrators This mode is the same as the General Mode plus the following additional functions will be developed Configuration of the devices the whole network and its parameters setup of the data gathering behavior as well as the configuration of the actuator behavior The administration of the user profiles is included in
18. Each parameter of the function cost derives from the parameters explained in Section 3 2 Except for the first two parameters that depend on the network topology and that are assumed to be fixed the rest are adapted during the whole routing process according to the information of the adjacent nodes Hence a setup phase is not required The first two parameters the most significant ones are cal culated at some point after getting the number of hops to each one of the sinks only once Let x be the minimum number of hops from a node to a certain sink only neigh bors that offer at most x 1 hops to this sink are taken into account as possible next hops In this way nodes which are farther away from the sinks than the routing node are not considered as next hops In order to explain how combinations of next nodes are cre ated let us consider a simple scenario with two sinks 51 and S2 and a node X with two neighbors and Na offers node X a path to sinks S and 55 with and 5 hops respectively Na offers node X a path to sinks S and S2 with 4 and 3 hops respectively With this information node X creates several combinations If the packet must be routed to 51 the next neighbor can be either N with hops or with 4 hops When routing a packet to S2 the next neighbor can only be N with 3 hops In case the packet must be routed to S and Sa there exist two combinations for S and Na for S2 with a tot
19. Gathering statistics about the network topology link qual ities and power consumption is an important application of the experimental system Once a stable network is es tablished the WSN can support the facility management of the TUHH by automatically forwarding meter readings and monitoring temperature humidity and carbon dioxide in lecture rooms Thus a green campus will be realized For a university of technology it is extremely important to fascinate visitors for technical study paths The GPS ap plication offers a convincing show case which will encourage pupils to choose the career of an engineer 4 REFERENCES 1 J Hill R Szewczyk A Woo S Hollar D Culler and K Pister System architecture directions for networked sensors SIGOPS Oper Syst Rev 34 5 93 104 2000 2 C Renner Energy efficient tdma schedules for data gathering in wireless sensor networks June 2008 3 V Turau and Weyer Location aware in network monitoring in wireless sensor networks In Proceedings of the Sensor Networks Workshop at Informatik 2004 Ulm Germany Sept 2004 Multi Client Systems in Wireless Sensor Networks Lasse Thiem Klaus Scholl Mario Schuster Dr Thomas Luckenbach Fraunhofer Institute for Open Communication Systems FOKUS Kaiserin Augusta Allee 31 10589 Berlin Germany 49 30 3463 7297 lasse thiem klaus scholl mario schuster thomas luckenbach fokus fraunhofer de ABSTRACT In this paper parts
20. The chosen smoothing factor a represented the best trade off between reaction time and peak suppression in the simulated scenarios is the number of changes in the neighbor list during the last observation interval 3 1 Neighborhood Change Detection Changes in the neighbor list are counted during each ob servation interval The counter is increased by one each time a node is added to the list or removed from it Depending on the routing protocol or design of implementation the ap proach can be integrated in the protocols or implemented as a cross layer If the approach is integrated in the pro tocols it can use already existing neighbor lists to calculate the neighborhood change In the case that the approach is implemented as a cross layer it has to keep track of the changes in the neighborhood by itself In the following the focus lies on the cross layer approach since it is independent of the routing protocol and the mac protocol 3 2 Neighborhood List The neighborhood list consists of neighbor entries Bach entry stores information about the neighbor e g time of first contact and time of last contact However the approach presented in this paper only requires the time of the last contact If a node receives a packet of another node it checks whether the originator of the packet is in the list In the case that the node is not in the list a new entry is created and inserted in the list Otherwise the existing corresponding e
21. This value is assigned during initialization for non hardware specific objects or hard coded for hardware specific objects The power manager contains a table with counters for every available sleep mode of the microcontroller used If a component is enabled it signals its deepest possible sleep mode to the power manager by in creasing the corresponding counter in the sleep mode table If a component is disabled the counter of its sleep mode is decreased If no event is pending the power manager iterates the sleep mode table starting at the lightest mode The first value different from zero is the deepest possible sleep mode To guarantee this the table is initialized with zero for all coun ters except the deepest mode which is initialized with one In contrast to TinyOS it is not necessary to evaluate the complete machine state which makes the changing of the lowest possible sleep mode very lightweight A state override is possible to prevent the system to go below a certain sleep mode or to sleep at all This is possible by using an energy manageable object which increments the counter in the power management table for a certain wanted mode This may be useful for reducing wakeup times The sleep mode counter table is the only hardware spe cific part of the power manager since it can have different sizes depending on the used microcontroller 38 5 Evaluation To show the efficiency and compatibility of the presented power manag
22. When data from different sources is present in sensor nodes generic compression algorithms will not always yield opti mum results In such application scenarios we expect our compression framework to outperform solutions that rely on a single static compressor and thus achieve high energy sav ings 5 REFERENCES 1 I Akyildiz W Su Y Sankarasubramaniam and E Cayirci A Survey on Sensor Networks EEE Communications Magazine 40 102 114 2002 2 Barr and Asanovi Energy Aware Lossless Data Compression In Proceedings of the First International Conference on Mobile Systems Applications and Services MobiSys 2003 3 M Buettner G Yee E Anderson and R Han X MAC A Short Preamble MAC Protocol For Duty Cycled Wireless Sensor Networks In Proceedings of the 4th International Conference on Embedded Networked Sensor Systems SenSys 2006 According to the developer homepage at http www cs fit edu mmahoney compression 4 10 11 P Dutta J Hui J Jeong S Kim C Sharp J Taneja G Tolle K Whitehouse and D Culler Trio Enabling Sustainable and Scalable Outdoor Wireless Sensor Network Deployments In Proceedings of the Fifth International Conference on Information Processing in Sensor Networks IPSN 2006 C Intanagonwiwat R Govindan and D Estrin Directed Diffusion A Scalable and Robust Communication Paradigm for Sensor Networks Proceedings of the 6th Annual Internati
23. algorithms may either run independently from each other or interact over common variables that are appended in a type safe way to the nodes The design with processors tasks and the previously pre sented models enables a rapid algorithm engineering with short development cycles per algorithm This makes Shawn an attractive choice especially for implementing and evalu ating high level algorithms iSense 5 is a hardware and software platform Its modular approach allows adapting the hardware and software used for a specific application exactly to the required functional ity A fundamental design guideline is to use only languages and tools that are well understood by a large user community Consequently the object oriented C programming lan guage and popular tools such as the Gnu Compiler Collec tion 7 GCC and the Eclipse 6 development framework are used iSense includes a tiny STL like implementation Thus implementations for standard containers such as lists maps and sets are already part of iSense Hence the learn ing curve for many system designers and solution developers is extremely flat iShell is a tool that is provided by iSense It provides an easy way for the communication between sensor networks and PCs Functionalities like serial and over the air pro gramming are supported as well as a serial terminal and a plug in system to facilitate the supplementation of indi vidual desired functionality as m
24. build number CPU type remaining ener gy and uptime Additional data is optional and node specific but always limited by the payload of a single network data packet 5 Haunt initiates updating of program and or configuration by any previously received image This is done in Ghost low level mode and finally invokes an implicit node reset Figure 3a shows an example for a possible remote software update procedure via a gateway node Depending on the requirements and remaining memory resources an additional clone function can be statically linked into the Ghost subsystem 6 Clone requests the currently running application code from a node By first sending a query a node can check its neighbourhood for available software types and versions Thereby newly deployed no des can integrate themselves almost autonomously into a running WSN by requesting the appropriate software from neighbour no des Figure 3b shows an example Yet this autonomy requires that 82 Destination node Gateway Destination node b Clone mode a Gateway mode Source node Figure 3 Ghost communication process each new node is equipped with a certain initial application contai ning the Ghost subsystem and the knowledge about its future role in the WSN This role is defined by the application type which is un ique for each kind of software and hardware combination e g rou ter for TelosB nodes ultrasound sensor for SNOWS nodes etc
25. gungen der Designmerkmale bei ei ner Implementierung erfordern Die Klassifizierung soll die Grundlage der in den folgenden Phasen zu findenden Hand lungsempfehlungen f r Sicherheitsmechanismen bilden In der zweiten Projektphase sollen m glichst umfassend ver ffentlichte Sicherheitsmechanismen und protokolle die f r den Einsatz in drahtlosen Sensornetzen konzipiert wurden zusammengestellt und hinsichtlich ihrer Anforderungen un tersucht werden Falls m glich soll jeweils eine Einsch t zung ber das durch den Sicherheitsmechanismus bereitge stellte Schutzniveau geliefert werden Anschlie end soll un tersucht werden ob sich ein Sicherheitsmechanismus einer oder mehrerer der gefundenen Klassen von Einsatzszena rios zuordnen l t Notwendig hierf r ist nat rlich dass die technischen Merkmale der Szenarios einer Klasse die Vor aussetzungen der Sicherheitsmechanismen erf llen Zu be r cksichtigen ist auch ob ein Mechanismus den gew nsch ten Schutzbedarf in einem Szenario erf llen kann Auf die se Weise entsteht ein Katalog von Einsatzempfehlungen f r konkrete Einsatzszenarios und es lassen sich leicht noch of fene Forschungsfragen ableiten etwa wenn f r eine Klasse keine geeigneten Sicherheitsmechanismen gefunden wurden 56 Im Rahmen des Projekts sollen ausdr cklich nicht vollst n dige Sicherheitsarchitekturen f r einzelne Einsatzszenarios entworfen werden Zun chst soll nur festgestel
26. remaining period Energy efficient MAC protocols specifi cally tailored to sensor networks such as 7 11 3 6 make use of such schedules and thus allow for significant energy savings We anticipate that combining such MAC proto cols with supplementary extensions to minimize the number of packet transmissions and the corresponding payload sizes can lead to even higher savings and an extended node life time 257 Current mA 1 off standby RX on TX on off on off idle Configuration Radio CPU Figure 1 Power consumption of tmote sky nodes numbers taken from 9 This paper compares two methods to reduce traffic in sensor networks see Section 2 and outlines their mode of opera tion Special emphasis is put on their applicability in WSNs as resource constrained devices generally exhibit character istics that differ from common desktop computers Subse quently we present our approach towards an adaptive packet compression framework for sensor network applications in Section 3 By compressing sensor data with the locally opti mal energy efficiency ratio energy can be preserved and thus the node lifetime extended The description of our vision is followed by an analysis of related work Finally conclusions and an outlook will be presented in Section 4 19 2 REDUCING TRAFFIC IN THE WSN Although several methods of reducing the size or number of packets in WSNs exist we introduce two common
27. the as sociation procedure was also studied in 11 The association process as outlined in the standard is not optimized for the case of large duty cycled or lossy networks with only one coordinator It does e g not handle the situation that a node wanting to join a PAN does not receive a beacon upon its active channel scan To solve this issue 11 proposes each device which is unable to associate at first attempt to retry to associate after an associationRetryInterval a 1 sec later As the authors focus on beaconed networks and the sensor nodes do never go to sleep state those results are of limited use for the case of non beaconed low power networks In 6 the authors establish analytical models for comput ing the time and energy consumptions for 802 15 4 specific mechanisms Furthermore typical power consumptions and goodput for devices and coordinators are derived and ver ified by simulation The reassociation procedure i e the case where a node loses the contact to the coordinator it as sociated with is handled the initial association procedure is not covered Furthermore duty cycling nodes are not consid ered the analysis is thus difficult to apply to our problem The authors of 4 exploit the hierarchical dependency re sulting from the 802 15 4 association procedure for a estab lishing a hierarchical routing scheme HERA as this algo rithm is called outperforms AODV which is proposed by the ZigBee Alliance for
28. 7 REFERENCES 1 D Babic and A J Hu Calysto scalable and precise extended static checking In ICSE 08 Proc of the 30th international conference on Software engineering 2008 2 Cadar V Ganesh Pawlowski L Dill and D R Engler EXE automatically generating 10 12 inputs of death In CCS 06 Proc of 18th conf on Computer and communications security 2006 S Chandra and T Reps Physical type checking for C SIGSOFT Softw Eng Notes 1999 H Chen and D Wagner MOPS an infrastructure for examining security properties of software In CCS 02 Proc of the 9th conference on Computer and communications security E Clarke D Kroening and F Lerda A Tool for Checking ANSI C Programs In Tools and Algorithms for the Construction and Analysis of Systems TACAS 2004 2004 N Cooprider W Archer E Eide D Gay and J Regehr Efficient memory safety for TinyOS In SenSys 07 Proc of the 5th international conference on Embedded networked sensor systems 2007 T A Henzinger R Jhala R Majumdar G C Necula G Sutre and W Weimer Temporal Safety Proofs for Systems Code In CAV 02 Proc of the 14th International Conference on Computer Aided Verification 2002 S Johnson Lint a C program checker Computer science technical report 65 Bell Laboratories 1977 J C King Symbolic execution and program testing Commun ACM 1976 R Kumar E Kohler and
29. Betrachtung der erreichten Genauigkeit sowie der Einfluss fehlerbehafteter Distanzmessungen im Vordergrund Der vorgestellte DLS Algorithmus ist um einige Besonderheiten erweitert worden weist aber auf dem Gebiet der Mobilit t von GSN Schw chen auf die im weiteren Verlauf der hier beschriebenen Projekte eliminiert werden sollen Das Projekt SLEWS arbeitet an der Weiterentwicklung von Technologien und Methoden f r den Einsatz von Beobachtungs und Fr hwarnsystemen bei Hangrutschungen Die OGC SWE Initiative liefert mit Spezifikationen und Datenmodellen umfangreiche Werkzeuge zur Integrierung von Instrumenten wie Geosensornetzwerken in ein interoperables Sensor Web Die Verwendung offener Standards und Schnittstellen sichert Herstellerneutralit t verbessert system bergreifende Interoperabilit t im Kontext nationaler und internationaler Fr hwarnsysteme und f hrt zur Entkopplung von anf lligkeitsbehafteten hierarchischen monolithischen Informationsstrukturen Wesentliche Herausforderungen bestehen in der funktionellen Kombination der SWE Dienste zur Lieferung von Echtzeitdaten und in der semantisch eindeutigen Schaffung von Namensr umen f r Ph nomen und Sensorparameter in SWE Datenstrukturen 7 Danksagung Dieses Arbeit wird durch die Deutsche Forschungsgemeinschaft DFG unter der Nummer BI467 17 1 Schl sselwort Geosens sowie durch das Bundesministerium f r Bildung und Forschung BMBF unter der Nummer 03G0662A Schl sselwo
30. DLS Reichenbach et al 2006 die Algorithmen bei gleichbleibender Pr zision vereinfachen Der DLS benutzt Distanzen zwischen den Knoten und basiert auf einer anschlie enden berbestimmten Trilateration Durch eine geschickte Verteilung der Berechnungen komplexe Aufgaben Matrixinversionen werden auf die ressourcenstarken Beacons ausgelagert und nach Fluten dieser ins GSN auf den Sensorknoten die sparsame Nachberechnung Multiplikation Addition einer 2x2 Matrix ausgef hrt wurden Einsparungen von bis zu 86 im Ressourcenverbrauch im Vergleich zu etablierten exakten Methoden erreicht 3 1 3 Ausrei er Messungen wenn sie denn durchgef hrt werden sind immer fehlerbehaftet Gerade die bekannten Distanzmessverfahren z B Signalempfangst rkemessung RSS in GSN sind sehr fehleranf llig und k nnen eine gro e Zahl Ausrei er produzieren Reichenbach und Timmermann 2006 Wie im Kapitel 2 aufgezeigt k nnen auf den Sensorknoten diverse Sensoren zur Erfassung physikalischer Ph nomene installiert sein Diese Sensormesswerte k nnen wenn sie mit definierten Regeln mathematisch mit r umlichen Eigenschaften in Verbindung gebracht werden zur Verbesserung der Position eingesetzt werden In Reichenbach et al 2008 wird der Anomaly Correction in Localization Algorithmus ACL vorgestellt mit dessen Hilfe Sensormesswerte zur Ausrei erdetektion genutzt werden Daf r wird ber dem Gebiet von Interesse eine Footprintkarte in Se
31. Dennoch hat in der Regel das je weilige Szenario einen entscheidenden Einfluss auf das De sign des Systems und legt dieses in weiten Teilen fest weswe gen man von den Desginmerkmalen eines Einsatzszenarios sprechen kann Sollten die oben genannten Szenarios tat s chlich breit in der Praxis umgesetzt werden spielen auto matisch Sicherheitsfragestellungen eine Rolle Dies betrifft sowohl die Informationssicherheit als auch die Ausfallsicher heit Die Auspr gung der technischen Merkmale hat bedeu tenden Einfluss auf die Auswahl geeigneter Sicherheitsme chanismen 3 SICHERHEITSMECHANISMEN Die oben genannten Einsatzszenarios werfen eine Reihe von Sicherheitsfragestellungen auf Im Fall der Patientenloka lisierung d rfen Unbefugte z B nicht beliebig auf sensible Patientendaten zugreifen k nnen Das Grenz berwachungs system darf durch einen Angreifer der das Gebiet passie ren will nicht au er Betrieb gesetzt werden und bei der Geb ude berwachung sollten absichtliche Fehlalarme nach M glichkeit verhindert werden Die Sicherheitsanforderun gen die in Sensornetzen von Belang sind lassen sich in den allermeisten F llen auf die abstrakten Anforderungen Inte grit t Authentizit t Vertraulichkeit von Nachrichten und Verf gbarkeit zur ckf hren Die ersten drei Anforderungen werden in IT Systemen durch kryptographische Verfahren realisiert die vierte in der Regel durch Infrastrukturma nahmen Im Falle von Sensorne
32. Heunecke 2008 verkn pft ein drahtloses Sensornetzwerk mit der Notwendigkeit die Position eines oder mehrerer Knoten in einem bergeordneten Lagebezugssystem zu bestimmen siehe hierzu Kapitel 3 Drahtlose ad hoc Geosensornetzwerke werden zuk nftig aus tausenden winzigen elektronischen kosteng nstigen Sensorknoten bestehen die ihre Umgebung berwachen einfache Rechenschritte ausf hren und miteinander kommunizieren k nnen Das Geosensornetzwerk konfiguriert sich unmittelbar nach Ausbringung der Knoten selbst Jeder einzelne Knoten ist in der Lage bei Bedarf aktiviert zu werden und solange zu arbeiten wie seine Energiequelle ausreicht Mittels Methoden wie Selbstheilung und Selbstorganisierung reagiert das Netzwerk auf Knotenausf lle und St rungen Je nach Auswahl der Detektoren ist ein GSN als universell anpassbares Instrument zur Beobachtung verschiedenster Ph nomene einsetzbar welches die gro r umige Erfassung von Umweltph nomenen unterschiedlichsten Umgebungen z B auch in schwer zug nglichen Regionen und bei sich bewegenden Objekten erlaubt Geosensornetzwerke sind zuk nftig als Quelle automatisierter Datengewinnungsmethoden f r verschiedenste GIS Anwendungsfelder sehr interessant siehe hierzu Kapitel 5 3 Positionierung in Geosensornetzwerken Eine der wichtigsten Aufgaben f r eine raumbezogene Datenauswertung ist bei einer zuf lligen Ausbringung der Sensoren die Bestimmung der Position jedes Einzelnen U
33. In order to allow immediate service of DSR we provided the source node with a slightly increased message buffer size We set the buffer size to 8 Our implementation buffers the messages com ing from upper layers until the first RREP message arrives From there on messages are immediately sent out to the next hop node until the route gets invalidated Our DSR implementation provides a timeout which can be used to invalidate the current route from source to desti nation This means when using this timeout a new route discovery will be initiated periodically However this is just an optional feature which is useful to adapt proactively to a dynamically changing network topology due to mobility or any kind of fading effects So far we were interested in performing measurements in a single source single destination traffic pattern Thus we have not implemented DSR route caching strategies so far 2 2 Greedy Routing 2 2 1 The Algorithm Principle The greedy routing principle 10 2 8 works as follows Each node is supposed to know its own location in terms of some given coordinate system Messages store the location of the destination node as well A current forwarding node has to acquire the position of its one hop neighbors and then choose the next hop as the neighbor whose location is the best one with respect to the current node s location the final destination s location and the localized metric being applied If the curre
34. M Wilhelm C Wegmann and J B Schmitt Wireless Client Puzzles in IEEE 802 11 Networks Security by Wireless In Proceedings of ACM Conference on Wireless Network Security WiSec 2008 pages 43 52 Alexandria VA USA March 2008 Reliable Multicast in Wireless Sensor Networks Gerald Wagenknecht Markus Anwander Marc Brogle and Torsten Braun Institute of Computer Science and Applied Mathematics University of Bern Neubrueckstrasse 10 3012 Bern Switzerland wagen anwander brogle braun iam unibe ch ABSTRACT Multicasting in Wireless Sensor Networks WSNs is an eff cient way to disseminate the same data to multiple receivers For critical tasks such as code updates reliability would be a desirable feature in order to use multicasting for such scenarios Due to the nature of WSNs several problems exist that make realizing an efficient reliable and energy consumption friendly implementation a challenging task In this paper we describe the challenges of such an implementa tion and propose a solution for designing a reliable multicast solution based on IP Multicast and Overlay Multicast We discuss several scenarios and depict the different advantages and limitations of the solutions proposed 1 INTRODUCTION 1 1 Wireless Sensor Networks A Wireless Sensor Network WSN consists of a number of sensor nodes which are limited in terms of energy CPU power and memory On the sensor nodes may run differ ent applications fo
35. Measuring Round Trip Times to Determine the Distance Between WLAN Nodes In R Boutaba K C Almeroth R Puigjaner S X Shen and J P Black editors NETWORKING volume 3462 of Lecture Notes in Computer Science pages 768 779 Springer 2005 M N Huxley Corrigenda Exponential Sums and Lattice Points II Proc London Math Soc 53 68 2 264 1994 H Keller Numerical Studies of the Gauss Lattice Problem Technical Report CRPC TR97699 Center for Research on Parallel Computation Houston Jan 1997 R Prasad and Ruggieri editors Applied Satellite Navigation Using GPS GALILEO and AugmentationSystems Artech House 2005 N B Priyantha The Cricket Indoor Location System PhD Thesis Massachusetts Institute of Technology June 2005 A Savvides Han and Strivastava Dynamic fine grained localization in Ad Hoc networks of sensors In MobiCom 01 Proceedings of the 7th annual international conference on Mobile computing and networking pages 166 179 New York NY USA 2001 ACM Press Improving Response Time of Sensor Networks by Scheduling Longest Flows First Nicos Gollan and Jens B Schmitt TU Kaiserslautern Distributed Computer Systems Lab box 3049 67653 Kaiserslautern Germany gollan schmitt informatik uni kl de ABSTRACT When deploying wireless sensor networks sink trees are a natural topology choice However those networks have an intuitive drawback the deeper down
36. Systems DES testbed is introduced Subsequently in Section 3 we discuss our design concepts for integration We compare our architecture and approach to other testbed setups in Section 4 The paper ends with a conclusion in Section 5 2 HYBRID TESTBED ARCHITECTURE At the time of this writing we are in the process to set up a hybrid testbed 9 at the Freie Universit t Berlin The testbed consists of hybrid nodes containing a wireless mesh router and a wireless sensor node in one shared enclosure So far we installed 35 of these hybrid nodes in our office rooms spanning three floors with the goal to deploy 100 nodes in total The WSN is our contribution to the European Union 49 WISEBED 3 project This multi level infrastructure of in terconnected testbeds pursues the integration of heteroge neous small scale devices for research on large scale The European Union OPNEX 1 project in contrast is focused on the design of architectures and protocols for multi hop wireless mesh networks based on IEEE 802 11 technology The corresponding mesh routers are used as foundation of the hybrid testbed Figure 1 depicts our general system setup which is based on a three tier architecture Mesh routers and sensor nodes are deployed in our institute in a planned but not uniform manner As part of future extensions adjacent buildings will be comprised The stationary mesh routers build the core backbone network representing tier 1 Tier 2 c
37. The framework is targeted to al low dynamic updates and modifications of the compressors and or compressor parameters during runtime This allows to exchange slow implementations of algorithms with opti mized ones and tune parameters to fulfill the application s needs 3 1 Related Work A variety of data compression and decompression respec tively algorithms are well known although only few of them were specifically tailored to fit the needs of sensor networks The emerging field of sensor networks however poses differ ent constraints on resulting algorithms than common desk top computers do Both processing power and available memory are tightly limited in sensor nodes and their tight energy budget must additionally be considered For these reasons heavyweight compressors such as partial predic tive matching PPM or similar algorithms that require big dictionaries to be stored in RAM cannot be run on the nodes The even more sophisticated PAQ compressor quotes a minimum RAM consumption of 35 megabytes although values range up to 1712MB Barr and Asanovi performed an analysis of several compression algorithms on a Compaq Personal Server handheld device in 2 The platform fea tured a 233MHz CPU with 32 megabytes of RAM render ing a direct mapping of the results to sensor node platforms equipped with low power CPUs and a few kilobytes of RAM impossible Other authors have noticed this discrepancy between highly dema
38. Vicari Kuhnert A Cross Layer Approach for Enabling Low Duty Cycled ZigBee Mesh Sensor Networks In ISWPC 08 Santorini Greece May 2008 8 Texas Instruments 2 4 GHz IEEE 802 15 4 ZigBee ready RF Transceiver Texas Instruments 2006 9 USC Information Sciences Institute The Network Simulator ns 2 http www isi edu nsnam ns 10 Wang and W Yang Energy Consumption Model for Power Management in Wireless Sensor Networks In SECON 07 San Diego CA USA June 2007 11 J Zheng and M J Lee A Comprehensive Performance Study of IEEE 802 15 4 chapter Sensor Network Operations pages 218 237 IEEE Press 2004 A Novel Approach for a Lightweight Crypto free Message Authentication in Wireless Sensor Networks Ivan Martinovic and Jens B Schmitt TU Kaiserslautern Distributed Computer Systems Lab box 3049 67653 Kaiserslautern Germany martinovic jschmitt informatik uni kl de ABSTRACT In this paper we propose a system leveraging the peculiarities of the wireless medium such as the broadcast nature of wireless com munication and the unpredictability of indoor signal propagation to achieve effective protection against attacks based on the injection of fake data in wireless sensor networks WSNs Using a real world WSN deployment and a realistic implementation of an at tacker we analyze this protection scheme and demonstrate that nei ther position change transmission power manipulatio
39. Zeitpunkt an dem die Anwendung den Sendeauftrag erteilt bis zur Versandbe st tigung der TinyOS Netzwerkschicht Nicht ber cksichtigt werden hier Energiewerte f r den Mikroprozessor und das An Ausschalten des CC2420 da diese f r den Vergleich der Werkzeuge nicht relevant sind Der Stromverbrauch beim Lauschen auf Pakete so wie zus tzlicher Energieaufwand bei Kollisionen bleibt ebenfalls unber cksichtigt da die notwendigen Zahlen nicht von allen verglichenen Werkzeugen geliefert wer den Deshalb stellen die Energiewerte aus den Simula toren sehr kleine untere Schranken f r den Energiever brauch von MICAz Knoten dar die in der Praxis nicht erreicht werden k nnen Die damit in verschiedenen Si mulatoren errechneten Werte lassen sich jedoch gut ver gleichen 4 AUSWERTUNG Im Folgenden werden die einzelnen Experimente und die Besonderheiten der einzelnen Werkezuge vorgestellt 4 1 GloMoSim Die Implementierung des Versuchs erfolgte als Proto koll auf Anwendungsebene Die von GloMoSim bereit gestellten Implementierungen der niedrigeren Netzwerk schichten wurden unver ndert bernommen und ledig lich in ihrer Parametrisierung gem den Vorgaben an gepasst Die Knotenplatzierung wurde direkt aus den einheitlich definierten Szenariobeschreibungen bernommen und in der Konfiguration mit NODE PLACEMENT FILE refe renziert Die Signalausbreitung wurde mit dem in Glo MoSim vorhandenem TWO RAY Modell simuliert wel
40. Zhou T He S Krishnamurthy and J A Stankovic Impact of radio irregularity on wireless sensor networks In Proc of the Sec ond International Conference on Mobile Systems Applications and Services MobiSys 2004 June 2004 23 M Zuniga and B Krishnamachari Analyzing the transitional region in low power wireless links In Proc of the Conference on Sensor and Ad Hoc Communications and Networks SECON 2004 SomSeD An Interdisciplinary Approach for Developing Wireless Sensor Networks 2 1 2 Sebastian Georgi Christoph Weyer Martin Stemick Christian Renner Felix Hackbarth Ulf Pilz Jens Eichmann lobias Pilsak Harald Luis Torres Klaus Dembowski Fabian Wagner Hamburg University of Technology georgi tu harburg de c weyer tu harburg de ABSTRACT The research field Self organized mobile Sensor and Data networks SomSeD is introduced Its purpose is the in vestigation of Wireless Sensor Networks WSN It benefits from interdisciplinary exchange between various institutes of the Hamburg University of Technology TUHH Due to different design constraints such as energy efficiency and package size compared to well known classical computer networks all aspects of the development of WSNs must be reconsidered This paper describes the advantage of having experts of various faculties both in computer science and electrical engineering in a single research field In addition to the i
41. adaptive wireless sensor networks One of the next steps will be to build up the Experiments Repository This contains the aim of establishing a reposi tory of applications for simulation as well as building up a testbed composed of real sensor nodes Further the testbed shall be connected to the Internet via the tool iShell that is provided by iSense That way all project partners will be able to test their algorithms on a common hardware plat form 5 REFERENCES 1 FP7 http cordis europa eu fp7 home_en html 2 FRONTS http fronts cti gr index php home 3 ns 2 Network simulator 2 http isi edu nsnam ns 4 Shawn http shawn sf net Buschmann and D Pfisterer iSense A modular hardware and software platform for wireless sensor networks Technical report 6 Fachgespr ch Drahtlose Sensornetze der GI ITG Fachgruppe Kommunikation und Verteilte Systeme 2007 6 Eclipse Foundation Eclipse an open development platform 2001 http www eclipse org 7 Free Software Foundation Inc Gnu Compiler Collection GCC 1984 http gcc gnu org 8 A Kr ller D Pfisterer Buschmann S Fekete and 5 Fischer Shawn A new approach to simulating wireless sensor networks In Proceedings of the 3rd Symposium on Design Analysis and Simulation of Distributed Systems DASD 05 pages 117 124 2005 103
42. by an existing infrastructure On the other hand the WSN hardware can also be beneficial for the WMN Furthermore there is a lack of long term experience with wireless sensor network applications since many studies are limited in time or use small setups only Few reports about long term performance studies reliability and maintenance of wireless sensor networks have been published Though Testbed Server Wired Backbone Backbone mesh routers a N Pi ri En P H 7 ra 52 k a Routing wa Non routing mesh clients F mesh clients pense v R A o Wireless Connection Wired Connection Mesh Routers A Extensions Sensor Nodes Figure 1 Architecture of the hybrid testbed con sisting of mesh routers mesh clients sensor nodes and the testbed server hosting management com ponents including databases Sensor nodes may be connected with wired extensions there are many real world applications with an intended run time of several years Life cycles of up to 10 years are a common demand in facility or animal monitoring But how can we execute experiments of adequate time span We present an approach to save and restore the network state at particular times The remainder of the paper is organized as follows In Sec tion 2 the architecture of the Distributed Embedded
43. channel scan i e it broadcasts a beacon request command and waits for a response for an ap plication specific time This procedure will be repeated on all or some of the available channels The PAN coordinator or a device which has already associated will send a bea con containing information about the PAN it belongs to in response to this request After having scanned all channels n chooses a PAN for associating among the PANSs it got to know of and exchanges a sequence of command messages with the sender of the corresponding response As mentioned earlier 11 proposes each device which is unable to associate at first attempt to retry to associate after an associationRetryInterval a 1 sec later For the case of a self organizing duty cycled network which may be not synchronized at the beginning this solution is not al ways successful Especially in sparse and low duty cycled networks a node may have physical neighbors i e nodes within its range but may be temporally isolated as it does not share waketime with the nodes in its radio range Us ing a fixed a will thus very likely not result in a successful association In this initial work we will point out directions for en abling a low power association procedure investigating the following strategies 1 choose a at random but dependent on the duty cycle eg a U 0 2pw T where 0 lt x lt 1 2 choose a at random but independent from the duty cycle e g
44. ches ein PROPAGATION LIMIT von 111 dBm vor sieht Die Parameter zur Funkschnittstelle wurden ent sprechend der MicaZ Spezifikation angepasst Insbeson dere wurde das SVR BOUNDED Modell zum Empfang von Datenpaketen ausgew hlt welches Signale ber dem RADIO RX SNR THRESHOLD von 10 dB fehlerfrei zu stellt und sonst verwirft Als MAC PROTOCOL kommt CSMA CA zum Einsatz Die Simulationsergebnisse sind Mittelwerte von je 1000 Durchl ufen mit variierenden SEEDs des GloMoSim Zufallsgenerators 42 TOSSIM TOSSIM ist ein Simulator der von TinyOS bereitgestellt wird Sein Radiomodell basiert auf der Signalverst r kung gain die jeweils zwischen zwei Knoten festge legt wird Da die Signalst rke mit der Entfernung ab nimmt ist gain immer negativ In diesem Experiment haben wir die gains mit Hilfe der von TOSSIM bereitge stellten Hilfsmittel generiert Als Eingaben dienten die Koordinaten der Knoten und mehrere Umgebungspara meter path loss exponent noise floor white Gaussian noise usw die dem Tutorial 12 entnommen wurden TOSSIM simuliert Umgebungsrauschen und Interferen global a 0 1 init 1 module SensorA Local State O idle 1 forward 2 drop 3 sleep la init 0 1 amp 1 0 gt pf la 1 amp b 1 amp d 1 amp c 1 Ll pf 1 amp 1 2 gt la 2 endmodule Figure 2 Prism Modell eines Moduls das einen Sensor knoten repr se
45. consuming In fact this expected speed up was the initial main motivation for developing the Ghost system at all Finally table 1 shows the energy requirements of a single node during various operation modes One can see clearly that updating the program memory imposes significant load on the power supply compared to regular operation Yet intentionally provoked node breakdowns due to current peaks or empty batteries were safely handled by the low level recovery function 5 CONCLUSION AND OUTLOOK In our paper we presented the Ghost remote maintenance subsys tem for wireless sensor and actor networks We showed that effi cient and fast software and configuration updates are possible wi thout the need for a predefined communication protocol or espe cially modified system software Instead Ghost integrates easily into any WSN application and accepts data from arbitrary sources wired or wireless Thereby it requires little memory and compu tational power Besides it supports role specific software deploy ment by either direct transmission via a gateway or by autonomous self maintenance of the nodes via cloning This way simultaneous updates are even possible in heterogeneous networks Finally the sophisticated low level module along with the optional encrypti 84 on allows extremely safe and secure operation for reliable remote maintenance Our current work in this field addresses viral data dis semination techniques power aw
46. device may require a different configura tion of the shared resource such as data rate or protocol Vir tual device hubs contain all state information necessary for multiplexing a shared device entity between an arbitrary number of virtual devices A single virtual device can be se lected as default and will be active whenever no other device is active The device selection is also used to forward inter rupts to the right virtual device A hub does not have a list of connected devices which allows introducing new virtual devices dynamically at runtime A virtual device entity is an abstract software device It is specialized by virtual device classes and is not intended to be used stand alone It extends the device entity In this way the virtual device becomes a unique device entity itself can refer ence its dependencies and provide its own power and configu ration management It can be connected to a virtual device hub which then functions as a switching facility for all virtual devices operating on the same shared device A virtual device class represents a set of virtual device enti ties with a common interface Usually a common functionality is also assumed Typical classes found on embedded devices would be serial byte I O e g communication ports peripheral connections or memory e g EEPROM external flash card Depending on the implementation access to the underlying hardware can be shared exclusive virtualized or emulated
47. environments we also address practical aspects like performance and resource requirements Some results from a real world installation will close this paper 1 INTRODUCTION In order to achieve a long lifetime for sensor actor networks hence forth simply called WSN these distributed systems are subject to regular maintenance cycles Besides hardware related issues like the renewal of power supplies or the replacement and attachment of modules software related modifications are also very common The latter allow application updates for integration of new functio nality or for fixing bugs Sometimes there is just the need to reset a node or to modify its configuration for changing its behaviour or individual role in the overall system The problem s scope and intensity differs according to the evoluti on stage of the system During software development frequent up dates test versions for few nodes can be expected The frequency diminishes rapidly with the final release but then affects significant ly more nodes within the original environment But still several updates might be required suddenly during the system runtime It is similar with hardware During development there are commonly few nodes but possibly they are already of different architectures Often these nodes are densely placed and easily accessible After system deployment their number and heterogeneity increases wi thin a quite ample environment Then some of them
48. into the network an attacker must find an appropriate configuration of wireless parameters such as physical position transmission frequency wireless channel and a trans mission power level to produce the received signal strength which satisfies the acceptance interval on all sensors If any of legitimate sensors detects the impersonation 1 it measures an RSS of a frame not included in acceptance interval the receiver is notified by a warning frame disclosing a sequence number of the injected frame Dynamic configuration is used to hinder an attacker in its brute force search for a successful configuration All WSN sensors are instantiated with the same PRNG seed which allows them to per mute different transmission parameters defined during the deploy ment phase So for example one configuration would be to transmit on channel 15 with a transmission level 10 another to transmit on channel 22 with a transmission level 15 As a result different con figurations are assigned with different acceptance intervals Even if an attacker is able to inject frames with a certain configuration the attack is only successful during the same acceptance intervals To allow periodic change of acceptance intervals the WSN uses a simple synchronization method where a basis station periodically broadcasts a beacon to signal the change of the transmission pa rameters e g every 2 minutes An example can be seen in Figure 4 which shows a time lin
49. is successful the fragment is written to the respective page of the external memory and an ACK packet is sent to the host This buffering of whole fragments of the image has two benefits First only a few ACKs are needed which improves transfer speed And second most data corrections take place inside the RAM which extends the lifetime of the external memory When the image transfer is finished the host can invoke the bootloader by sending a special command This command also contains the number of the slot to be loaded This information is stored in the EEPROM to preserve it for the bootloader After the description of the transfer process a closer look at the AirFlash packet is taken The first field contains the AirFlash ID that distinguishes between up and downlink of an AirFlash connection Thus other nodes with AirFlash connections are not disturbed by packets from other nodes 13 The Packet ID is a simple counter which is used for packet ordering The Command field contains commands to the AirFlash module The Storage page identifies the current page to write and the offset points to the actual subfragment of this page It follows the data size and the data field which can contain up to 16 bytes The last field is a CRC checksum to detect packet errors Summarizing the main features of the AirFlash module are listed below Implementation as component thus easy integration in TinyOS applications Runs as bac
50. lead to disconnected topologies in dense networks 3 ADAPTIVE LINK ESTIMATOR Since the proposed adaptive link reliability estimator can be used independently from the neighborhood management protocol it is specified separately in this section As de scribed in 4 the PRR is estimated by observing the success ful delivery of periodical broadcast packets If application specific packet rate is sufficient the estimator information can be sent piggybacked in the data packets in order to re duce the periodical broadcast packets EWMA in agile mode is a fast responding link estimator while in stable mode a low mean square error is achieved The main goal of the adap tive link estimator proposed in this paper is to combine the strength of both modes The adaptive link estimator ALE uses EWMA as a basis It has the capability to adjust the parameter a depending on 24 300 Agile EWMA Stable EWMA Daa ee E 5 200 amp 150 4 1001 50 4 r 01 02 03 04 05 06 07 08 09 1 PRR Figure 1 Crossing Time the current estimated link quality improve the crossing time the initial PRR is set to the threshold value PRRoaa whenever the first packet is received from a node In this pa per is set to 0 5 Additionally when the PRR is be low PRRoag or new link must be estimated the agile mode is used in order to estimate quickly the real link quality ALE
51. long term link estimators and routing protocols We present our approaches on the identification of temporarily available and unavailable links and evaluate these in Section 4 Our inter est in short term link estimation is motivated by two key ob servations indicated by research 3 13 21 and our own mea surements 1 Links of intermediate quality amount to about half the number of high quality stable links see Figure 1 a and 2 this percentage grows with the physical distance see Figure 1 b Although links of intermediate quality offer further choices for routing and often promise long distance con nectivity Figure 1 c shows that this class of links is sub ject to large and frequent temporal variations Their dynamic connectivity poses a special challenge to any link estimator Long term link estimators are not designed to identify short term link dynamics As a result they adapt slowly to chang ing link conditions limiting their use to the identification of long term stable links 3 1 Deriving a Short term Link Estimator Commonly links in a wireless network can be classified into three categories good links that are reliable in the long term intermediate unreliable links often with frequently changing quality and bad links that very seldom transmit a packet successfully Figure 1 a shows that the ratio between good links and intermediate links is 2 1 in our testbed mea surements Furthermore our measurements indic
52. networks Table 1 Summary of differences between academic wireless sensor networks and industrial wireless sensor networks ee Academic WSN Industrial WSN Deployment random planned unsupervised supervised Infrastructure only to a certain extent 4 CONCLUSION Three typical applications of wireless sensor networks in an in dustrial setting have been described Their usage environment was compared with the assumptions often made by academic research work The BMBF funded ZESAN project develops solutions for reliable energy efficient and self organizing wireless sensor networks addressing requirements of prominent industrial applica tion scenarios The gap between topics addressed by academic research and the needs of the envisaged industrial applications needs to be closed Industrial wireless sensor networks are usually planned get deployed in a deterministic way and are supervised during their operation and can make use of infrastructure compo nents Some of those infrastructure components are used to inte grate the industrial wireless sensor network into the plant system And finally the number of nodes used in industrial sensor net works is likely to be lower than assumed in some academic re search and will typically range from tens to at most some hun dreds The assumptions made by academic research make it some times difficult to transfer the results to industrial applications having different requirements and si
53. node s total break down caused by unpredictable errors while haunting This is why the Ghost subsystem is divided into two parts As the high level task runs as part of the application and in parallel to the user tasks it is permanently ready to process Ghost packets Fig 2 If a haunt command arrives the Ghost task checks for sufficient remaining energy to perform the expensive update ope ration If energy is low haunting is deferred to allow a possibly available energy harvester to recharge the power supply As soon as enough energy is available or if the charge level can not be deter mined at all the Ghost task stops the operating system and passes control to the low level functions for haunting the system Checking the remaining energy before each update is very import ant since many sensor nodes are based on microcontrollers with program flash ROM Besides many applications are statically lin ked and hard to modify at runtime In both cases the node s pro gram memory must be erased before installing the new software A power breakdown between erasing and complete reflashing would indispensably corrupt the node and require manual repair if this is possible at all Indeed the energy requirements for updating a node may not be underestimated They depend on the electrical charac teristics of program and flash memory and on the image size The latter defines the number of read write cycles and in consequence the total du
54. now receives a list message with new nodes it changes the source address of the list message splits the list if required and forwards it them further If a forwarding node has to become a branching node it pre pares the overlay link to the source of the list message splits the message and forwards the new list messages further as described before This new branching node tells the source of the original list message which receivers it handles in the future Therefore the previous branching node removes the overlay link that previously was using this new branching node as forwarder Upon reactivation of the modified tree the overlay link connections are opened from the source via branching nodes to the the receivers New and old receivers then become aware of their new group membership or of the change of branching nodes for existing group member ships Nodes can also be removed using remove list messages accordingly and the forwarding and branching nodes have enough information as with adding new nodes to modify the resulting tree In the source driven centralized approach the source node determines all required branching nodes ahead Therefore the source also creates the complete distribution tree that is required for a multicast group The branching nodes are then notified process the information and further forward these notifications If new receivers need to be added to a tree while it is inactive the sour
55. of at most 118 bytes long which is passed to the MAC as a MSDU MAC Service Data Unit The MAC layer adds its header of 9 bytes converts it into MPDU MAC 59 Number of CFP Slots of CFP Slots Number CFP Slots needed sah CFP Slots requested x Number of CFP Slots 60 Length of Data Packets Bytes a 60 Length of Data Packets Bytes GTS Devices Figure 4 The number of CFP slots actually needed for the length of given data packets with acknowledgment request is shown in a whereas b depicts the same for unacknowledgment request The comparison for the bandwidth underutilization problem of the original standard with our approach is shown in c which highlights the major improvement given by our approach Protocol Data Unit and passes it to the PHY layer which also adds the header of 6 bytes Further 11 bytes are required for an optional acknowledgement request To re ceive acknowledgment device needs aTurnaroundTime 12 symbols to change its radio from the TX to RX mode or vice versa Additionally we have to consider the interframe spacing IFS which separates two successive frames sent by the device Its value is 12 symbols for the MPDU lt 18 oth erwise it is 40 symbols The number of CFP slots required for the device is shown in Equation 4 For an application where the devices genera
56. of link changes 25 4 UI 4 i i 0 0 500 1000 1500 2000 2500 3000 3500 4000 time rounds Figure 3 Comparison with network density 24 and the number of link changes are shown over time Ma halle provides very fast neighborhood detection and reaches 100 connectivity Due to the large density of 24 poten tial neighbors LEEP and the basic protocol are providing only a slow convergence Connectivity is not fully reached by these two approaches after adding nodes The number of link changes during the stable phases is also decreased in Mahalle 5 CONCLUSION The presented neighborhood management protocol Mahalle and the underlying adaptive link estimator have shown good results in terms of agility stability symmetry and stability Mahalle outperforms in terms of stabilization speed The next steps will be porting Mahalle to TinyOS and using it in a real long term deployment at our campus 6 ACKNOWLEDGEMENT We would like to thank Ting Fu Chen for his help in devel oping a preliminary version of Mahalle 7 REFERENCES 1 Gnawali TinyOS Extension Proposal TEP 124 The Link Estimation Exchange Protocol LEEP Feb 2007 http www tinyos net tinyos 2 x doc pdf tep124 pdf 2 Srinivasan and Levis RSSI is Under Appreciated In Proceedings of the Third Workshop on Embedded Networked Sensors EmNets 06 Harvard University Cambridge MA USA May 2006
57. of neighbors that a node can store a simple neighborhood protocol can lead to disconnected network Therefore a neighborhood management must ensure these four quality criteria This paper presents Mahalle a new neighbor management protocol based on adaptive link es timator Mahalle optimizes the neighborhood relations ac cording to agility stability symmetry and connectivity 2 RELATED WORK Research has mainly focused on link estimation so far Link quality can be assessed by the physical or logical proper ties Currently available hardware is providing these phys ical properties with a high accuracy 2 The link quality indicator LQI or received signal strength indicator RSSI are determined by the radio transceiver However the val ues of these metrics are highly hardware specific and must be calibrated for each hardware setting Packet reception rate PRR is a widely used logical prop erty indicating the link reliability 5 PRR is calculated from the rate of successfully received periodical broadcast packets Periodic sending is done by discretizing time into rounds In each round every node send its broadcast packet Besides an identifier of the sender these packets contain a sequence number that is used to determine packet loss In contrast to the physical properties PRR is the link relia bility experienced by applications but comes at the extra effort of periodical broadcasts The main drawback of all metric
58. or completely new protocols algorithms and strategies can be designed Building on the previous work the participants of FRONTS will examine strategies and algorithms which are focused on solving global tasks in the network This will result in a set of algorithms that contain a collective exploration of an unknown area by considering both active and passive mobility in the network algorithms for target tracking in dynamic environments with either a moving sink to which data must be sent or a mobile agent that follows the target and algorithms for network connectivity maintenance with actively moving nodes 3 EXPERIMENTS REPOSITORY The previous section described the project objectives A decisive point are the practical aims of the project such as evaluation and validation of developed algorithms We will contribute an Evaluation Package to obtain valuable feed back for the project s theoretical insights This package will consist of a WSN simulator appropriate hardware and at least the provision of one sensor network testbed composed of about 50 sensor nodes The common central testbed will be accessible via Internet for all project partners to test pro tocols algorithms and applications We had to cope with the following requirements concerning suitable software and hardware technologies e We need a simulation framework that aims particu larly at high level algorithms and a quick and easy development process e
59. recent advances in providing real world measurements of basic wireless data communication primitives in wireless sensor networks At the time of writ ing we have been focusing on end to end communication be tween a source and destination node We believe that a basic understanding of such basic primitives will as well provide insights in more advanced sensor network specific data col lection primitives Moreover sensor networks are not only about data collection trees For instance a data sink issu ing a sensor request into a specific geographic region a data sink issuing a request to a certain area identified by a spe cific node or a data source trying to connect to a specific actuator are just a few examples which show that sensor networks require basic unicast communication primitives as well Data communication protocols can be classified in topology based and geographic ones Surveys can be found in 9 and 3 respectively Topology based communication requires either proactive or reactive exchange of global information before message forwarding from source to destination can take place Geographic message communication in contrast does not require such global message exchange In such routing mechanism it is assumed that each node knows its current physical location Based on this additional informa tion a routing decision can be performed in a pure localized manner i e a forwarding decision requires only informa tion abou
60. requested from the cross layer inter face by REMC and SNOMC TSS TCP Support for Sensor Nodes 2 supports optimizations of TCP specific mecha nisms such as intermediate caching local retransmission and acknowledgment recovery and regeneration Additional energy saving is achieved by disabling the radio interface by the MAC layer when no transmission is required 71 Cross Layer Interface Figure 2 Protocol stack of a wireless sensor node using reliable unicast and multicast communication 3 DISCUSSION amp CONCLUSION We have shown several possible design concepts for reliable multicast in WSNs Each approach its own advantages and limitations depending on the scenario in which it is used For small scale WSNs the centralized approach helps to save energy and resources on the sensor nodes because the source generally the base station is handling the tree man agement The sensor nodes do not need to store a lot of status information The de centralized approach is useful in large scale environ ments where robustness and easier tree construction can be achieved by letting the sensor nodes manage the tree con struction and group handling themselves The Overlay Multicast approach is easy to implement but triggers more control messages in the underlaying layers as well as in the overlay management layer By distinguishing between active and inactive trees we can reduce the energy consumption due to the fact that we onl
61. shown that all clients were able to communicate to each other also in the multi hop manner 5 CONCLUSION amp OUTLOOK In this paper an approach for Multi Client Systems MCS for management of wireless sensor networks has been presented The first MCS implementation for a mobile device running on a Windows Mobile operating system was introduced as well as the foreseen backbone architecture The wireless backbone on the second level is build up on full meshed WLAN router based on OLSR It is shown that also the MCS can be part of this network Future work will contain further development of the Multi Client System including user right management for e g guest habitant and administrator Topology visualization in form of a typical tree view or similar graphical representations on top of a bitmap like a floorplan is planned too Studies of user acceptance are planned as well to get feedback for the enhancement of the GUI in the wireless sensor networking domain 6 ACKNOWLEDGEMENT The presented work will be part of the AutHoNe project results AutHoNe is a BMBF German Federal Ministry of Education and Research funded project In the German part of the project the Partners are Fraunhofer FOKUS Hirschmann Automation and Control GmbH Siemens CT and TU Munich The project is being carried out as part of a CELTIC cluster within a EUREKA initiative 7 REFERENCES 1 IEEE 802 15 4 4 Homepage The Institute of Electrical and Elect
62. switches from agile into stable mode after 30 rounds in case that the PRR is greater than PRR aa during this period for a more accurate estimation This mechanism is called raising mechanism which improves the crossing time of ALE compared to EWMA In Figure 1 the crossing time of the different estimators is compared The crossing time of the stable EWMA is very high compared to the agile mode Due to the threshold PRR a and the raising mechanism the crossing time of ALE is even faster than the agile EWMA especially for good links Observe that the crossing time of ALE is smaller than 30 for all high values of PRR This justifies the choice made above for the number of rounds to switch from the agile to stable mode Another mechanism that improves ALE is the so called drop ping mechanism With this mechanism a good link is trusted even if currently no communication is possible Doing so the stability of the resulting network topology is increased Whenever a good link with PRR gt PRRgooa suffers from consecutive packet loss the PRR is kept constant for 60 rounds This number is a parameter of ALE and specifies the time that is conceded to a node to recover from a tran sient failure In Figure 2 the behavior of EWMA and ALE is compared over time As shown in both figures ALE outperforms the agile mode of EWMA in terms of mean square error Fig ure 2 a shows a link with a fixed real PRR of 0 89 and a short link interruption of 50 p
63. test if the Safe TinyOS toolchain installation is working properly After marking the array index variable as symbolic we immediately de tected the known out of bound pointer error int ratio event message_t Receive receive message_t bufPtr void payload uint8_t len if len sizeof receive_fail_msg_t return bufPtr else receive_fail_msg_t rcm receive_fail_msg_t payload possible division by zero here depending on received message ratio rem gt good rcm gt total return bufPtr Listing 5 Possible division by zero error make kleenet test KLEE ERROR divide by zero Listing 6 KleeNet detects the div by zero error In listing 5 we demonstrate the usability of KleeNet for finding possible bugs without annotating application source code any further A received message is processed without sanitizing the received message which is a typical fault of students new to programming KleeNet rapidly detects this mistake and warns the developer Lst 6 Overall after our initial tests we have confirmed the follow ing key benefits of KleeNet Usability A programmer can test the code with minimum manual effort and without any previous knowledge about the checking tool Coverage KleeNet covers all possible execution paths and checks all possible data values before application deploy ment Integration KleeNet is invoked by simply adding an extra build flag enabling the perma
64. the networking of different WSN and other networking components in the home area is required For this purpose the German project Autonomous Home Networking AutHoNe has been established and is funded by the German Ministry of Education and Research BMBF One of the goals of AutHoNe is to provide users a consistent and transparent approach to access data sources and services in the home range in an easy way including functions for the definition of rules and constraints to influence the autonomous behavior of the network elements A suitable graphical user interface for these functions will be one result of the project and it is introduced in this paper This kind of user interface will help normal users to use and BMBF AutHoNe label 01BN0702 manage their home networks even in the complex network environments of the future The remainder of this paper is structured as follows Chapter 2 presents related work Chapter 3 describes the developed graphical user interface with its functionality Section 4 introduces the foreseen backbone architecture for the Multi Client System Chapter 5 will give an outlook for further implementations and will conclude this paper 2 RELATED WORK The project e home 4 contains a home network on the basis of Local Operating Networks LON This technology allows an information exchange between single devices and the creation of scenarios for single actuators In 5 a home automation system is present
65. the nodes the more link breaks occur in the network Gerharz et al 8 simulated the probability distribution function of the link duration time in a network where the nodes move according to a random waypoint mo bility model They showed that the end to end reliability can be increased by selecting the next hop depending on the estimated link duration time Another approach is to take advantage from nodes with correlated movement since rela tive movement speed is mainly responsible for link breaks These nodes should be selected as next hop due to their more reliable links as a consequence of their correlated movement The problem is to detect the correlated node movement if no position information is available Consider a scenario on a highway in one direction There will be some fast driv ing cars some with average speed and some slow moving trucks On one hand if there are more trucks on the road than other cars the routing protocol should prefer trucks to forward data since their movement is strongly correlated re sulting in a high link duration time On the other hand fast movement cars are also able to build a stable network if they represent the majority of cars on the highway Thus a met ric is required that is able to estimate the current relative movement speed of a node Each node is able to estimate its speed by keeping track of the presence of its surrounding nodes A change in its neighborhood indicates that the node
66. the verge to industrial use Several aspects of wireless sensor networks are appealing for industrial use there is no need for wiring which may be a huge cost factor or simply not practical e g in the case of moving parts and as the installation is very flexible it can be used also for temporary installations However it is hard to transfer results of the latest research di rectly to industrial wireless sensor networks as assumptions on the setup and environment made by research are significantly differ ent compared to important industrial application scenarios Section 2 describes three typical industrial use cases for wireless sensor networks Section 3 sets into contrast assumptions and requirements of academic sensor networks and industrial sensor networks in several areas Section 4 concludes the paper 2 SCENARIOS Some typical use cases are described that are drawn from process and factory automation industry used e g in refineries or manu facturing automation to illustrate typical industrial application environment of wireless sensor networks 2 1 Condition Monitoring A condition monitoring system supervises a set of machines to obtain information about their condition and to schedule mainte nance only if needed thereby reducing down time of machinery and costs Hans Joachim Hof Siemens AG Corporate Technology hans joachim hof siemens com Rudolf Sollacher Siemens AG Corporate Technology rudolf sollacher si
67. to the group s ad dress and the routers in the network then distribute the data according to the multicast tree that has been setup before Although IP Multicast has been available for a while it has not been widely deployed in the Internet today due to dif ferent reasons configuration ISP agreements etc To offer multicast functionality to the end user the concept of Appli cation Level Multicast 3 ALM often also called Overlay Multicast has been introduced With ALM which is based on the Peer to Peer 4 P2P paradigm end systems build the multicast tree among themselves rather than relying on routers to handle multicasting on their behalf Numerous research has been done about multicast in WSNs In 5 a multicast protocol called BAM Branch Aggrega tion Multicast is presented which supports single hop link layer multicast and multi hop multicast via branch aggre gation VLM Very Lightweight Mobile Multicast 6 is a multicast routing protocol for sensor nodes which is imple mented on top of the MAC protocol It provides multicast from a base station to any sensor node unicast connections from a sensor node to the base station and supports mobil ity In 7 the authors present an effective all in one solution for unicasting anycasting and multicasting in wireless sen sor networks and wireless mesh networks The authors of 8 adapt ADMR Adaptive Demand driven Multicast Rout ing a multicast protocol for M
68. too far from the sinks more nodes are involved in the routing of messages increasing the computational and energy cost There exist some other parameters that can be useful to balance the load among all nodes in the network When choosing the next hop it is preferable to use lightly loaded neighbors so as to reduce the number of collisions and not to exhaust the energy of nodes An equivalent met ric to the number of messages sent is the number of sources a node serves Additionally the number of served sources is more stable than the messages sent as it considers the traffic coming from a source independently from the sinks it is addressed to It would be desirable to have certain information about what the path looks like in the next hops before choosing a certain neighbor Since the information available in sensor networks is rather expensive to get and to store nodes can only count on local information The number of paths is defined as the number of source sink paths passing through a given node By counting the number of paths a node can get an insight of how loaded the next nodes will be because at some moment these paths are split In order to balance the load evenly through the network neighbors that have little remaining energy should be avoided as relay nodes As a metric for measuring the remaining en ergy we consider the number of messages a node has sent since it is operative In case the batteries are replac
69. und Clientsysteme notwendig sein 5 1 2 Semantische Anforderungen Die semantisch eindeutige Verwendung von Ausdr cken ist einer der wichtigsten Punkte bei der korrekten Erstellung von Datenmodellen in SensorML und O amp M Dokumenten SWE Spezifikationen beschreiben zwar den syntaktischen Aufbau von Informationen jedoch nicht ihren semantischen Inhalt Die semantisch unscharfe Verwendung von Ausdr cken erschwert die automatische Verarbeitung und Lesbarkeit f r Au enstehende Um die verwechslungsfreie Anwendung von Begriffen wie Ph nomennamen Sensortypen und Einheiten zu gew hrleisten ist die Schaffung von kontrollierten Vokabularen notwendig Ein kontrolliertes Vokabular erm glicht das Spezifizieren von Begriffen durch die Verkn pfung mit Metadaten und die Zuordnung in spezielle Anwendungsbereiche 6 Fazit Die Entwicklung ressourcensparender pr ziser Lokalisierungs algorithmen erh hen die Skalierbarkeit und Robustheit von autonomen ad hoc Geosensornetzwerken An der Universit t Rostock wurden im aktuellen Projekt Algorithmen entwickelt die in einem GSN energieeffizient zur Lokalisierung eingesetzt werden k nnen Eventuelle Ausrei er und grobe Fehler sind in 87 einem ersten Schritt mit dem ACL Algorithmus aufgedeckt und eliminiert worden Die geod tische Ausgleichung stellt hierbei Mittel zur Verf gung die f r die weitere Verbesserung vorgestellter Algorithmen eingesetzt werden k nnen Dabei stehen eine genauere
70. used so an automatic de activation of receiving is not possible In contrast the driver module in REFLEX knows that it is used because it is part of the event flow Additionally the receiving driver module is managed by the underlying group management and thus is only active when implicitly activated 6 Conclusion We presented an efficient way of determining the deepest possible sleep mode in event driven systems and its imple mentation in the REFLEX operation system The presented power management is lightweight and hardware indepen dent Measurements show that this approach seams to per form around 50 better than the one of TinyOS for the given application and regardless the noise of the measurements Further examination of additional applications is needed to confirm the results 7 References 1 Texas Instruments MSP430 series Webpage http www msp430 com 2 J Hill R Szewczyk A Woo S Hollar D Culler and K Pister System Archi tecture Directions for Networked Sensors In the 9th International Conference on Architectural Support for Programming Languages and Operating Systems ASPLOS IX 2000 3 TinyOS Microcontroller Power Management Documentation TEP112 Web page http www tinyos net tinyos 2 x doc html tep112 html 4 C C Han R S Ram Kumar E Kohler and M Srivastava A dynamic oper atingsystem for sensor nodes In Proc of the 3rd international conference on Mobile systems applications and ser
71. used components current sensor node platforms can perform 4 000 to 2 000 000 CPU cycles 8 in stead of transmitting a single byte while still maintaining a positive energy balance 3 VISION The vision of our research is to develop a modular compres sion framework which can be deployed on nodes in a WSN It comprises a set of different compression and decompres sion algorithms which can be adaptively selected and even combined to achieve additional savings By compiling the algorithms into packages different compression methods can be dynamically selected exchanged and updated during run time As many current sensor network applications follow pre defined structure definitions in their output data format ting different compression algorithms will typically produce differently sized output The framework can either rely on application defined preferences regarding the preferred com pression engine or compress an exemplary set of sensor data with its available compressors prior to selecting the algo rithm with the highest energy efficiency While this analysis might consume more energy before regular operation it only needs to be performed once and will subsequently use the locally optimal compression algorithm Additionally applications can also specify delay bounds for the compression of their packets This mechanism ensures that lengthy algorithms are avoided when low latency trans missions are required
72. used in the field An additional advantage is that the AirFlash module can be easily integrated into any TinyOS application There it stays in the background when not needed and thus keeps the main application running smoothly Furthermore additional features can be easily implemented since the TinyOS part is completely open source A feature that is going to be added is the capability to disseminate an image to a whole WSN at once 7 REFERENCES 1 Hui J et al 2008 TOSBootM nc URL http tinyos cvs sourceforge net 2 Crossbow Technology 2008 XMesh User s Manual Retrieved June 05 2008 URL http www xbow com Products productdetails aspx sid 154 3 Chlipala A 2003 Deluge Data Dissemination for Network Reprogramming at Scale Technical Report UC Berkeley 4 Crossbow Technology 2008 IRIS data sheet Retrieved June 02 2008 URL http www xbow com Products productdetails aspx sid 264 5 Atmel Corporation 2007 ATmegal281 data sheet Retrieved June 05 2008 URL http www atmel com dyn resources prod_documents doc2549 pdf 6 UC Berkeley 2004 TinyOS Community Forum Retrieved June 05 2008 URL http www tinyos net 7 Boah A 2008 Over the Air Programming of Wireless Sensor Nodes Project Thesis Hamburg University of Technology Deadlock free Resource Arbitration for Sensor Nodes Heiko Will Jochen Schiller Freie Universit t Berlin Berlin Germany baar hwill schiller inf fu berl
73. usually equipped with a microcontroller and a radio transceiver A challenge in large networks is to keep the firmware of the nodes up to date Once the network is deployed it is hardly possible to access every node physically and to upload a new firmware Therefore this paper describes a technical solution for this problem using the radio transceiver for the upload procedure The presented solution focuses on heterogeneous WSNs that contain several groups of nodes and where each group fulfills a different task Keywords WSN OTAP Network Reprogramming AirFlash 1 INTRODUCTION WSN technology is a promising candidate to provide monitoring over large areas and for long time periods The network consists of a large number of small and inexpensive nodes Each node contains a sensing computation and communication unit The WSN technology will be applied in dangerous areas where human beings have no access Therefore network maintenance or even firmware updates cannot be accomplished based on a physical access to the nodes and to the network But the nodes can be accessed over a wireless connection Thus in order to update the firmware on a single node or in the entire network a procedure based on wireless access will be used In this paper an Over the Air Programming OTAP technique will be used which is based on the wireless access and transmission scheme In this case an image of the node firmware will be transmitted via the wirel
74. weeks although they cannot be done in one continuous session 1 MOTIVATION Wireless sensor networks WSN have been in the focus of research in the last decade and are deemed mature for many application areas for which they were intended to Several fundamental problems have been identified e g routing ad dressing transport layer issues and middleware various ap proaches are consequentially proposed Unfortunately most knowledge and understanding is based either on simulation studies or small deployments of wireless sensor networks Few real world WSN deployments resemble the envisioned kind of networks where vast numbers of nodes solving a col lective task have to rely on limited energy resources While these setups are existing e g wild life monitoring 13 many installations are in urban environments sensors mon itor decrepit bridges and buildings support elderly in as sisted living facilities or are part of home automation sys tems These kind of applications demand WSNs to integrate into existing network infrastructures like a wireless mesh network WMN New challenges emerge since the applica tion specific devices are often performance limited and use feature reduced operation systems They were not devel oped with integration and support of common often heavy weight protocols in mind We discuss the integration aspect twofold On one hand the WSN can be part of the overall architecture provided
75. 34 this mode too The current state of the graphical user interface implementation of a MCS is shown in Figure 2 with the rule editor interface 4 BACKBONE ARCHITECTURE This section describes the envisioned backbone architecture of the AutHoNe project The backbone architecture design follows a two tier approach On the lowest level the first tier several wireless technologies can exists using different kinds of communication technologies These technologies are connected through a second full meshed WiFi WLAN network to bring them into the IP world This IP infrastructure is the second tier To build up a WLAN backbone for the WSN technologies the routers WL500gP from ASUS are used 266MHz 8MB flash The routers are able to perform ad hoc networking including mesh networking based on OLSR 6 This system is running on a Linux operating system based on OpenWRT 7 In the prototype evaluation of the system the routers has been equipped with the Integration CompXS USB IEEE 802 15 4 dongle This dongle is fully compliant to the ZigBee 2006 specification 8 and allows a full speed USB 2 0 communication It has an integrated antenna and the typical transmission range is above 30m An Xbee module is included for a second revised version of the gateways This provides an USB port and gives the opportunity to use antennas for enlarging the observing area To enlarge the covered area a seven dBi omnidirectional rod antenna has been used For t
76. 9 26 Mar 2008 10 D Koutsonikolas S Das Y Hu and I Stojmenovic Hierarchical Geographic Multicast Routing for Wireless Sensor Networks International Conference on Sensor Technologies and Applications SENSORCOMM 07 Valencia Spain Oct 2007 11 J A Sanchez Ruiz and I Stojmenovic Energy Efficient Geographic Multicast Routing for Sensor and Actuator Networks Computer Communications 30 13 2519 2531 Jun 2007 112 OMNeT Discrete Event Simulation System Website http www omnetpp org 13 A Dunkels Gr nvall Voigt Contiki a Lightweight and Flexible Operating System for Tiny Networked Sensors EmNetS 04 Tampa FL USA Nov 2004 First Results of a Performance Comparison of Dynamic Source Routing versus Greedy Routing in Real World Sensor Network Deployments Extended Abstract Hannes Frey University of Paderborn Pohlweg 47 49 33098 Paderborn Germany hannes frey uni paderborn de ABSTRACT This work covers our first results of an empirical perfor mance comparison between dynamic source routing and greedy routing in a real world wireless sensor network de ployment The test environment is based on a 7 x 7 grid of 49 battery operated Tmote Sky sensor nodes We briefly sketch the protocol implementations and the experimental setup used in this work and present our findings from the measurement data we have gathered so far 1 INTRODUCTION This work describes our
77. ANETS on a real wireless sensor node MICAz They show that the adaption is not a trivial task and a number of problems have to be solved The authors of 9 analyze IP Multicast and show that it is possible to use it in WSNs Further there are several multi cast solutions for WSNs which are based on the geographical position of the sensor nodes in the network 10 11 1 3 Structure of the Paper The remainder of the paper is structured as follows In the next Section multicasting in WSNs is described showing the challenges different designs of multicast and a protocol stack for IP based communication is proposed Section 69 discusses the advantages and limitations of our proposed solutions An outlook in Section 4 closes this paper 2 MULTICAST IN WIRELESS SENSOR NETWORKS 2 1 Challenges Due to the nature of WSNs the IP Multicast implementa tion can not be simply ported from existing solutions for wired networks In WSNs energy memory and CPU power are limited This implies the following challenges for mul ticast in such environments In wired networks routers are handling packet replication and forwarding clients just send and receive simple IP UDP datagrams On the other hand WSNs would need to introduce the router function ality for IP Multicast management into each sensor node Group management is normally concentrated on the routers that communicate with each other to handle multicast trees The management for
78. After the API operation has finished the device is deactivated It is left to the power man agement to power down inactive devices later During both activation and power down the optional control functions of involved devices are called Activation of virtual devices from powered off state is a key operation It is based on the fact that API operations are synchronous and no other API operations are to be invoked from within interrupts For few operations it can be necessary to access virtual device methods from interrupts e g timers At this point it is up to the developer to manually activate another device and switch back to the previous one after wards Special care has to be applied to verify that any device which might be preempted during an operation is able to pause and continue 223 Multiplexing Virtual Devices Using hubs virtual devices are multiplexed on a shared re source Only a single virtual device can be selected at a hub at any time This addition is necessary when interrupt events need to be forwarded to the device which currently uses the associated device and to provide a mechanism for default configuration Hubs also help to reduce overhead for device deactivation because usually all virtual devices on a hub de device access activate powered off device access reactivate access finished power management deactivate V other virtual device on same hub becomes active V depen
79. Beitr ge des Fachgespr chs Drahtlose Sensornetze 2008 zusammengefasst Ziel dieses Fachgesprachs ist es Wissenschaftlerinnen und Wissenschaftler aus diesem Gebiet die M glichkeit zu einem informellen Austausch zu geben wobei immer auch Teilnehmer aus der Industrieforschung willkommen sind die auch in diesem Jahr wieder teilnehmen Das Fachgespr ch ist eine betont informelle Veranstaltung der GI ITG Fachgruppe Kommunikation und Verteilte Systeme www kuvs de Es ist ausdr cklich keine weitere Konferenz mit ihrem gro en Overhead und der Anforderung fertige und m glichst wasserdichte Ergebnisse zu pr sentieren sondern es dient auch ganz explizit dazu mit Neueinsteigern auf der Suche nach ihrem Thema zu diskutieren und herauszufinden wo die Herausforderungen an d e zuk nftige Forschung berhaupt liegen Das Fachgespr ch Drahtlose Sensornetze 2008 findet in Berlin statt in den R umen der Freien Universit t Berlin aber in Kooperation mit der ScatterWeb GmbH Auch dies ein Novum es zeigt dass das Fachgespr ch doch deutlich mehr als nur ein nettes Beisammensein unter einem Motto ist F r die Organisation des Rahmens und der Abendveranstaltung geb hrt Dank den beiden Mitgliedern im Organisationskomitee Kirsten Terfloth und Georg Wittenburg aber auch Stefanie Bahe welche die redaktionelle Betreuung des Tagungsbands tibernommen hat vielen anderen Mitgliedern der AG Technische Informatik der FU Berlin und nat rlich a
80. CHITECTURE The purpose of our resource arbitration architecture is both to provide a hardware abstraction layer on top of the underlying hardware and to make it possible to switch off hardware peripherals when they are not used to conserve energy The operating system and its applications can run on different platforms using platform specific implementations of the HAL Stackable hardware platforms like the ScatterWeb MSB 430 1 that allow run time addition of new boards and hardware devices present a novel challenge for the resource arbitration architecture and HAL The architecture must be able to dynamically adapt to new device drivers being added at run time potentially stored in on board memory of the additional hardware Our architecture supports this by using one way dependencies and allows for run time extension Our resource arbitration architecture is deadlock free by design Existing approaches to resource arbitration for sensor nodes 7 rely on explicit locking to determine when a peri pheral should be switched off Explicit locking can however lead to deadlock In contrast our architecture uses atomic System Design without Hardware Abstraction Operating System and Applications Radio Driver SD Card Driver Driver SPI Driver SPI Driver SPI Driver UART Driver USARTO USART1 SPI SPI 2 MHz SPI UART 56 kb s Compass Terminal Radio SD Card Circuit Circuit Clock System System Desig
81. Das Projekt gliedert sich in zwei Teile Zun chst soll eine systematische bersicht ber Einsatzsze narios f r Sensornetze die schon umgesetzt wurden oder von 55 N Sicherheits Einsatzszenarios mechanismen techn Voraussetzungen Schutzbedarf Bewertung 4 Klassifikation Zuordnung Abbildung 1 Vorgehen zur Klassifikation Ein satzszenarios denen dies in naher Zukunft zu erwarten ist erstellt werden Dabei soll versucht werden die technischen Designmerkma le Abschnitt 2 4 die durch das Szenario festgelegt werden zu bestimmen F r ein konkretes Einsatzszenario kann auch eine gewisse Bandbreite bei einem Merkmal vorliegen d h unterschiedliche Implementierungen f hren zu unterschied lichen technischen Merkmalen Gleichzeitig soll m glicher Schutzbedarf in den einzelnen Szenarios ermittelt werden d h es soll gepr ft werden welche Gef hrdungen im jewei ligen Einsatzszenario auftreten k nnen und wie gravierend diese Gef hrdungen sind Beispielsweise hat ein System zur Grenz berwachung einen hohen Schutzbedarf ein System zur Umwelt berwachung ein eher niedriges Anhand der gefundenen Merkmalsauspr gungen soll anschlies send eine Klassifikation der Einsatzszenarios stattfinden Ei ne Klasse soll dabei aus Szenarios bestehen die hnliche oder die gleiche Auspr
82. G seed each sensor periodically changes the transmission frequency and thus it dynamically changes its acceptance intervals The amount of false positives i e legitimate frames discarded due to signal variation is low at 0 78 sures an RSS increase of 10dBm Figure 3 b Consequently a single frequency change may cause RSS to increase or decrease on different indoor receivers at the same time and the same position of the sender This finding has important implication in a security context it allows us to shuffle the order of sensors RSS merely by switching between different transmission frequencies without changing the placement of the sensors For example if both indoor sensors transmit on channel 11 the order of their RSS is approxi mately equal 1 However changing the transmission frequency e g using channel 19 results in gt This prop erty we exploit to establish a dynamic configuration feature in our protection mechanism 3 COUNTERACTING INJECTION ATTACK BY WIRELESS Leveraging the unpredictable and chaotic nature of signal propaga tion our protection method is based on a single assumption two wireless transmissions from different physical positions cannot pro duce exactly the same signal propagation properties To facilitate this wireless peculiarity we propose and implement two mecha nisms for crypto free authentication of frames exchanged within an indoor WSN 1 acceptance int
83. HARTMUT RITTER KIRSTEN TERFLOTH GEORG WITTENBURG JOCHEN SCHILLER HRSG 7 GI ITG KUVS FACHGESPRACH DRAHTLOSE SENSORNETZE FREIE UNIVERSIT T BERLIN INSTITUT F R INFORMATIK TAKUSTR 9 14195 BERLIN TECHNICAL REPORT B 08 12 FREIE UNIVERSIT T BERLIN FACHBEREICH MATHEMATIK UND INFORMATIK 7 GI ITG KUVS FACHGESPRACH DRAHTLOSE SENSORNETZE AM 25 UND 26 SEPTEMBER 2008 IN BERLIN HARTMUT RITTER KIRSTEN TERFLOTH GEORG WITTENBURG JOCHEN SCHILLER HRSG INHALT BUG HUNTING IN SENSOR NETWORK APPLICA R Sasnauskas J Bitsch Link Alizai Wehrle OVER THE AIR PROGRAMMING OF WIRELESS SENSOR NODES nnnnnneensnnennenennn 11 M Stemick A Boah H Rohling DEADLOCK FREE RESOURCE ARBITRATION FOR SENSOR NODES 2 15 M Baar H Will J Schiller A Dunkels AN APPROACH TOWARDS ADAPTIVE PAYLOAD COMPRESSION IN WIRELESS SENSOR WORKS ee 19 A Reinhardt Hollick Steinmetz CONNECTIVITY AWARE NEIGHBORHOOD MANAGEMENT PROTOCOL IN WIRELESS SENSOR ea 23 Ch Weyer S Untersch tz V Turau CHALLENGES IN SHORT TERM WIRELESS LINK QUALITY ESTIMATION ccccccsscccssscesseceeseeeseeees 27 Alizai Landsiedel Wehrle A Becher SOMSED AN INTERDISCIPLINARY APPROACH FOR DEVELOPING NR Seesen 29 S Georgi Ch Weyer Stemick Ch Renner Hackbarth U Pilz J Eichmann Pilsak
84. Hoc and Sensor Networks WMAN April 2005 16 P Wang and T Wang Adaptive routing for sensor networks using reinforcement learning In Proc of the Sixth IEEE International Con ference on Computer and Information Technology CIT 2006 17 Y Wang M Martonosi and L S Peh A supervised learning ap proach for routing optimizations in wireless sensor networks In Proc of the 2nd International Workshop on Multi hop Ad hoc Networks from Theory to Reality REALMAN 2006 18 Welsh Werner Allen Lorincz Marcillo J Johnson M Ruiz and J Lees Sensor networks for high resolution monitor ing of volcanic activity In SOSP 05 Proc of the twentieth ACM symposium on Operating systems principles 2005 19 G Wittenburg K Terfloth F L Villafuerte T Naumowicz H Ritter and J H Schiller Fence monitoring experimental evaluation of a use case for wireless sensor networks In In Proc of 4th European Conference on Wirless Sensor Networks EWSN 2007 20 A Woo T Tong and D Culler Taming the underlying challenges of reliable multihop routing in sensor networks In Proc of the Ist Inter national Conference on Embedded Networked Sensor Systems Sen Sys November 2003 21 J Zhao and R Govindan Understanding packet delivery perfor mance in dense wireless sensor networks In Proc of the 1st Interna tional Conference on Embedded Networked Sensor Systems SenSys November 2003 22 G
85. ISBN 0 471 68472 4 Wiley Oct 2005 4 D Gay Levis von Behren Welsh E Brewer and D Culler The nesc language A holistic approach to networked embedded systems In In ACM SIGPLAN Conference on Programming Language Design and Implementation 2003 5 M Heissenbiittel and T Braun BLR Beacon less routing algorithm for mobile ad hoc networks Elsevier s Computer Communications Journal 2003 6 J Szewczyk A Woo S Hollar D Culler S J Pister System architecture directions for networked sensors In Architectural Support for Programming Languages and Operating Systems pages 93 104 2000 7 D Johnson and A Maltz Dynamic source routing in ad hoc wireless networks In Mobile Computing Kluwer Academic Publishers 1996 8 J Kuruvila A Nayak and I Stojmenovic Greedy localized routing for maximizing probability of delivery in wireless ad hoc networks with a realistic physical layer In CD Proceedings of the 1st International Workshop on AlgorithmS for Wireless And mobile Networks A SWAN at MobiQuitous Aug 2004 0 Royer and C K Toh A review of current routing protocols for ad hoc mobile wireless networks IEEE Personal Communications 6 2 46 55 Apr 1999 10 Takagi and L Kleinrock Optimal transmission ranges for randomly distributed packet radio terminals Transactions on Communications 32 3 246 257 Mar 1984 75
86. Italy January 2008 Poster Presentation 13 Jens B Schmitt Frank A Zdarsky and Ivan Martinovic Performance Bounds in Feed Forward Networks under Blind Multiplexing Technical Report 349 06 University of Kaiserslautern Germany April 2006 APPENDIX A BASICSENSOR NETWORK CALCULUS This is a very basic overview of the Sensor Network Cal culus SNC Detailed explanation of the SNC can be found in 2 To apply the SNC the network topology has to be known to some degree For example a tree structured network topology with a sink at the root and n sensor nodes can be used Next the network traffic has to be described in terms of so called arrival curves for each node An arrival curve defines an upper bound for the input traffic of a node Leaf nodes in the network have to handle traffic according to the sensing function they perform for example a node might sense an event and create a data packet at a maximum rate of one packet every second This sensing pattern can be expressed as an arrival curve a Non leaf nodes handle traffic according to their own sensing pattern and the traffic they receive from other nodes To calculate the output the so called service curve is used The service curve specifies the worst case forwarding capabilities of a node The necessary forwarding latencies 48 are defined by the nodes forwarding characteristics From the arrival and service curves it is possible to cal c
87. Knoten und jeder einzelne Knoten ber hin reichend viele Nachbarn in Funkreichweite verf gt also dass die Topologie ausreichend dicht ist e Einige Schl sselvereinbarungsverfahren wie etwa 10 ben tigen einzelne leistungf hige Knoten als Cluster heads was zu heterogener Hardware im Netz f hrt Damit ist h ufig nicht klar f r welches Einsatzszenario sich ein konkreter Sicherheitsmechanismus tats chlich eignet oder ob so ein Einsatzszenario berhaupt existiert Zudem bauen einige Verfahren auf anderen auf z B wird f r viele abgesi cherte Routingverfahren eine anf ngliche Schl sselvereinba rung zwischen benachbarten Knoten angenommen so dass diese h herwertigen Verfahren wiederum auf solche Szena rios einschr nkt sind die das zugrundeliegende Verfahren erlauben 4 KLASSIFIZIERUNG VON EINSATZSZE NARIOS Aus den oben beschrieben Gr nden plant das BSI das ge nannte Projekt zur Klassifizierung von Einsatzszenarios f r Sensornetze in dem dar ber hinaus Einsatzempfehlungen f r konkrete Sicherheitsmechanismen in den einzelnen Sze narios entwickelt werden sollen Diese Klassifizierung soll ei nerseits Hilfestellung bei der Auswahl von Sicherheitsmecha nismen in der Praxis bieten und andererseits noch zu l sende Forschungsfragen hinsichtlich Sicherheit in drahtlosen Sen sornetzen offenlegen Dar ber hinaus soll die Klassifizierung eine Vergleichbarkeit unterschiedlicher Mechanismen erlau ben
88. Likewise nodes can stay up to date by observing their environment for newer versions of their own software In contrast to the common method where a source node gateway pushes the software to each affected node cloning allows updates to be pulled over the whole network without explicit routing or flooding This desired virus like spreading is extremely useful for very large networks where the in dividual handling of each single node would become too complex Of course the network protocol must remain compatible between versions to support any kind of long term self maintenance 2 2 Security Secure data transmission is very critical in wireless sensor net works This is especially true for the deployment and maintenance of software and configuration images since a tapping attacker might draw conclusions about operation and vulnerabilities of the overall network Of course prevention of hostile code injections or node takeovers is also mandatory The obvious technique is to encrypt all communication Yet the used algorithm must provide an adequate security level at appro priate resource requirements Furthermore the key selection needs careful consideration Where an individual key for each node pro vides highest security especially if nodes can be stolen along with the secret a simultaneous updating of several nodes by a single encrypted image 15 not possible any more The same holds for clo ning since key exchange methods among t
89. M Srivastava Harbor software based memory protection for sensor nodes In IPSN 07 Proc of the 6th international conference on Information processing in sensor networks 2007 C Lattner and V Adve LLVM A Compilation Framework for Lifelong Program Analysis amp Transformation In CGO 04 Proc of the international symposium on Code generation and optimization 2004 P Levis S Madden J Polastre R Szewczyk K Whitehouse A Woo D Gay J Hill M Welsh E Brewer and D Culler TinyOS An Operating System for Sensor Networks In Ambient Intelligence 2005 J Parr and W Quong a predicated Il k parser generator Software Practice and Experience July 1995 H Shen J Wang L Ping and K Sun Securing C Programs by Dynamic Type Checking In ISPEC 2006 N Volanschi A Portable Compiler Integrated Approach to Permanent Checking In ASE 06 Proc of the 21st IEEE ACM International Conference on Automated Software Engineering 2006 Over the Air Programming of Wireless Sensor Nodes Martin Stemick m stemick tuhh de Anthony Boah anthony boah tuhh de Hermann Rohling rohling tuhh de Hamburg University of Technology Eissendorfer Str 40 21073 Hamburg ABSTRACT Self organizing Wireless Sensor Networks WSNs have gained much attention in recent research and industrial activities Such WSNs enable measurements and information dissemination over large areas A common node in a WSN is
90. SA December 2007 7 Jens B Schmitt Frank A Zdarsky and Markus Fidler Delay Bounds under Arbitrary Multiplexing Technical Report 360 07 University of Kaiserslautern Germany July 2007 8 Jens Schmitt Frank A Zdarsky and Ivan Martinovic Improving performance bounds in feed forward networks by paying multiplexing only once In 14th GI ITG Conference on Measurement Modeling and Evaluation of Computer and Communication Systems MMB 2008 Dortmund Germany March 2008 GI ITG 9 Jean Yves Le Boudec and Thiran Network Calculus A Theory of Deterministic Queuing Systems for the Internet volume 2050 of Lecture Notes in Computer Science Springer 2001 10 Anis Koubaa Mario Alves and Eduardo Tovar Modeling and Worst Case Dimensioning of Cluster Iree Wireless Sensor Networks In 27th IEEE International Real Time Systems Symposium RTSS 06 pages 412 421 Rio de Janeiro Brazil 2006 IEEE Computer Society 11 Huimin She Zhonghai Lu Axel Jantsch Li Rong Zheng and Dian Zhou Traffic splitting with network calculus for mesh sensor networks In FGCN 07 Proceedings of the Future Generation Communication and Networking FGCN 2007 pages 368 373 Washington DC USA 2007 IEEE Computer Society 12 Petcharat Suriyachai Utz Roedig and Andrew Scott Implementation of a Deterministic Wireless Sensor Network In Proceedings of the 5th IEEE European Workshop on Wireless Sensor Networks EWSN2008 Bologna
91. V Configuration Active Route Timeout 1 58 Hello Message Interval Net Traversal Time Node Traversal Time Table 2 SBR Configuration Hello Message Interval Decrease Routing Interval Short Hello Message Interval Short Hello Message TTL Hello Message Time To Live simulation runs with different seeds of the traffic and mo bility model All error bars show the 99 percent confidence level of the collected statistics whereas histograms represent the average of 20 simulation runs The configurations of the routing protocols are shown in Table 1 and Table 2 The chosen configurations leave enough room for performance improvements of the end to end reliability of the protocols Three different delay metrics are used to simulate the im pact that the deferring of routing messages has on the next hop selection and on the reliability The first metric for wards the routing messages immediately second met ric calculates the forwarding delay according the approach presented in Section 3 and is in the following referred as Relative Speed Estimation RSE metric The forwarding delay of the third metric is chosen in respect to the current absolute node speed 41 Node Speed Distribution The end to end reliability of routing protocols is strongly correlated to the movement pattern of the nodes in the net work For that reason a closer look is taken on the absolute speed of the nodes at the time they forward a packet Fig
92. We need a simulator that copes with simulations of large scale networks with tens to hundreds of thou sands of sensor nodes in an adequate time frame e We need a simulator that realizes realistic and univer sally valid i e hardware independent evaluations e We also need working and easily maintained hardware for our practical evaluations e To narrow the gap between virtuality simulations and reality experimentations on real hardware it is desirable to have the possibility of easy transfers of implemented algorithms from one to the other In the following technologies that address our above speci fied needs are described They will be part of our Evaluation Package Shawn 4 is a discrete event simulator for sensor networks It has been primarily designed for simulating large scale net works with up to a million of nodes with an algorithmic point of view Instead of simulating a phenomenon itself it simulates the effects This approach leads to an essential performance gain Shawn finishes simulations in minutes where say Ns 2 3 is running for more than a day 8 Other design focus points are flexibility and extensibility All crucial parts that can influence the simulation are de signed as exchangeable models First of all is the commu nication model that defines whether two nodes in the net work can in principle communicate with each other Next is the edge model that is responsible for neighborhood rep r
93. ability Fig 2 b signal strengths measured from an outdoor transmission on indoor sensors refer to topology shown in a The order of measured RSS is dominated by physical characteristics of environment and signal propagation effects 50 31 28 55 27 26 25 60 24 23 20 9 70 75 80 85 90 95 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 channel output power MNU NO a Position 1 50 31 30 27 26 25 60 24 23 20 70 75 80 85 90 95 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 channel b Position 2 output power NOOO Figure 3 Results of sampling RSS on all available wireless channels on two different physical positions both receivers are lt 2m from the sender The darker area shows lower signal strength right vertical scale depicts measured dBm values while lack of coloring implies 100 frame loss 66 wireless sensor sensors and an attacker placed outside the room as shown in Figure 2 The wireless senors are based on a MicaZ platform w th CC2420 radios allowing for 32 different transmis sion power settings and 8 bit resolution of received signal strength The power measurements are reported in RSSI Received Signal Strength Indicator yet the conversion to dBm is easily computed by Pygm RSSIvyar RSSlorrser where RSSlorrser 45 2 1 Unpredictability of Signal P
94. ackets in round 1000 This figure depicts the improvements during startup of ALE and especially during the short interruption compared to stable EWMA Figure 2 b shows a link with a PRR of 0 89 which is dropping during round 600 and 900 to a PRR of 0 4 and is interrupted during round 1200 and 1500 Here the improve ments introduced by the dropping mechanism are depicted If the PRR is dropping below PRRoaa the estimation mode of ALE switches back to the agile mode in order to react faster on link changes I 0 9 7 i Why 4 ji 0 8 ry 071 0 6 0 5 041 0 3 4 Agile EWMA 0 1 4 Stable Eo ALE 200 400 600 800 1000 1200 1400 1600 1800 2000 round a Real PRR 0 89 and short message lost PRR S TA 03 N 081 if ji I 0 7 06 ee 04 0 31 PRR 0 1 Stable EWMA ALE 0 200 400 600 800 1000 1200 1400 1600 1800 2000 round b Variation of PRR over time of Agile EWMA Figure 2 Comparison between EWMA and ALE Normally a link is considered as a neighbor if the link quality is above a given threshold here defined by PRRgooa and is often set to 0 8 In contrast a neighbor is disregarded if its quality drops below Doing so oscillation effects can occur Therefore a hysteresis is used to decrease oscillation effects as much as possible Several
95. ag in the EEPROM if a previous write operation failed Additionally all CRCs of the image are verified To this end an image slot inside the external memory is organized in pages of 256 bytes where each page contains a 2 byte CRC A slot also contains an additional 2 byte CRC which is used to verify the CRCs of all pages Thus the integrity of an image can be guaranteed If everything is correct the selected image is written to the program memory The bootloader then sets the status flags and resets command flags in the EEPROM and the program counter is set to the start address of the transferred image The application inside the image is then executed If an error is detected in the image to load the bootloader will not write it to the program memory and sets the respective error flags in the EEPROM Then the control is given back to the main application There is still the possibility that an undetected error occurs while writing to the program memory If this happens and the program counter is reset to the application it will skip these illegal code parts and return to the bootloader There the exceptional invocation is detected since the command flags are not properly set Then the bootloader loads the Golden Image The fourth slot in the external memory is reserved for this The Golden Image contains the main AirFlash functionality and provides a fall back solution if something fails during reprogramming Still it has to be discu
96. ain time it inserts local data instead of strictly following the LFF definition This tries to avoid the worst case by introducing a guaran teed service for data entering the network based on the assumption that bottlenecks only occur very close to the sink where a lot of flows with high priority exist We want to investigate te last point in a little more detail 4 1 Impact of Priority Injection To model the impact of a minimum service guarantee to data entering the network we introduce the notion of a backdoor service curve Intuitively we want to achieve that after an amount of time an original data packet is scheduled even if there are other packets from other higher priority flows scheduled With M as the maximum packet size and as the link capacity this can be modelled as a rate latency curve Bosc B p4M A node 2 following that scheme and with an input a then offers a minimum service curve i eff V bsc to its own traffic while offering Bi ai Bi usc to forwarded flows If we now model the simple network from Figure 2 with Bosc P 2 at each node we get h 1 1 82 3 V G1 2 dg h ma Ba N Bg 2 2 4 which means an improvement for the whole network since it allows node 1 to forward its own data even though it is experiencing heavy load from nodes 2 and 3 At the same time overcompensation is avoided Further study of priority injection is the subject of futu
97. al cost of 6 or Na for both 51 and S2 with a cost of 7 Among the possible combinations only those K combina tions that provide a lower number of hops are stored The higher K is the more different possibilities a node has to for ward a packet however a larger computation delay is also expected If the delivery delay must be minimized this pa rameter must have a large weight Bach combination has associated a number of neighbors N in Equation 1 The smaller the weight of neighbors is the larger the probability that the paths remain unsplit keeping the total energy consumption low The energy of the nodes is independent from the rest of the parameters as it measures the number of messages a node has sent or forwarded since it is active A large weight for the energy means that load in the network is more balanced The three remaining parameters are closely related to each other The number of sources a node has served in the last epoch is related to the high load indicator The maximum num ber of served sources is not known before hand When this number is reaching its maximum the number of contentions increases and therefore the load indicator is active As long as the number of served sources is far from its maximum load indicator is not active the cost due to the number of messages sent only depends on the number of sources and SOUT CES paths energy Ww N 4 N Ws we highload 1 w3 Ho
98. and disadvan tages of localized over centralized data communication pro tocols with real world measurements We focused on one representative of localized and one representative of global routing so far The space of possible future investigations is endless and we anticipate a discussion on the most interest ing ones to follow We think that a final justification when localized is bet ter than a globalized approach requires many independent measurements coming from independent research groups Even measurements repeating experimental setups and ex periments of certain protocols are of importance here We advertise that a single result of one paper should not ex clude other papers performing the same measurements and probably achieving varying results This is however often not that accepted in network research like it is accepted in other disciplines like physics for instance 6 REFERENCES 1 P Buonadonna J Hill and D Culler Active message communication for tiny networked sensors In Proceedings of the Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies INFOCOM 01 Apr 2001 2 G Finn Routing and addressing problems in large metropolitan scale internetworks Technical Report ISI RR 87 180 Information Sciences Institute ISI Mar 1987 3 H Frey and I Stojmenovic Geographic and energy aware routing in sensor networks In I Stojmenovic editor Handobook on Sensor Networks
99. and if v v v 2 09 2 o e of oF Q 00 lt 25 bytes 00 Low 0 Allocation 00 1 period 01 lt 50 bytes 01 Medium 0 RX 01 2 periods 10 lt 75 bytes 10 Critical 1 Deallocation 10 3 periods 11 lt 118 bytes 11 Most critical 11 4 periods Figure 3 Revised GTS characteristics field where first two bits specify the length of a data packet and the next two bits define the delay condition of the packet Bits four and five taken from the original standard define the GTS direction and GTS request type The last two bits are used for periodic appli cations it finds its GTS entry it can utilize it by sending receiving data After using its GTS slots the device makes its ra dio off again The device can send new GTS request via its CAP to the coordinator and if the request is accepted its GTS entry will be added to the upcoming beacon The node requests for GTS slot by specifying its data delay and en ergy conditions and the coordinator who runs an admission control AC decides the fate of the request In order to cope with the aforementioned limitations of the protocol we change the bit structure of 8 bit GTS charac teristic field where the device rather than sending fixed slot length sends its data and delay specification to the coordi nator as shown in Figure 3 The coordinator who runs an AC decides the slot length for the device AC works on th
100. anged with any alteration of sender s con figuration of antenna orientation physical position or even a trans mission frequency To demonstrate the impact of frequency changes on RSS we sampled transmissions on every wireless channel with out changing the positions of the sensors The results measured at two different receivers are depicted in Figure 3 a and b So for example if the transmission is changed from channel 13 to chan nel 19 i e 2405 MHz to 2445 MHz one receiver experiences RSS decrease of 20dBm Figure 3 a Yet if the transmission is changed from channel 13 to channel 22 another receiver mea channel 11 12 13 4 15 16 17 18 19 20 2 2 23 2 25 26 11 2 13 14 15 6 17 18 19 20 23 2 23 2 25 26 11 12 13 14 15 6 17 18 19 20 21 2 23 24 25 26 Receiver Verifier 3 Sensor 4 Verifier 2 Verifier 1 Rejected Sequence number Figure 4 Time line of legitimate transmissions verified by 3 sensors and 1 receiver The figure shows acceptance intervals at each node gray area and received frames black dots as accepted frames crosses as rejected Using the same PRN
101. are hardly or not accessible at all The conventional method to update sensor no des on demand by means of mobile computers and debug interfaces is safe and secure indeed but in any case extremely complex an noying or even impossible To counter this problem the Ghost sub system offers an efficient method for accomplishing software and configuration modifications via image distribution over the com munication infrastructure which is available anyway within such networks Since data dissemination protocols have received wide attention within the WSN community several similar approaches like Typhoon 7 FiGaRo 8 and Deluge 6 already exist Yet such systems commonly provide their own customized communi cation routing protocol or even require a special runtime environ ment on the nodes In our opinion both premises are adverse since remote maintenance is important indeed but should neither defi ne the actual application s communication nor affect the overall Gateway Workstation Node Figure 1 Ghost infrastructure for wireless remote maintenance system performance This is why Ghost was designed to use any available protocol and needs no modifications or extension of the used operating system Instead our focus aims on resource aware integration into any software besides providing efficient and relia ble operation In this paper we present the central concepts of the Ghost remote maintenance subsystem but refer to radio communi
102. are operation in general and ener gy harvesting in particular Table 1 Energy requirements during sensor node operation low system load no radio heavy system load no radio data reception radio and buffering flash haunting process copy flash to ROM 6 REFERENCES 1 M Baunach Speed Reliability and Energy Efficiency of HashSlot Communication in WSN Based Localization Systems In Verdone 11 pages 74 89 2 Baunach Kolla and C M hlberger SNoW A versatile ultra low power modular node for wireless ad hoc sensor networking In J Marron editor 5 G ITG KuVS Fachgespr ch Drahtlose Sensornetze pages 55 59 Stuttgart 17 18 July 2006 Institut f r Parallele und Verteilte Systeme 3 M Baunach R Kolla and C M hlberger Introduction to a Small Modular Adept Real Time Operating System In Distributed Systems Group editor 6 Fachgespr ch Sensornetzwerke pages 1 4 Aachen 16 17 July 2007 RWTH Aachen University 4 M Baunach R Kolla and C M hlberger SNoW Bat A high precise WSN based location system Technical Report 424 Institut f r Informatik Universit t W rzburg May 2007 5 Baunach Kolla and C M hlberger SNoW modular platform for sophisticated real time wireless sensor networking Technical Report 399 Institut f r Informatik Universit t W rzburg Jan 2007 6 J W Hui and E Culler The dynamic behavior of a data dissemination prot
103. as the flow of interest To that end we define the function h f as the number of hops passed by flow f at node n Define the set Fp p f Fass half gt hn f The effective service curve offered to a flow of interest then becomes Bn eff Bn gt On f fer x The expected results of that scheme are improved delay bounds for longer flows while the bounds for the shortest flows show no significant change This can be intuitively explained by LFF only removing potential interference from the longer flows while still considering all cross traffic for the shortest ones No flow exeriences a penalty For illustration we look at a small network which is shown in Figure 2 and calculate the delay bounds for the flows originating at the nodes 1 2 and 3 which will be called di with i 1 2 3 Each node generates input limited by token bucket arrival curve 71 1 and exhibits a for warding behavior lower bounded by a rate latency minimum service curve 241 The PMOO analysis under arbitrary multiplexing yields di h a 61 a2 b2 as Bs d2 h a2 82 81 Bs da 8 9 8 a2 b2 Under LFF flows 2 and 3 get priority over flow 1 and the term for da and d3 becomes da d3 h a2 82 Q G1 az B3 4 h as Q G1 32 4 We see an improvement for the longer flows There is no im
104. ash partitions the external memory into three so called slots of 120kB for firmware images and a fourth slot of 20kB for the Golden Image The remaining 132kB can be used by data logging applications For programming of the node the open source framework TinyOS 2 x is used 6 The choice for TinyOS was made since it provides powerful high level interfaces to access the node s hardware 4 AIRFLASH IMPLEMENTATION At the node s location there are three basic tasks which must be performed by AirFlash Receive image via wireless transfer Store image in external slot Load image from slot to program memory For the last task the microcontroller s internal and external memory must be accessed For this purpose usually a so called bootloader application is used This bootloader resides in a dedicated partition 8kB of the program memory of the ATmegal281 and is able to access the remaining 120kB of memory and execute applications therein The bootloader is also able to access the external memory see Figure 2 Since the partition dedicated for the bootloader is too small to contain the complete AirFlash functionality task one and two of the above list must be performed by an additional software module Such an AirFlash module that controls the wireless image transfer must be included into the node s main application There it stays in the background and is only activated when awoken by a wireless command If the AirFlash module
105. at it is possible to modify the topology of a network by deferring the forwarding of routing information None 07 Speed Estimation MB Absolute Speed 021 0 1 u lih be ke iu 1 3 9 11 143 15 AF 18 Node Speed in mis Figure 2 SBR Node Speed Histogram 80 None f T Rel Speed Estimation Absolute Speed oo 065 BR a ital E 5 08 0 55 2 4 6 10 12 14 16 18 20 22 Maximum Speed in m s Figure 3 AODV End To End Reliability 6 REFERENCES 1 C E Perkins E M Belding Royer and S R Das Ad Hoc On Demand Distance Vector AODV Routing RFC 3561 IETF MANET Working Group 2003 2 A Klein and Tran Gia A Statistic Based Approach towards Routing in Mesh Networks IEEE MASS pp 1 6 Pisa Italy October 2007 3 J M Francois G Leduc and S Martin Learning Movement Patterns in Mobile Networks a Generic Approach in Proceedings of European Wireless 2004 pp 128 134 Barcelona Spain February 2004 4 W Su S J Lee M Gerla Mobility Prediction and Routing Ad Hoc Wireless Network Int Journal of Network Management Vol 11 No 1 pp 3 30 January 2000 5 Y B Ko and N H Vaidya Location Aided Routing LAR in Mobile Ad Hoc Networks in Proceedings of ACM IEEE MOBICOM 98 pp 66 75 USA 1998 6 N Frangiadakis et al Realistic Mobility Pattern Generator Design and Application in P
106. ate that any link no matter of what long term quality can temporarily change its characteristics and thereby temporarily become a reliable link for routing or become an unreliable one We consider an intermediate or broken link temporarily available when it successfully transmits a number of pack 28 ets over a short interval We define a corresponding thresh old based on the link s long term reliability for example a link of intermediate quality needs to transmit less pack ets before being considered temporarily available than a link of bad quality Similarly a number of successfully transmit ted packets indicate that a reliable link 15 currently available while failed ones indicate that a link is currently not avail able Overall we expect that a single successful transmission indicates that a long term reliable link is currently available Furthermore short term link estimator does not send probe packets to test for link availability it bases on packet overhearing Hence a node overhears packets sent by neigh boring nodes and collects statistics on the current reachabil ity When a node considers the incoming direction of an un reliable link temporarily stable and concludes that it offers a routing improvement it sends a message to the link neighbor to inform it about a short term link availability The neigh boring node may then consider routing subsequent packets over the newly available link If the node is selec
107. ath Prediction Algoritm Evaluation 13th IEEE Int Symp on PIMRC Vol 2 pp 765 769 Athen Greece September 2002 7 C K Toh Long Lived Ad Hoc Routing Based on the Concept of Associativity IETF Draft 1999 8 M Gerharz C de Waal M Frank and P Martini Link Stability in Mobile Wireless Ad Hoc Networks in Proceedings of the 27th Annual IEEE Conference on Local Computer Networks LCN pp 30 39 2002 9 J Yoon M Liu and B Noble Random Waypoint Considered Harmful in Proceedings of INFOCOM 2003 San Francisco USA April 2003 ne Speed Estimation Absolute Speed gt 09 5 0 85 5 gt 0 75 2 Figure 4 SBR End To End Reliability 4 6 8 10 12 14 16 18 20 22 Maximum Speed in m s Ghost Software and Configuration Distribution for Wireless Sensor Actor Networks Marcel Baunach Department of Computer Engineering Am Hubland University of W rzburg baunach informatik uni wuerzburg de ABSTRACT Wireless sensor and actor networks commonly consist of a rat KA oy her large number of more or less widely distributed nodes The gt ae resulting spatial complexity arises severe software related main tenance problems like application code and configuration updates This paper presents the Ghost subsystem for efficient remote soft ware maintenance in such networks Besides safety security and operation in heterogeneous
108. ation and experiment specific settings both for sensor and mesh network in an uniform structure 2 2 Wireless Mesh Router The mesh router component of the hybrid nodes is based on an Alix2c2 mainboard featuring a 500 MHz AMD Geode LX800 CPU and 256 MB RAM A customized Debian Linux is used as operation system A minimum of three IEEE 802 11 based NICs are attached to a mesh router over a powered USB hub Due to experimental results and driver availabil ity 7 for Linux we chose Logilink WL0025 adapters based on the RT2501U 2 chipset 2 3 Wireless Sensor Network Node Our current generation of sensor nodes is called MSB A2 and was developed at the Freie Universit t Berlin as part of the ScatterWeb 5 working group The sensor nodes are connected to each hybrid node via USB The ARM7 based devices can be configured dynamically at runtime up to 60 MHz depending on the sensor network application and 50 energy requirements The Chipcon CC1100 transceiver in our configuration uses the ISM band at 868 to 870 MHz with a maximum data rate of 500 kbps The WSN nodes can be equipped with a multitude of sensors using either simple general purpose input output pins GPIO or an on board mini USB port Additional peripheral devices can also be connected to the mesh routers We aspire full integration of WPANs TETRA and cellular networks GSM Our current research projects have taught us that convergence of WSNs and oth
109. ation development process 2 RELATED WORK To the best of our knowledge currently there are no frame works for automatic bug detection in sensor network appli cations before deployment at bit accurate semantics For bringing memory safe executions of applications at runtime we only know of the related efforts 6 10 where the sole representative with fined grained memory safety at level is Safe TinyOS Safe TinyOS adds dynamic memory checks during compila tion which allows to catch unsafe pointer and array opera tions without corrupting the RAM Overall this results in Features FOS Automatic code instrumentation Target platform independence Assembly level bug detection Off line bug detection All possible input values checked Automatic test case generation Runtime safety enforcement No additional resource usage Table 1 Comparison Safe TinyOS and KleeNet 13 increase in the code size and 5 2 increase in CPU us age As with any dynamic assertion checking Safe TinyOS can detect program bugs only eventually after the software is deployed Therefore still many corner case bugs circumvent this testing technique KleeNet on the other hand offers offline bug detection with automatic code instrumentation In doing so it doesn t consume any system resources and ensures memory safety by treating the inputs symbolically i e checks any program variable for its all possible input values A
110. ation is usually only achiev able by configuring all nodes of the network manually and defining their states one after another With our approach we will be able to record a successful experiment run and make use of the gathered information for future application scenarios To provide such a feature our system has to be able to save each node s state at arbitrary times That includes global variables stack content CPU and transceiver status One of the arising challenges is to do this operations in the short est time possible The sensor nodes have to continue to be an active part of the network with minor downtime In con trast the restoration of the state is not time critical but requires a network wide switch to signal all nodes to con tinue their tasks NCPs rely on a shared timebase for all sensor nodes which can be provided by the mesh network We are still undecided whether such a feature will interfere with the WSN and introduce short but unwarranted down times of nodes An initial evaluation will examine if peri odical NCPs in discrete time intervals are feasible or just infrequent ones are tolerable 3 3 Interfacing the Sensor Nodes One of our more ambitious concepts focuses on the integra tion of the used radio transceiver into the Linux kernel as a driver module The sensor node shall be provided as an addi tional network interface card In the context of the ISO OSI reference model the sensor node repre
111. auter kabel gebundener Sensorsysteme realisiert Hier bieten sich draht lose Sensornetze als kosteng nstige und flexible Alternative 53 an In 6 wird ein berblick ber die zahlreichen protoypi schen Systeme dieser Art gegeben Messungen werden hier bei in den meisten F llen mit Hilfe von Beschleunigungs bzw Ersch tterungssensoren oder Dehnungsmessstreifen er hoben Im ersten Fall m ssen die Messungen zumindest zeitweise mit einer relativ hohen Samplingrate durchge f hrt werden Viele Systeme k nnen Messungen direkt auf den Knoten verarbeiten dazu sind z B FFT oder Wavelet Transformationen notwendig was Energievorteile gegen ber einer direkten bertragung der Me werte an eine Basissta tion hat H ufig wird auch eine zweischichtige Gesamtarchi tektur der Systeme vorgeschlagen Auf der unteren Schicht befinden sich die eigentlichen Sensorknoten die die Me werte erheben Einer Gruppe solcher Knoten wird jeweils ein Zentralknoten auf der oberen Schicht zugeordnet Dieser sammelt und verarbeitet die Me werte der Sensorknoten Die Zentralknoten verf gen ber leistungsf higere Hardwa re und k nnen auch untereinander ber gr ere Distanzen kommunizieren 2 2 Patienten berwachung Die Verf gbarkeit von konsteng nstigen und nicht invasiven medizinischen Sensoren macht den Einsatz von Sensornet zen zur Patientenkontrolle in Kliniken interessant Solche Systeme lassen sich dazu nutzen Patie
112. ave been developed ranging from static code analysis formal verification to full state model checking Numerous tools exist and are freely available to use 4 5 7 15 Hence our first step was to employ them for testing sensor network software written in widely spread TinyOS platform 12 We encountered the following prob lems due to which the available tools are mostly not used at all and the developers fall back on manual code debugging techniques e Sensor network applications are tightly integrated with the whole operating system leading to time consuming manual code modification in order to perform the ac tual testing e Most of the tools perform only static code analysis with limited support for C semantics Since sensor network applications mainly process data from the en vironment possible runtime errors might stay unde tected e None of the tested tools can offer a push button bug finding technology and the usage learning curve is mostly too steep for a typical developer We have discovered that recent research efforts in the area of C code checking are now targeting the usability and full automation as primary design goals 1 2 The philosophy is to detect only definite errors but automatically at full ex ecution path coverage and with minimum manual effort by employing symbolic checking techniques 9 We think that these efforts could finally achieve the integration of sound testing tools into the applic
113. avoidance mechanism is presented to regain service attention for lower priority devices that need more GTSs for data transmissions A simulation model and an analytical model are developed to investigate the performance of AGA scheme The work in 5 suggests a multi beacon superframe MBS structure with multiple sub beacon intervals for different slot sizes in a superframe and a greedy GTS allocation GGA algo rithm where device determines the most appropriate slot sizes based on their traffic characteristics They claim to have significant improve in the bandwidth utilization at the expense of only a very small increase of the device s active periods A distance based real time off line periodic mes sage scheduling algorithm is proposed in 9 which generates BO SO and GTS information to schedule the given message set 4 SYSTEM DESCRIPTION We consider an IEEE 802 15 4 based network which works in star topology with beacon enabled mode and uses 2 4 GHz frequency band with O QPSK modulation 8 The coordi nator collects data from its CFP regarding the devices that need GTS slots assembles the beacon and transmits it Af ter that CAP starts which follows the optional CFP On the other hand the device first synchronizes itself with the coor dinator makes its receiver on just before the beacon arrival time and receives the beacon That follows CAP and then CFP where the device first checks for its GTS slots
114. ay Infrastructure components are likely to be placed in a way that optimizes the coverage of the industrial wireless sensor network Figure 4 shows a typical industrial sensor network setting several sensor nodes communicate with one of multiple gateways GWs The gateways are connected to the wired or wireless factory intranet These gateways are placed in a way that minimizes the path length in the industrial sensor network The use of infrastructure is also advantageous because plant op erators usually require access to devices for information and de bugging purposes 3 3 Operation In academic research projects it is often assumed that a wireless sensor network is deployed in a public space and once deployed works without administration and unsupervised In industrial applications hoverer and in particular in process and factory automation wireless sensor networks are usually deployed in access restricted areas a Factory Intranet gt bee er Figure 4 Typical industrial wireless sensor network setting with two gateways Plant premises are rarely accessible without physical access con trol This affects the security design of industrial wireless sensor network protocols as a corruption of a sensor node due to physical access is less likely To ensure an uninterrupted uptime plant an industrial sensor network will typically not be unsupervised but rather be watched carefully 24 7 A typical sensor
115. bitrarily aligned anchors is another interesting issue Hence we want to proof some upper or lower bounds for the number of re quired reference points within such systems 7 REFERENCES 1 P Bahl and V N Padmanabhan RADAR An In Building RF Based User Location and Tracking System In INFOCOM 2 pages 775 784 2000 2 Baunach Speed Reliability and Energy Efficiency of HashSlot Communication in WSN Based Localization Systems In R Verdone editor EWSN volume 4913 of Lecture Notes in Computer Science pages 74 89 Springer 2008 3 M Baunach R Kolla and C M hlberger A Method for Self Organizing Communication in WSN Based Localization Systems HashSlot LCN pages 825 832 IEEE Computer Society 2007 4 M Baunach Kolla and M hlberger SNoW Bat A high precise WSN based location system Technical Report 424 Institut fiir Informatik Universitat Wiirzburg May 2007 5 N Bulusu J Heidemann and D Estrin GPS less low cost outdoor localization for very small devices 44 10 11 12 Personal Communications IEEE see also IEEE Wireless Communications 7 5 28 34 2000 Y Fukuju M Minami H Morikawa and T Aoyama DOLPHIN An Autonomous Indoor Positioning System in Ubiquitous Computing Environment In WSTFES 03 Proceedings of the IEEE Workshop on Software Technologies for Future Embedded Systems page 53 IEEE Computer Society 2003 A Giinther and C Hoene
116. bors on the grid nodes was set to 25cm To have a reasonable multi hop experiential scenario transmission power of each node was set to 2 out of a range between 0 and 31 We investigated that a trans mission power 1 rendered many message losses due to weak signal strength A signal strength above 2 kept too many nodes connected thus rendering the topology almost fully connected To control the experiment the node at the top left and the node at the bottom right of the 7 x 7 grid have been attached via USB to a notebook computer We implemented a Java application to generate the traffic and collect measurement data from the final destination node The Java application tells the attached source node at the top left corner to send a data packet periodically to the destination node at the bottom right corner The time interval between two message transmissions was set to 2 seconds 4 RESULTS The graphs depicted in Fig 1 compare the performance of DSR and greedy routing in terms of message delay success rate and hop count measurements are plotted against different invalidation times for DSR route caches and greedy neighborhood tables Each point in the graph is the average of 1000 measurements Figure la shows the average delay from source to destination we observed for varying cache invalidation times Clearly a longer invalidation delta supports a lower delay Routes can be used for a longer time before route dis
117. bride Algorithmus muss flexibel genug sein um alle m glichen Knotenausf lle zu kompensieren und damit sowohl eine dauerhafte verteilte als auch eine zentralisierte Berechnung zuzulassen was zu einer l ngeren Lebensdauer des Netzwerkes f hrt 3 1 2 Lokalisierungsmethoden In der Literatur wird zwischen zwei M glichkeiten unterschieden um eine Lokalisierung in Sensornetzwerken zu erm glichen Born 86 et al 2008 Die approximativen Algorithmen nutzen entweder N herungsbeziehungen oder nur sehr ungenaue Positionen bzw Distanzen als Eingangsdaten und bestimmen eine eher grobe Position mit wenig Rechenaufwand Dazu nutzen sie zum Beispiel Schwerpunktbildung WCL Blumenthal et al 2005 oder Fl chen berlagerung APIT He et al 2003 Sie k nnen Fehler einfacher verzeihen und sind damit 1 A sehr robust Allerdings ist der Algorithmusfehler selbst daf r verantwortlich dass sie auch bei fehlerfreien Eingangswerten nie eine exakte Position ermitteln Dies ist wiederum ein Vorteil der zweiten Klasse den exakten Algorithmen Sie k nnen bei entsprechend genauen Eingangswerten auch exakte Ausgangswerte produzieren Born et al 2006 Jedoch ben tigen sie daf r 1 A viel Rechenzeit und Speicher wodurch eine uneingeschr nkte Ausf hrung auf ressourcenlimitierten Sensorknoten nicht m glich ist Dennoch gibt es erste Ans tze die durch z B eine sinnvolle Verteilung der Subaufgaben Distributed Least Squares Algorithmus
118. bsolute speed In the following a look at the structure of an urban net work is taken On one hand there are hotspots in urban areas e g pedestrian areas or areas around places of inter est where the user density is very high On the other hand there are low density areas in the outlying districts of a city The nodes in such a scenario can be divided into two ma jor groups The first group consists of slow moving or fixes nodes that remain within the area close to a hotspot The other group is represented by nodes which move more or less directly and fast from one hotspot to another If two users in one hotspot decide to move on to the next hotspot they will be able to directly communicate with each other during the travel due to their correlated movement This paper is organized as follows Related work is dis cussed in Section 2 The approach that takes advantage from correlated node movement is introduced in Section 3 Furthermore the section includes a description how to im plement the method in AODV and SBR Simulation results are presented in Section 4 Finally we summarize the results and conclude with our fields of research in the future 2 RELATED WORK Routing protocols that are designed for mobile wireless networks have to deal with frequent topology changes It is clear that the number of topology changes corresponds to the transmission range and the mobility pattern of the nodes A shorter transmission range de
119. cation 4 5 are not appropriate in case data must reach several sinks Most commonly they simply duplicate the same solution Recently some new protocols deal with the problem of routing data to several sinks in WSN Never theless they usually create a tree with the source at the top and the sinks as leaves However these trees are created in dependently from each other Hence it is possible that some nodes are included in many paths between sources and sinks As a result such nodes spend too much energy forwarding packets and their batteries get exhausted quickly Whereas the total amount of energy spent by the WSN can be opti mal if each of the trees is also optimal this expense can be concentrated among a small number of nodes This means that such nodes deplete their batteries very fast while other nodes remain almost unused Finally this can lead to the partition of the network avoiding the transmission of data Our goal in this paper is to enable the adaptation of rout ing paths for many to many communication appropriate for these new multi sink scenarios In order to select the next node s towards the sink a cost function is evaluated This cost function maintains a trade off between low delay and low power consumption while keeping the load of the net work evenly balanced The function takes into account the number of hops to the sinks as well as the load produced by all the sources in the network It balanc
120. ce calculates the new and or modified branching nodes Only these nodes are notified about the changes in the tree branching nodes that do not need to be changed require no notification Removing re ceivers from the tree is done similarly branching nodes that need to be modified or removed are notified accordingly For the centralized receiver driven approach the join mes sages from the receivers are forwarded to the source which manages the tree as described for the source driven central ized approach In the receiver driven de centralized approach receivers send the join message to their neighbor responsible for the default route If this node is not a forwarder or branch ing node for that group it becomes a forwarding node only knowing that it is on the path of an overlay link when the tree would become active and forwards the join message further Intermediate nodes which are already branching nodes of the requested group drop the join message and pre pare the overlay link to the new receiver Forwarding nodes receiving join messages become new branching nodes pre pare the new overlay link and send this information about becoming a branching node towards the source dropping the original join message A branching node receiving such a message modifies its overlay link in that direction There fore the overlay link of which the new branching node has been acting just a forwarder before is removed and replaced by an overlay li
121. ck and Vibration Digest 38 2 91 128 2006 7 D Malan Fulford Jones Welsh and S Moulton CodeBlue An ad hoc sensor network infrastructure for emergency medical care In MobiSys 2004 Workshop on Applications of Mobile Embedded Systems 2004 8 A Perrig Szewczyk V Wen D Culler and J Tygar Spins Security protocols for sensor networks In Proceedings of the Seventh Annual International Conference on Mobile Computing and Networks MobiCom 2001 pages 189 199 2001 9 R mer and Mattern The design space of wireless sensor networks JEEE Wireless Communication 11 6 54 61 2004 10 Y Cheng and D Agrawal An improved key distribution mechanism for large scale hierarchical wireless sensor networks Ad Hoc Networks 5 35 48 2007 Real time Bandwidth and Energy Efficient IEEE 802 15 4 for Medical Applications Pardeep Kumar Mesut Gunes Abd Al Basset Al Mamou Jochen Schiller Institute of Computer Science Freie Universitat Berlin Berlin Germany opkumar guenes almamou schiller inf fu berlin de ABSTRACT The recently standardized IEEE 802 15 4 protocol provides many appealing features such as low power low cost low data rate simplicity and timeliness guarantee for delay bound energy and bandwidth critical applications How ever when treating applications with such requirements as in the medical field several problems arise triggering the need of an efficient en
122. covery is initi ated by DSR or neighbor rediscovery is initiated by Greedy respectively However compared to Greedy DSR is more susceptible for short route invalidation times Delay is in creasing significantly in this case Those large delays come from large setup times until the DSR route is established We were interested as well how long it takes until a DSR route is established Figure 1b depicts the average setup time over the route invalidation time If invalidation time decreases for invalidation time 0 route discovery starts for each new message the setup time increases significantly We interpret this result by the fact that route discovery will overlap with previous ones thus it may take a longer time until a message is handled by a node We were also interested in the performance of DSR com pared to Greedy after the route of DSR is established As depicted in Figure 1 once the route is established DSR performs with a lower delay compared to Greedy Figure 14 shows the success rate of Greedy and DSR over the invalidation timeouts We observed a low and almost in variant success rate of about 20 for Greedy while the suc cess rate increases for DSR over invalidation timeout length Route setup in DSR favors good path over bad ones auto matically In Greedy we select the neighbor node closest to the destination We believe that Greedy can be improved significantly at this place by selecting a different m
123. creases the average number of neighbors of a node In addition the neighbor hood of a node changes more often since nodes that pass by remain a shorter amount of time within the coverage area of the node However usually it is not possible to modify the transmission range or the movement of a node The rout ing protocols can follow different strategies to deal with the mobility problem One strategy which is used by a large number of protocols is to predict the movement and the connectivity of the nodes in advance to minimize the number of link breaks and the amount of routing overhead The prediction can be based on several characteristics e g bit error rate packet loss re ceived signal strength location information obtained from the Global Positioning System GPS or learned movement patterns as shown by 3 The Flow Oriented Routing Pro tocol FORP 4 uses the GPS information to predict the mobility in advance Therefore it is able to compute the link expiration time for each hop along the route The pre dicted link expiration time is used to calculate the route expiration time which can be regarded as a metric to evalu ate an end to end route in respect to stability and reliability Other routing protocols like Location Aided Routing LAR 5 utilize the GPS information to reduce flooding overhead Learning algorithms can be used to predict the movement pattern of the nodes in the case that they have enough com putational powe
124. d dynamic source routing and greedy rout ing in nesC 4 under the TinyOS 2 0 execution environment 6 Wireless communication between immediate neighbor ing sensor nodes and wired serial communication between sensor node and host computer is implemented on the ba sis of Active Messaging 1 The maximum allowed message size was fixed to 64 bytes payload The following provides a brief outline on both implemented protocols and motivates the rationale behind our implementation decisions 2 1 Dynamic Source Routing 2 1 1 The Algorithm Principle Dynamic source routing 7 DSR in short basically works as follows The source node sends out a route request message RREQ which is repeated by all intermediate nodes once they receive it for the first time Additional RREQ receptions are just ignored Every RREQ message copy stores the sequence of nodes it has visited so far In this way several copies of the RREQ message will eventually reach the destination node which can then pick up one of these messages to send a route reply RREP back along the reverse of the path stored in the RREQ messages To achieve this the RREP message just stores the sequence of nodes which have to be visited from source to destination Once the message arrives at the source 73 node the reverse of the path stored in the RREQ message can finally be used to send data towards the destination node 2 1 2 Implementation Route setup in DSR requires some time
125. de conditions ACKNOWLEDGEMENT This work has been performed as part of the ZESAN project partly funded by the Federal Ministry of Education and Research under the funding number 01BN0712A The paper represents the opinion of the authors The authors would like to acknowledge the contributions of their colleagues participating in the ZESAN pro ject 95 Vergleichbarkeit von Ans tzen zur Netzwerkanalyse in drahtlosen Sensornetzen Joachim Wilke Institut f r Telematik Universit t Karlsruhe TH wilke ira uka de Zinaida Benenson Institut f r Informatik Universit t Mannheim zina uni mannheim de Frank Werner Institut f r Theoretische Informatik Universit t Karlsruhe TH werner ira uka de Simon Kellner Institut f r Betriebs und Dialogsysteme Universit t Karlsruhe TH Markus Bestehorn Institut f r Programmstrukturen und Datenorganisation Universit t Karlsruhe TH bestehorn ira uka de Erik Oliver Bla Networking amp Security EURECOM France blass eurecom fr kellner ira uka de Kurzfassung In diesem Dokument wird die Vergleichbarkeit verschie dener Netzwerk Simulatoren Emulatoren und theore tischer Analyse in drahtlosen Sensornetzen untersucht Die Ergebnisse dieses im Rahmen des ZeuS Forschungs projektes durchgef hrten Experiments zeigen die St r ken und Schw chen der verschiedenen Ans tze auf 1 EINLEITUNG Im Rahmen des ZeuS Forschungsverbundes einem Pro jekt zur Erforschung v
126. deler 14 5 is used to simulate the impact of the cross layer approach on the end to end reliability and the selection of forwarding nodes of AODV and SBR in re spect to their current relative speed Instead of using the OPNET AODV model the model is implemented as spec ified in the RFC The physical layer is replaced by a disc model to minimize side effects Furthermore the radio range of the nodes is set to 200 meters and the transmit data rate is set to 256 kb s The nodes use Carrier Sense Multiple Ac cess CSMA as Media Access Control MAC protocol with a simple back off algorithm At the beginning of each simu lation 100 mobile nodes are randomly placed on a square of 1000 by 1000 meters All nodes move according to a random waypoint model with a constant pause time of zero seconds and a minimum node speed of 1 m s The maximum node speed is increased from 2 m s to 20 m s in steps of 2 m s 10 nodes in the network select a uniform destination at the beginning of the simulation These nodes generate packets according to an exponential distribution with a mean value of 2 seconds and a constant packet size of 1024 bits The traffic model is started after 300 seconds to minimize the impact of the tran sient phase of the random waypoint model The duration of the simulations is set to 1400 seconds Statistics are collected after 400 seconds to allow the stabi lization of the network The results are calculated from 20 Table 1 AOD
127. dency needed by other device to become active power down only applicable if a hub is used Figure 3 Virtual device State Machine diagram 17 pend on the same device entities Virtual devices may be used without hub Interrupt handlers shall use hubs to forward any event or data to the associated handler ofthe currently active device on the hub 2 2 4 Power Management We propose a fine grained multi layer power manage ment Each layer uses the same power down API such that the virtual device state compare Figure 3 remains consistent at all times Device Implementation Devices can deactivate them selves immediately after an operation transparent to the HAL core This is especially reasonable for devic es that consume a lot of power and complete in a single operation Examples are AD converters or some temperature sensors Virtual Device Hub The HAL powers down vir tual device if another device becomes activate on the same hub or a device has to be deactivated due to conflicting dependencies F Operating System The system kernel powers down inactive devices before the microcontroller is set to low power mode Virtual devices that are selected as default on a hub will never be inactive and thus are not powered down by the ker nel For kernel power management the HAL needs to keep track of inactive devices Depending on the microcontroller architecture it can be effective to disable all unused ci
128. density will be chosen to guarantee a connected network even if some nodes experience permanent failure The well known IRIS platform is utilized and TinyOS 2 x 1 forms the basis for the software development This demonstration system gives the opportunity to gather long term measurements of the behavior of WSNs It will fulfill two major tasks Collecting status information of the network itself such as neighborhood link quality successful packet transmission rates and power supply monitoring Running for several months a good estimation on stability and reliability of such a WSN can be achieved Besides status monitoring convenient applications will be demonstrated All nodes are regularly measuring the tem perature which is graphically displayed The network will also be able to forward position information of mobile nodes that are equipped with GPS receivers The architecture is based on a routing tree with one distinct data sink All acquired data is stored inside a database and visualized using Google Maps In a first version most nodes are battery equipped and a few nodes are powered by an autonomous supply unit developed at the TUHH This unit uses a solar cell that is dimensioned to guarantee a stable power supply of the backbone routing network even under unfavorable weather conditions 3 1 Applications Hardware and software development for WSNs will be vali dated by performing long term tests on the campus network
129. dure dur ing the localization stage no replies from beacons to clients are required anymore This way the hardware for clients and beacons remains the same for deployment and local ization Still this deployment procedure implies two open issues which need a closer examination 1 First the reference client needs exact knowledge about its own position now For example this can be done manually but an inaccuracy of a couple of millimeters already produces drastic errors during localization 2 Next the number of required known positions of the reference client during deployment should be kept as small as possible to save energy time and costs mainly manpower Because the first item depends on the craftsmanship of the calibrating persons as well as the accuracy of the used mea suring instruments we will initially focus on the second item in the next section 5 CALIBRATION In this section we address the localization system calibra tion after installation of the static beacons at the anchor plane The goal is to determine accurate anchor positions while achieving a short duration for this exceptionally im portant process As already described in section 4 we use certain carefully selected reference points at well known po sitions to allow each anchor to be to measure at least four distances in a three dimensional space Of course increas ing this number yields higher precision but also consumes more time and power T
130. e Currently sensor network application developers are con fronted with a number of domain specific complications The constrained memory and CPU resources on sensor nodes re sult in using low level type unsafe languages without dy namic type checking and memory protection Similarly be cause the applications are highly data flow oriented the cor rect exception handling at full coverage is a challenging task Moreover due to highly distributed and faulty nature of sen sor nodes some of the program bugs are detected only after the software is deployed C language has been the main choice for developing sensor network applications It provides great flexibility expres siveness and in particular the required low resource foot print However the absence of dynamic type checking neces sitates very careful programming because many sensor OS s do not support memory protection We argue that besides the particular sensor application semantic the majority of bugs comes from general programming flaws such as mem ory out of bounds references null pointer dereferences or wrong type conversions Especially the widely used casting between pointers to structures may lead to well hidden type errors But in most cases the given language features are simply misused by novice programmers Starting with lint 8 tool nearly three decades ago there has been enormous effort spent on the error removal in C pro grams Many techniques h
131. e wurde im Jahr 1998 ma geblich gepr gt und malt eine dreidimensionale Repr sentation der Erdoberfl che verschiedenster Aufl sungen zur Unterbringung ausgedehnter Mengen von georeferenzierten Daten aus OGC 1998 Grundbestandteil dieser Digitalisierung ist eine zum Teil bereits heute gegebene Allgegenw rtigkeit von untereinander vernetzten Sensoren Die fortw hrende Miniaturisierung von Hardware die Effizienzsteigerung der Ressourcennutzung und die Verf gbarkeit preisg nstiger massentauglicher Sensor komponenten erschlie t ein immer breiteres Anwendungsfeld f r den Einsatz von Sensornetzwerken Zuk nftige Sensornetzwerke bestehen aus gro en Mengen von einfach auszubringenden Sensoreinheiten die sich selbst organisieren drahtlos miteinander kommunizieren Messungen durchf hren und auswerten k nnen und die Beobachtung verschiedenster Gebiete und Ph nomene erm glichen Ziel dieses Beitrags ist es Aspekte des Einsatzes von Sensornetzwerken vorzustellen die im Rahmen von Forschungsprojekten der Professur f r Geod sie und Geoinformatik der Universit t Rostock entwickelt werden Zentraler Ansatz des seit Fr hjahr 2007 vom Bundesministerium f r Bildung und Forschung BMBF im Rahmen des Geotechnologien Programms gef rderten Verbundprojekts SLEWS Sensor based Landslide Early Warning System ist die Konzeption und prototypische Entwicklung von Methoden und Technologien f r flexible und ressourceneffektive Alarm und Fr h
132. e basis of total cluster load remaining CAP size device s data and delay specifications and device s recent CFP usage In this way bandwidth under utilization and the restriction of entertaining at most seven GTS requests are avoided To an nul the constant GTS expiration the coordinator uses the period bits of the GTS characteristics field and performs GTS expiration dynamically Rather than using one whole byte in the superframe to indicate the presence of the CFP part we use one available reserve bit in the superframe 5 PERFORMANCE EVALUATION We have implemented the IEEE 802 15 4 protocol on the Tinyos 2 x 3 platform based on our system model Moteiv Tmote 2 nodes with Chipcon CC2420 transceivers are used for the implementation Now we will discuss our analysis for the application specific optimal parameters setting The values of BO and SO vary between 0 and 14 whereas the values of BI and SD lie between 15 36 ms and 251 6 s We can decide for an application specific value of BI For exam ple the value of BI can be 12 for the periodic application where nodes send their data at an interval of 60 seconds The maximum number of CFP slots for the different values of SD can be calculated by Equation 3 In order to calculate the number of slots actually needed to the device we first calcu late the total length of data packets which are generated by the device The upper layers of the device generate the data frame
133. e worn by patients should automatically but timely alert doc tors or emergency services when the patient suffers from se vere disease and nodes must lively inform the security and emergency services about the persons wounded by a ter rorist bomb blast In such applications usually many short range and battery operated devices are deployed Hence the 2 gt lt gt or Superframe Duration SD Beacon Interval Active gt heave Figure 1 The MAC superframe is divided into 16 equally sized slots Optionally it can have an ac tive and an inactive part The active part is further divided into the CAP and the optional CFP part bandwidth and energy efficiency also demand special atten tion so that many devices can communicate at a time and can survive longer without any human interaction In clinical diagnostics traditional paper based patient mon itoring is difficult complex and expensive The increase in the world population diseases and the natural and non natural incidents and with the inadequate number of doc tors and clinicians available around especially in the third world countries highlights the need of an automatic system Such system should efficiently but timely alert medical staff for any mishap or remind them for their scheduled check ups regardless to the personal s physical location It can also be used to store patie
134. e 2nd international conference on Mobile systems applica tions and services 2004 9 C E Koksal and H Balakrishnan Quality aware routing metrics for time varying wireless mesh networks EEE Journal on Selected Ar eas in Communications 24 11 1984 1994 November 2006 30 10 S B Lee K J Kwak and A T Campbell Solicitation based for warding for sensor networks In Proc of the Conference on Sensor and Ad Hoc Communications and Networks SECON September 2006 11 K Lorincz and M Welsh A robust decentralized approach to rf based location tracking Technical report Harvard University 2004 12 J Polastre R Szewczyk and D Culler Telos enabling ultra low power wireless research In Proc of the Fourth International Sympo sium on Information Processing in Sensor Networks IPSN 2003 13 N Reijers G Halkes and K Langendoen Link layer measurements in sensor networks In st IEEE Int Conf on Mobile Ad hoc and Sensor Systems October 2004 14 K Srinivasan P Dutta A Tavakoli and P Levis Understanding the causes of packet delivery success and failure in dense wireless sen sor networks In Proceedings of the 4th International Conference on Embedded Networked Sensor Systems SenSys November 2006 15 P von Rickenbach S Schmid R Wattenhofer and A Zollinger A robust interference model for wireless ad hoc networks In 5th Inter national Workshop on Algorithms for Wireless Mobile Ad
135. e absolute speed metric increase the performance of AODV in the case of more mobile sce narios The higher reliability results from more stable routes as a consequence of the longer link duration time The SBR protocol achieves the highest end to end reliabil ity if its routing messages are deferred according to the RSE metric which is in contrast to AODV where the absolute speed metric offered a slightly better performance than the RSE metric This effect can be explained as follows Hello Messages are periodically flooded by the SBR protocol in the network which results in a higher routing overhead Thus the nodes have more precise information of their neighbor hood For that reason the SBR protocol takes more advan tage from the deferring of messages than AODV 5 CONCLUSION AND OUTLOOK In this paper a new delay based approach was introduced that is able to increase the end to end reliability of protocols in mobile networks by deferring the forwarding of routing messages Optimized configurations were used instead of using the default ones since these configurations only achieve acceptable performance in networks with low mobility A closer look was taken on the speed of the nodes that are selected by the protocols to forward data traffic The simulation results showed that the nodes are able to estimate their relative movement speed by keeping track of changes in their neighborhood lists Furthermore the results pointed out th
136. e is highly affected by the implemented routing metric A variety of different metrics aiming on different aspects like success rate latency or energy efficiency have been considered in the literature In this work we focus on the basic metric from 2 send the message to the node which minimizes the distance to the destination However here we suspect main improvement potential in future experimental studies and anticipate in corporating other metrics in future performance studies as well Location information is an important ingredient of any lo calized data communication protocol Here we focused on the data communication aspects only We implemented a TinyOS module which provides an interface to request a node s position The current implementation just returns a preconfigured location value which is set in the phase of mote flashing 3 EXPERIMENTAL SETUP The following described experiments were performed on a regular 7 x 7 grid of 49 battery operated Tmote Sky sensor nodes see www sentilla com The network was deployed indoors in a single room Nodes were located on the ground No obstacles have been placed in between them Greedy routing does not require absolute node positions Any virtual coordinate system which supports computing distances between nodes is sufficient We have set the node positions according to increasing integer values for x and y coordinates in the 7 x 7 grid The distance between two neigh
137. e of monitoring phase without the pres ence of an attacker The sender s transmissions are verified on four sensors three sensors and a base station Although the acceptance intervals dynamically change gray areas the legitimate sender has 67 Scenario 1 Interval width No of sensors Interval width Scenario 3 interval width Scenario 2 1 1572 2 01772 1 10 5121010 2 1121816121146 11 019141210 311501411018 11611311315 13181716 O 4 115 115 115 112 01615 15 14 16 Table 1 Number of secure channels i e channels immune to injection attacks if one channel is broken no problem of correctly changing its transmission properties to ful fill acceptance intervals on other four sensors 3 1 Attack Scenario To analyze the impact of dynamic configuration change during an injection attack the PRNG seed the transmission frequencies and the attacker s RSS on legitimate sensors is made public hence this protection method does not depend on any secret The single as sumption is that an attacker cannot gain a physical access to the indoor WSN and cannot transmit from the same physical position as an impersonated sensor To increase the efficiency of injections we manually searched for an appropriate configuration of physical position antenna orientation and power level until we were able to use at least one frequency to successfully inject fake frames at the receiver We were interested in the number of accep
138. e two variables beacon order BO and superframe order SO BO describes the interval at which the coordinator transmits its beacons i e Beacon Interval BI and is given by 1 SO describes the length of the active portion i e Superframe Duration SD and is given by 2 The parameter aBaseSuperframeDura tion depends on the frequency range SD and BI are the decisive parameters to define the duty cycle of the network BI aBaseSuperFrameDuration gao 1 0 lt BO lt 14 SD aBaseSuper FrameDwuration D 2 0 lt SO lt BO lt 14 2 1 IEEE 802 15 4 Suitability and Limitations for Medical Applications Figure 2 shows an health care scenario where different pa tients are to be treated in a room The traffic generated by the nodes deployed in the room can be classified in two types periodic and aperiodic Periodic traffic contains the routine check up values for the patients and physical status of the room Usually these values don t have strict timeliness re quirements Therefore normal contention access method CSMA CA is used for such traffic However the aperi 58 oo gt gt u es ae ee Sage 4 o Pi v T K er i jf 7
139. ecise localization in geosensor networks Proceedings of the 4th International Conference on GIScience 2006 321 337 M nster Germany 15 Reichenbach F Born A Nash Ed Strehlow Ch Timmermann D Bill 2008 Improving Localization in Geosensor Networks Through Use of Sensor Measurement Data Proceedings of the 5th International Conference on GIScience 2008 Park City USA in press 16 Reichenbach F und Timmermann D 2006 Indoor Localization with Low Complexity in Wireless Sensor Networks In 4th International IEEE Conference on Industrial Informatics S 1018 10 Self Adaptive Load Balancing for Many To Many Communication in Wireless Sensor Networks Manuel Gonzalo Klaus Herrmann Kurt Rothermel Universit t Stuttgart Universit tsstrasse 38 Stuttgart Germany firstname lastname ipvs uni stuttgart de ABSTRACT New scenarios in wireless sensor networks where several independent sinks can ask for different kinds of data are currently emerging Former algorithms that route data to a centralized sink are not appropriate for these scenarios First solutions for these multi source to multi sink scenarios simply create independent trees for each source As a re sult some nodes become hot spots since they are included in several trees whereas nodes around them remain unused In this paper we propose a new algorithm that balances the load evenly in the network extending the network lifetime while st
140. ecuire Performance analysis of IEEE 802 15 4 contention free period through real time industrial maintenance applications 9 S Yoo D Kim Phan Y Doh E Choi and J Huh Scheduling support for guaranteed time services in IEEE 802 15 4 low rate WPAN 11th international conference on Embedded and Real Time Computing Systems and Applications IEEE 2005 A Closer Look at the Association Procedure Low Power 802 15 4 Multihop Sensor Networks Barbara Staehle University of W rzburg Institute of Computer Science bstaehle informatik uni wuerzburg de ABSTRACT IEEE 802 15 4 proposes the advantage of a standardized low power low data rate communication stack and is therefore also an option for deploying low power wireless sensor net works WSNs Most studies of 802 15 4 based WSNs con centrate on the operational phase and neglected the initial startup phase This bears however also potentials for energy savings as the 802 15 4 association procedure has to be exe cuted to make the network operational and is not optimized for low power networks In this study we point out direc tions how to perform the association in a self organizing and energy Saving way 1 INTRODUCTION Among the existing standardized wireless communication solutions the IEEE 802 15 4 standard 5 seems to be the most suitable for WSN purposes as it is targeted at low power low bandwidth networks characteristics which match most sensor network a
141. ed This system allows to get information of surrounded sensors and to put actuator states at a central place As a central user control unit a device with the name Funk Haus Zentrale FHZ 1300 was developed 3 MULTI CLIENT SYSTEMS Usually networks in a home environment should work autonomous without the intervention of users Nevertheless it is also necessary for users to have the possibility to access the communication infrastructure and attached components Sometimes it could be interesting to have knowledge about the network state or to control individual devices manually Therefore a development has been started to implement Multi Client Systems Appliances 3G WLAN Bluetooth Figure 1 Multi Client User interfaces 3 1 Client Systems Different end devices respectively clients with different capabilities are planned for the MCS At a first choice mobile end devices like mobile phones smart phones or pocket PCs are good candidates because of their mobile handset characteristics These devices should have access to the gateways over a local WiFi or Bluetooth connection or over a 3G GPRS UMTS link Other 33 possible MCS devices are standard personal computer environments like desktop PC notebooks ultra mobile PCs or at least PC based appliances for instance TVs or touch screens 3 2 Graphical User Interface AutHoNe Rule Editor OF a 14 58 heatingregulation dormitory Sere F 3 Te od
142. ed this counter is reset In case a highly loaded scenario is considered the influence of contention and collisions must also be taken into account In such a case it is desirable to avoid sending messages to nodes that have detected such problems because a high probability that these packets get lost exists Measuring the times a message could not be sent because the medium was busy gives a good measurement of contention 3 3 Assumptions Each sink defines its interests about a certain kind of sensor reading and disseminates this information through the net work Each source has a list of all the sinks it must serve and each node knows the number of hops necessary to reach each sink from each neighbor Each node maintains a list of its neighbors and their associ ated information number of sources number of paths the consumed energy and the high load indicator This infor mation is piggybacked in the data packets that nodes trans mit and is overheard by all neighbors This consumes at the most 4 bytes of the payload We define an epoch as the amount of time between two successive updates of this in formation Nodes always transmit their information of the previous epoch 3 4 Cost function In our approach we have defined the cost function depicted in Equation 1 The neighbor or set of neighbors that mini cost nj Sk gt hops n Sk w2 N ws k mize this cost function are chosen as forwarding nodes
143. eivers and or about other branching nodes Forwarding nodes have less or no information about the multicast state and just forward the multicast data to one neighbor We will also limit our discussion to core based trees where only the dedicated root node will disseminate the data while other senders would need to transmit the data to the root node first for dissemination An example topology with some branching nodes forwarding nodes and three group mem bers is shown in Figure 1 2 2 1 Overlay Multicast For the source driven scenario we can use a de centralized as well as a centralized approach Generally we distinguish between active and inactive multicast trees While data is transmitted to a multicast group the tree is in the active state with all required TCP connections for the overlay links established New nodes are not allowed to join the group 70 branching fowarding group node node member Figure 1 Example topology showing branching nodes forwarding nodes and group members while its tree is active though the joins are cached and pro cessed later When no data needs to be transmitted the tree is inactive and all required TCP connections to build the distribution overlay are closed Nodes can only join to a multicast group while its tree is inactive This limitation ensures that subscribed members get all data for a dissem ination session because late joins are avoided In the de centralized and source driven app
144. ello message interval Ah and rep resents the maximum forwarding delay Thus covers the function of a delay limiter The second factor of 2 is influ enced by and is used to divide the maximum forwarding delay into smaller steps A smaller value increases the delay for a smaller number of neighbor changes A value of ten is used in the simulations since it results in a good accuracy of differentiation in a large spectrum of neighbor list changes The additional delay has to be chosen accord ing to the net diameter in number of hops the underlying medium access layer and the traffic load of the network In most scenarios an additional delay of several millisec onds is quite sufficient to modify the topology of the net work The impact on the end to end delay of data traffic is minimized due to the fact that the cross layer approach only defers routing messages Proactive protocols are less affected by the artificial delay of the routing messages since end to end routes are maintained between all nodes in the network However the duration that is required to establish a new end to end route in reactive and hybrid protocols is increased by the additional delay Nevertheless the routes are more stable which results in less link breaks Therefore the number of route repairs is minimized Thus the overall end to end delay can be reduced while achieving a higher end to end reliability 4 SIMULATION RESULTS The OPNET Mo
145. emens com Ulrike Meyer Siemens AG Corporate Technology meyer ulrike siemens com Norbert Vicari Siemens AG Corporate Technology norbert vicari siemens com Figure 1 Condition monitoring of events Sensor nodes are attached to the machines to analyze the current state of the machine see Figure 1 The number of sensor nodes is likely to be rather small as only a single machine or a manufactur ing cell will be monitored by a single wireless sensor network For example t may be analyzed if a vent is out of balance Sen sor nodes periodically send reports to a central server control station where they are evaluated by a plant operator If a problem is identified maintenance may be scheduled 2 2 Temporary Monitoring In process automation mass losses may occur from time to time It may be hard to localize the source of the mass loss by observa tion or using only the measurements of installed sensors A tem porary monitoring system like the one shown in Figure 2 may be used to locate the source of mass loss The monitoring system uses wirelessly connected sensors e g ultrasonic flow sensors that can be easily installed where and when needed and removed again when the leakage has been found The temporarily in stalled monitoring system provides temporally and spatially more fine grained information as the placement of temporary sensors can be denser than the placement of regular built in sensor nodes 93 Insta
146. ement scheme we devised a simple experi ment In the next section we first explain the experimental setup After that the results will be discussed 5 1 Experimental Setup The measurement setup is shown in figure 2 Besides the device under measurement the setup contains a shunt an adjustable power supply unit an oscilloscope and a PC The PC is used to control the oscilloscope and the power supply unit The energy consumption P is calculated by the voltage and current which flows into the board formula 1 Because direct measurement of the current is not possible it 18 deter mined by using formula 2 P Uboard Ipoaradt 1 I board 7 Ushunt R shunt 2 The start of a measurement can be accurately determined by using a trigger signal generated by the application running on the microcontroller itself It must be pointed out that the region of the measured values leads to a certain amount of noise Thus there could be an error of about 10 but the values are suitable to show trends Sense e Adjustable Power en Supply Microcontrollerboard Shunt Hameg 8143 G 5 e l Trigger paar RS232 1 i B 68232 USB USB Oscilloscope PicoScope 5204 Figure 2 Measurement Setup The measurements where done using a TMoteSky 8 equipped with a Texas Instruments MSP 430 serie
147. ements concerning software and hardware tech nologies and the technologies to cope with these needs are described in Section 3 2 EUROPEAN PROJECT FRONTS The aim of FRONTS is to understand the theoretical back grounds of large scale adaptive heterogeneous wireless sen sor networks These networks consist of nodes that provide different capabilities and act in a dynamic environment Dy namics can either be active or passive When dealing with active mobility nodes are able to move autonomously to des ignated parts of the network Passive mobility means that nodes are connected to moving objects like vehicles or hu man beings With FRONTS we will be able to get a clear grasp of such networks It will be crucial not only for under standing the net behavior but also for supporting algorithm development in nearly all related areas The goals of FRONTS can be summarized as follows e Design a unifying framework e Obtain a set of design rules e Provide distributed adaptation techniques e Provide laws on adaptation e Obtain knowledge of combining heterogeneous nodes e Provide a set of algorithms and e Evaluate with both simulation and experiments The main issue is to develop mathematical models that de scribe the focused adaptive large scale networks of heteroge neous nodes in dynamic environments Such models include mobility for dealing with different kinds of dynamics com putation models that consider the l
148. en e Aktivit tszyklus Anwendungen k nnen h chst unterschiedliche Anfor derungen an den Aktivit tszyklus eines Knoten stel len Anwendungen zur Klimabeobachtung z B erfor dern i d R Messungen im Stunden oder Minutenab stand die Stabilit tspr fung von Bauwerken ben tigt mehrere hundert Messungen pro Sekunde ber einen begrenzten Zeitraum Knoten k nnen dar ber hinaus ausschlie lich zeitgesteuert oder durch u ere Einfl s se aktiviert werden e sr e des Netzes Die Anzahl der eigesetzten Sensorknoten kann zwi schen einigen wenigen und mehreren Hundert schwan ken e Vorhandene Infrastruktur Sensornetze verf gen in der Regel ber ein oder meh rere Datensenken an die letztlich die erhobenen Da ten bermittelt werden Es ist dar berhinaus denkbar dass dezidierte Router oder Zwischensenken existieren e Mobilit t der Knoten Knoten k nnen ber die gesamte Betriebszeit entweder ortsfest sein oder sich aktiv oder passiv bewegen e Vorhandene Aktorik Sensornetze k nnen ohne menschliches Eingreifen ak tive Regelungsfunktionen bernehmen und z B Venti le Klimaanlagen oder Alarmeinrichtungen ansteuern e Datenaggregation Me daten k nnen entweder direkt an eine zentrale Ba sisstation bermittelt werden wo sie dann ausgewer tet werden oder sie k nnen lokal auf bestimmten Knoten vorverarbeitet werden z B Berechnung des Durchschnitts Minimums oder Maximums untersch
149. en Zugriff unabh ngig von der unterliegenden Umsetzung zu gew hrleisten Prim rer Nutzen ist die verbesserte Auffindbarkeit von Sensorsystemen und ihre automatisierte Verwendung als Informationsressource Ohne erforderliches Vorwissen k nnen Sensoren und Messwerte anhand von zeitlichen r umlichen oder ph nomenbasierten Kriterien ausfindig gemacht und komplexe Informationen ber Detektoren z B Kalibrierung Arten von Messungen Messreihen und Beschaffenheit von Messdaten Datenqualit t eingeholt werden 5 GSN in Fr hwarnsystemen Ziel des Projekts SLEWS ist die Umsetzung eines dienstbasierten Fr hwarn und Informationssystems unter Ausbringung massentauglicher Detektoreinheiten ber ein funkbasiertes Geosensornetzwerk zur Detektierung von Hangrutschungen siehe Arnhardt et al 2007 Die Verwendung eines GSN verspricht eine gute r umliche Abdeckung in einem akzeptablen Verh ltnis zum Installationsaufwand Der Einsatzbereich der zugrunde liegenden Geodateninfrastruktur GDI siehe Nebert et al 2004 umfasst die Datengewinnung und Aufarbeitung die Datenanalyse zur Informationssch pfung und Visualisierung bis hin zur Verteilung von expliziten Warn und Hinweismeldungen an entsprechende Organe des Katastrophenschutzes Dabei soll die Orchestrierung flexibler ber das WWW verbundener Dienstkomponenten den Funktionsablauf von klassischen black box artigen Fr hwarnsystemen emulieren und erweitern Vorteile sind sow
150. en a typical initial toler ance interval A with which probability are the nodes able to associate In the optimal case sa n 1 for all nodes but in some scenarios it could happen that not all nodes are able to associate to the network resulting in sa n lt 1 3 2 2 Association Time t4 n The association time is obtained from simulations but it may be broken down analytically too ta n to n na n tscan c na n 1 a tta 1 The first component of ta n to n denotes the randomly distributed time node n waits between the deployment of the network and its first attempt to associate The num ber of association attempts na n is greater or equal than 1 Node n will thus scan na n times the channel need ing a time tscan c which is application specific 5 depends additionally on the number of channels to scan c which we adopt to be equal to 11 Upon an unsuccessful as sociation attempt n will wait for a before retrying until it requires finally a timespan ta for exchanging the association command messages 5 3 2 3 Association Power Consumptions Our interest is on layer 3 and below we therefore ne glect power consumptions for data acquisition handling etc and focus on estimating the energy consumptions of the transceiver For this purpose we abstract the sensor node radio unit to a state machine and use the periods a sen sor node spends in each state to estimate the transceiver power consump
151. er technologies is crucial for industrial adoption and to solve real world problems 3 TESTBED DESIGN CONCEPTS In this section we discuss the integration of the sensor nodes into our existing wireless hybrid testbed We elaborate the possibilities in regard of configuration and management 3 1 Exploiting the Infrastructure The mesh routers connected by ethernet to the testbed server constitute the backbone infrastructure for the wireless sen sor network The core part of our wireless sensor testbed in tegration WSTI approach consists of the USB connection between sensor node and mesh router which provides a serial terminal and flash interface The latter is utilized to repro gram all sensor nodes with an updated firmware image or additional software since experimentation with sensor net works requires frequent reprogramming with additional and bug fixed software Additionally approaches to distribute binary images in a sensor network do already exist 11 We intend to use them as a foundation to implement an over the air flashing feature In case such an over the air flashing process fails we fall back to the wired infrastructure and make use of the described flash interface provided by the USB connection therefore minimizing the need of manually handling the sensor nodes This will be achieved by placing a firmware image on the testbed server The mesh routers ac cess the firmware image and flash it on their locally mounted se
152. ergiebedarfes 5 ZUSAMMENFASSUNG In Abbildung 3 werden die Ergebnisse des Experiments zusammengefasst Alle Diagramme verwenden die selbe Legende die in Abbildung 3d dargestellt ist Abbildung 3a zeigt die Resultate des Energieverbrauchs den die verschiedenen Werkzeuge f r die ausgew hlten Topologien errechnet haben Trotz fehlender vollst n diger bereinstimmung sind Parallelen erkennbar et wa bei der Reihenfolge der Topologien bez glich ihres Energieverbrauchs Einem direkten Vergleich der Ener gieverbr uche stehen zudem die in einigen Werkzeugen fehlenden M glichkeiten Kollisionen und Paketwieder holungen mit einzubeziehen entgegen Weitere Details sind durch Aufschl sselung der Ergeb nisse in versendete und empfangene Pakete in den Abbil dungen 3b und 3c erkennbar Die Unterschiede im Ener gieverbrauch sind demzufolge haupts chlich auf Diffe renzen bei der Zahl der empfangenen Pakete zur ck zuf hren was auf die teilweise unterschiedlichen Sen dereichweiten und damit ver nderte Konnektivit tsgra phen und Paketverlustmodelle zur ckzuf hren ist Letztendlich gibt es deutliche Unterschiede bei der Zahl der erreichten Knoten Abbildung 3d die neben den 99 Vergleich Energieverbrauch 6 T T T T T T T Energie mAs 0 4 0 5 0 6 Wahrscheinlichkeit a Energieverbrauch im Vergleich 0 7 0 8 Vergleich empfangener Nachrichten 70 T T T T T T T
153. ergy consumption and the average consumption must also be minimal The main quantities that have been evaluated are e Percentage of packets delivered to the sinks e Total energy spent by the network in terms of the amount of messages sent in the network e The standard deviation of the node energy consump tion e The delay in the network in terms of the number of hops until a message reaches the sinks The algorithm has been implemented on top of TinyOS 8 and evaluated using the TOSSIM 9 simulator For the first experiments a regular grid has been used where the nodes can communicate with their four neighbors The algorithm has also been evaluated with a random topology where the number of neighbors varies from 2 to 8 All scenarios comprise 100 nodes 20 nodes are data sources and 8 are sinks Each sink asks for information of 10 different sources Sources send a reading every 5 seconds they are not synchronized with each other Each simulation lasted 2500 s which means that each source sent 500 readings The 91 genommen aoe of 700 Megat ve see Nera cca 9 600 ae aa ia 5 Pench een ee ee Q std dev without LB m 1 std dev with LB ee ee mean w
154. ervals and ii dynamic config uration The acceptance intervals are used to detect frames whose transmission properties are significantly different from those of le gitimate sensors while the dynamic configuration of transmission properties within a WSN increases the unpredictability of signal propagat on Similar to Figure 2 the legitimate sensors were de ployed on the ceiling of our university lab and the experiment was divided in two phases 1 Deployment phase Installation of sensors and configuration of transmission parameters including initial measurements using different power levels and transmission frequencies for defining acceptance intervals 2 Monitoring phase The network is in regular operation while being subject to an injection attack executed from various physical positions outside the deployment area The acceptance intervals are defined as u ko u ko and based on empirical data measured during the deployment phase where u is the sample median o the standard deviation of a sample and gt 1 is an environment dependent constant defining the width of the interval i e it describes within how many standard deviations the RSS is still considered to be legitimate During the deployment phase each sensor uses initial transmissions to create acceptance intervals on every other sensor 1 similar to signalprints used in 3 and a more advanced approach discussed in 1 To success fully inject fake frames
155. erwendeten Betriebssystems TinyOS bzw durch eigene Hardware zur Knotenverwaltung Beide Ans t ze werden gegenw rtig implementiert und in zuk nftige Publikationen einflie en 6 REFERENCES 1 Atmel Corporation ATMEL ATmegal28 datasheet http www atmel com dyn resources prod_documents doc2467 pdf 2002 E O Blab Sicherer aggregierender Datentransport in drahtlosen Sensornetzen Dissertation Universit tsverlag Karlsruhe Karlsruhe Germany June 2007 ISBN 978 3 86644 142 2 E O Bla L Tiede and M Zitterbart An energy efficient and reliable mechanism for data transport in wireless sensor networks In Proceedings of International Conference on Networked Sensing Systems pages 211 216 Chicago USA May 2006 ISBN 2 3 100 4 5 6 7 8 9 10 11 12 0 9743611 3 5 D Cavin Y Sasson and A Schiper On the accuracy of MANET simulators In POMC 02 Proceedings of the second ACM international workshop on Principles of mobile computing pages 38 43 New York NY USA 2002 ACM Crossbow Inc RADIO RF concepts and TOS radio stack 2006 http www eol ucar edu isf facilities isa internal CrossBow PresentationOverheads Dayl_Sect06_ RFConcepts pdf A Hinton M Kwiatkowska G Norman and D Parker PRISM A Tool for Automatic Verification of Probabilistic Systems In H Hermanns and J Palsberg editors Proc 12th Int Conf on Tools and Algorit
156. es FFDs and RFDs Handling the radio transceiver and physical medium energy detection link quality channel selection clear channel assessment are the main features of the PHY layer Whereas the MAC layer features include the beacon management channel ac cess guaranteed time slots GTS management frame val idation acknowledgments association disassociation and the security mechanisms The MAC protocol supports two operational modes selected by the coordinator The non beacon enabled mode in which MAC is simply ruled by non slotted CSMA CA and the beacon enabled mode in which beacons are periodically broadcast by the coordinator in the MAC superframe The MAC superframe structure is defined in Figure 1 The beacon is always transmitted in the first slot and is used to synchronize the attached devices and to describe the su perframe structure During the contention access period CAP the devices compete with each other using a slotted CSMA CA mechanism For the applications having latency and specific bandwidth requirements the coordinator may dedicate portions of the active superframe to that applica tion These portions are called guaranteed time slots GTSs and they form the contention free period CFP which ap pears at the end of the active part following the CAP The coordinator may allocate up to seven of these GTSs and a GTS may occupy more than one slot period The structure of the superframe is described by th
157. es the load based only on one hop information a requisite in WSN It does not require a setup phase because the different parameters in the cost function change according to the traffic in the network This cost function uses different weights for each one of its parameters which can be updated by the sinks Therefore an agent could set a higher priority for one of the metrics against the other ones e g low delay with high energy consumption The remainder of this paper is organized as follows Related work is reviewed in the next Section In Section 3 we present our algorithm and motivate all the parameters taken into account Furthermore we present the cost function used to select the next node s to the sinks Section 4 shows an evaluation of our protocol Finally Section 5 presents some conclusions and the next steps in our work 2 RELATED WORK Regarding scenarios with multi source and one sink that ad dress load balancing the protocol presented 7 forms an initial tree that gets adapted using topological knowledge In some scenarios where several sinks exist the task con sists of simply routing the packets from the sources to only one of the sinks Normally the closest sink is selected in order to save energy and make the communication more re liable In 2 the authors present an algorithm for maxi mizing the lifetime of sensor networks by reducing the total 89 energy consumption and balancing the energy
158. esentation It provides access to links between nodes in a graph structure using the communication model to decide whether there is a link between any two nodes There is a very simple edge model with no memory overhead but wast ing computation time up to a very fast model that requires much memory Finally there is the transmission model that is used to transmit messages It uses the edge model to se lect potential receivers of a sent message There are several models available that can also be put into a chain of mul tiple models Beginning with a reliable model that simply transmits every message there are also models for simulat ing message loss and delay as well as CSMA and TDMA All of these models can be arbitrary combined which leads to very flexible and powerful options in simulation behavior Also models can easily be exchanged to run an algorithm in different scenarios without any additional effort The simulation controller manages the process of simulation and provides access to the simulated world with all available nodes This allows for an easy development of centralized algorithms It is possible to write a task that is executed once and has access to the whole simulation via the con troller However Shawn has been primarily designed for simulating distributed algorithms Each node in the network can contain multiple independent processors whereby each processor may implement an individual algorithm These
159. ess link The image information will be stored inside the node The node will then load this new image and execute it These tasks already imply the main challenges in designing an OTAP application The OTAP application presented in this paper is called AirFlash and is especially designed to maintain experimental networks These are networks which vary very much in size and are reprogrammed quite often Additionally such networks are mostly heterogeneous i e the nodes inside the network run different applications The next section introduces the concept and features of the AirFlash application This includes also a comparison with already known OTAP applications After that the used node platform is introduced and important implementation aspects are addressed 2 AIRFLASH CONCEPT In a lab environment a node is reprogrammed by connecting it via cable to a PC that directly uploads the new firmware into the node s memory Instead of a cable AirFlash uses a wireless connection to the node to upload the firmware This connection is established by a so called Gateway node which is connected to the PC via a serial port The Gateway then sends out the firmware image to an addressed node The node will receive the image load it into its memory and execute it These are the main functions of the AirFlash application These functions are controlled by a host software running on the PC see Figure 1 Host Software Gateway
160. esting purposes a RFID reader a Bluetooth device and an UMTS GPRS modem for 3G connections has been included and tetsted Gateways Senso Actor Sensor Actor Senso Acor Sensor Actor Technology 1 Technology Technology 3 Technology N ZigBee RFID ISM i i 5 wy er Figure 4 Backbone architecture 4 1 OLSR Mesh based on IEEE 802 11 The system uses aWLAN mesh protocol based on the optimized link state routing protocol OLSR implemented by the olsr org community 9 This routing protocol is part of the used OpenWRT distribution called kamikaze To visualize the mesh backbone network the system uses an add on provided by 10 This add on allows a user to analyze the link quality between several meshed Wifi nodes This can be used to build up a stable working backbone network 4 2 OLSR on Multi Client Systems It is possible that every MCS can be become part of the WiFi meshed network At least for two aspects attention has to be paid On the one hand the end device needs of course an interface for IEEE 802 11 a b g 11 and on the other hand an implementation of the OLSR routing protocol is required Several devices are being tested with the implementation of the olsr org community A development for Windows mobile clients from 12 is also tested For testing purposes a network was installed with over 15 Asus routers several Notebooks running Windows XPTM and also one mobile pocket pc client It was
161. etric i e a metric which also favors high quality links over bad ones We observed a higher success of DSR on one hand but on the other hand DSR does not provide as much of the pay load which is provided in Greedy packets DSR has to store the entire route from source to destination in the message as well We were thus interested in the question how many bytes per second can be transmitted with DSR and Greedy respectively As depicted in Figure le Greedy achieves a better data rate than DSR for low cache invalidation times We think that using a metric as discussed previously will enable greedy as well to perform as good as or even better than DSR when invalidation times are increased Finally we were interested on the length of the paths taken by DSR and Greedy Figure 1f shows that Greedy takes about one hop more than DSR We observed that the paths taken by DSR are often following the network boundary By the time of writing for Greedy our log files do not reveal the paths taken 5 CONCLUSION In this work we presented an empirical performance com parison between a localized and a global routing approach Although this work is not the first one providing measure ments in a real world sensor network deployment we think that in answering the question on what are the advantages and disadvantages of localized over globalized approaches such empirical results are still lacking Our intention is to substantiate advantages
162. g algorithms Careful design of these algorithms helps to decrease protocol overhead and retransmissions and thus improves energy efficiency Cheap and small sensor nodes can only be realized if a high integration can be achieved The Institute of Nanoelectron ics offers the complete competence in the design of integrated circuits in up to date process technology The Institute of Microsystem Technology is capable of build ing very complex and sensitive sensors with small size A fo cus lies on the design of medical sensors that are distributed all over the human body An obvious application for a WSN The robustness against malicious network influences is in vestigated by the Institute for Security in Distributed Ap plications Operating mostly in ISM frequency bands and using only low transmission power WSNs are sensitive to all kinds of interferers but a wilful disturbance of the network must also be inhibited Many decisions depend on the chosen network application Long term stability autonomous operation of the network and maximum tolerable delays to establish stable control loops are necessary investigations performed by the Institute of Automation and the Institute of Control Systems 32 3 CAMPUS NETWORK One important goal of the research field SomSeD is the de ployment of a WSN on the TUHH campus An entire net work of sensor nodes will be installed permanently at the buildings to cover the whole main campus The
163. gien Nachrichtenl inge 56 Bytes 5 ee p 10 2 0 4 7 6 0 8 1 0 Ue CSMA CA H he der Antenne h 1 5 Meter R 38400 Bis 5 Stromverbrauch Funk CPU Energieverbrauch Nachricht versenden Nachricht empfangen IFunk 16 mA 5 Icpu 5 mA 1 3 siehe Abschnitt 3 2 Pm 185 2 uAs Pm 42 uAs Table 1 Simulations Parameter 3 2 Energie Die wohl interessanteste Metrik in Sensornetzen den Energieverbrauch mit unterschiedlichen Simulatoren zu erfassen ist eine Herausforderung f r sich Einerseits soll der simulierte Verbrauchswert nahe an der Reali t t liegen andererseits ist beim Einsatz verschiedener Werkzeuge darauf zu achten dass die Ergebnisse ver gleichbar bleiben Da einige der eingesetzten Simulato ren nur auf Netzwerkebene arbeiten bleibt als gemein same Datenbasis leider nur Paketempfang und versand brig Daher wurden diese Operationen auf realen Kno ten MICAz vermessen Um den Eigenschaften der Energieressource Rechnung zu tragen wurde die der Batterie entnommene Ladung mit Hilfe eines NI USB 6210 9 gemessen Dem Emp fang einer 56 Byte gro en Nachricht wurde ein Ladungs wert von 42 uAs und eine Dauer von 1 85 ms im Durch schnitt zugeordnet gemessen vom Beginn des Empfangs am CC2420 bis zur Benachrichtigung der Anwendung Der Versand eines Pakets derselben Gr e dauert hinge gen im Durchschnitt 8 34 ms und ben tigt dabei 0 185 mAs an Ladung gemessen vom
164. gorithms solving global tasks like exploration of unknown areas or target tracking We provide a coherent package con sisting of a simulation framework hardware and a remotely accessible testbed building a powerful tool for evaluation and validation of the results of FRONTS Our framework allows for easy exchangeability between simulation and real world experiment as well as rapid engineering Keywords Wireless Sensor Networks Simulation iSense Shawn 1 INTRODUCTION We are now on the verge of Wireless Sensor Networks WSNs leaving the landrush phase and becoming more and more mainstream Many complex issues have been addressed and we see a number of operating WSNs This shifts the focus from initial questions about how to get working networks at all to those about the shape of future networks and their applications Being relieved from dealing with complex low level issues we now can concentrate on high level applica tions and protocols It is our belief that soon sensor networks will exist with a number of properties Heterogeneity WSNs will no longer consist of identical devices The need to maintain networks over long pe riods of time will result in different software and hard ware generations from one family to coexist Also there will be completely different devices in networks due to interactions between networks the need for spe cialized sensory in certain areas and because networks will grow organical
165. hancement of the IEEE 802 15 4 pro tocol including the fine tuning of its parameters In this paper we discuss the suitability of the protocol and explore its limitations for health care applications We propose a solution to overcome those limitations and derive the most efficient IEEE 802 15 4 based network parameters setting for urgent medical services Categories and Subject Descriptors C 2 2 Computer Communication Networks Network Protocols General Terms Design performance standardization Keywords IEEE 802 15 4 real time bandwidth energy efficiency med ical applications 1 INTRODUCTION AND MOTIVATION Recent trend towards achieving communications between short range devices has laid down the foundation of a simple low cost low power low QoS and low data rate communi cation network called as low rate wireless personal area net work LR WPAN IEEE 802 15 4 1 which deals with the PHY and MAC layers for such networks is an emerging pro tocol for multidimensional applications such as surveillance and security home automation flood fire seismic detec tion and health care scenarios Though its pertinency and usage to many application areas has already been witnessed many open questions are there to be answered before IEEE 802 15 4 based networks can be efficiently utilized on a large scale especially for the environments where timeliness is vi tal For example a sensor node embedded in an e textil
166. has moved away from the other nodes or the other nodes are moving away from it However routing or data traffic are required to detect changes in the neighborhood Otherwise the nodes are not able to keep their neighbor lists up to date In general the routing traffic is sufficient to maintain fresh information in the neighbor list since even reactive protocols like AODV transmit hello messages periodically The basic idea of this delay based approach is to defer the forwarding of routing messages depending on the change of the neighborhood More changes in the neighborhood result in a higher delay of the routing messages Therefore the routing protocols choose nodes with a lower relative node speed since these nodes forward routing information more quickly It is obvious that a mobile node or a cluster of mobile nodes with correlated movement may have peaks in 78 their neighborhood change if they move through certain ar eas e g crossing of a road or an area of high node density Thus a mechanism is required to reduce the variation of the metric and make it robust against short temporary changes The exponential weighted moving average algorithm is used to minimize the impact of peaks The neighborhood change metric e is calculated according to Equation 1 which was empirically determined from a set of mobile scenarios with different mobility models node densities and traffic pat terns a e 1 l a X a 0 9 1
167. he management console with the feature to monitor the sensor network state in real time and interact with the sensor networks behavior by sending commands to sensor nodes 3 2 Long term Experiments Safety and security critical applications require long term experiments and studies that seldom are possible in a con tinuous manner Due to several reasons it might be neces sary to interrupt and continue test runs at arbitrary times Often we need to have a comparable environment during the whole experiment without much variance in interference and fading These kind of experiments usually have to be done at night shift when no link interruptions due to crowds of people exist or interfering radio devices are operated To expand the time span over the duration of a night we have to make a snapshot of the whole network state that can be restored later on The continuation after hardware fail ures and software updates are similar scenarios to be consid ered These issues are addressed by our concept under the name of network check points NCPs The restoration of the latest consistent network state also avoids repetition of interrupted experiments In many cases only selected time intervals containing particular events of experiments are of interest Using NCPs it will be possible to observe and eval uate how different implementations perform under the same given circumstances e g an update of a routing protocol This event oriented experiment
168. he round robin schedule consists of N rounds so that each entry of the node table is rebroadcasted in every round This information is used by the nodes that are receiving the packets Node table entries contain of the following information a flag that indicates if the node is in the neighbor list esti mated inbound PRR the received outbound PRR the num ber of neighbors and overlapping neighbors a flag indicating symmetry and the age of the entry Whenever a node re ceives a broadcast from a node that is contained in its node table the corresponding entry is updated Each packet is used to estimate the inbound PRR When the packet con tains the identifier of the receiving node the outbound PRR in the packet is stored in the table From all packets during N rounds the node can also identify the number of neigh bors of the sending node and the number of overlapping neighbors the nodes that are common for both When a packet is received by an unknown node and the node table has a free entry the node is inserted into the prepara tion list if the node table is full the packet is ignored If the estimated PRR of the inserted node is above PRRin the 25 node is a candidate to be placed into the neighbor list If the node stays longer than 50 rounds in the preparation list without entering the neighbor list the node is deleted from the preparation list and added to the blacklist If a node is in the neighbor list and its PRR
169. high mean square error In 4 the stable EWMA is preferred to the agile in order to produce more stable link estimations over time On top of such a link estimator in 5 a neighborhood protocol that is aware of asymmetrical links is described The real ization of this especially with a neighbor table size of 40 entries is not further explained TinyOS 2 x uses the Link Estimation Exchange Protocol LEEP to estimate link reliability and manage neighbor hood information 1 The Extra Expected number of Trans missions EETX is used as a link quality metric EETX is the expected value of a geometrically distribution with the probability PRR not counting the successful transmission Therefore EETX is defined as EETX 1 PRR 1 The bidirectional EETX value is based on the product of inbound and outbound packet reception rate that are discrete recur sively estimated Inbound PRR is estimated as the reception rate at node A by receiving packets from node B The re ception rate from node A at node B is the outbound PRR of node A The neighborhood table size is limited to 10 neigh bors in TinyOS To share the outbound quality information the complete neighborhood table must be exchanged If the complete link information does not fit into one packet a round robin schedule is established If the neighborhood ta ble is full and a new node is detected the neighbor with a low EETX value if existent is evicted This policy is simple but can
170. hms for the Construction and Analysis of Systems TACAS 06 volume 3920 of LNCS pages 441 444 Springer 2006 S Kellner M Pink D Meier and E O Bla Towards a Realistic Energy Model for Wireless Sensor Networks In 5th Ann Conf on Wireless on Demand Network Systems and Services WONS 2008 pages 97 100 2008 H Lee A Cerpa and P Levis Improving wireless simulation through noise modeling In IPSN 07 Proc 6th Int Conf on Inf Processing in Sensor Networks pages 21 30 New York NY USA 2007 ACM NI USB 6210 USB Multifunction DAQ http sine ni com nips cds view p lang de nid 203189 SUN Microsystems Inc Small Programmable Object Technology SPOT TinyOS Tutorial Lesson 11 TOSSIM http docs tinyos net M Zuniga Building a network toplogy for TOSSIM http www tinyos net FRONTS Foundations of Adaptive Networked Societies of Tiny Artefacts Tobias Baumgartner Alexander Kr ller and Sandor Fekete IBR Algorithms Group Braunschweig University of Technology Braunschweig Germany t baumgartner a kroeller s fekete tu bs de ABSTRACT The European project FRONTS aims at understanding adap tive large scale heterogeneous wireless sensor networks act ing in a dynamic environment This includes designing dif ferent mathematical models beyond the state of the art for such networks Based on these models a new network in frastructure will be developed followed by a large set of al
171. hop on Embedded Networked Sensors Tampa Florida USA 11 2004 An Approach towards Adaptive Payload Compression in Wireless Sensor Networks Extended Abstract Andreas Reinhardt Matthias Hollick Ralf Steinmetz Multimedia Communications Lab Technische Universitat Darmstadt Merckstr 25 64283 Darmstadt Andreas Reinhardt kom tu darmstadt de ABSTRACT Most nodes in current wireless sensor networks are battery powered and hence strongly constrained in their energy bud get While a variety of energy efficient MAC protocols specif ically tailored to sensor networks has been developed the data rate limitation of the underlying hardware still repre sents a lower bound for the time required to transfer pack ets and thus directly contributes to the energy requirement for transmissions Further energy savings for given plat forms can only be achieved by downsizing the packet e g by means of in network processing or data compression In this paper we present our approach towards an adaptive packet compression framework for sensor network applications that compresses sensor data with the locally optimal energy eff ciency ratio 1 INTRODUCTION Wireless Sensor Network WSN deployments commonly comprise sensor nodes which distributedly take measure ments process the data and subsequently forward the re sults to other nodes or external sinks 1 Most existing platforms are powered by batteries and hence inherently limited
172. hus the number of reference points shall be minimized but must still allow proper and complete operation Therefore we define the ratio d reference points anchors to be as central metric for the solution s quality and try to maxi mize Q Since we consider a grid like pattern for the anchors we also expect a regular pattern for the optimum reference point legend coverage zone SOE Figure 2 Some basic layouts each consists of four overlapping circles of equal radii forming an explicit coverage area solid black alignment SNoW Bat uses a regular grid for several rea sons It is easy to install contributes to the performance of the self organizing HashSlot 2 3 wireless communica tion protocol and allows easy adaption to the environmental conditions The last point allows to easily provide complete localization service coverage throughout the entire opera tional area As figure 1 shows the selected grid constant L depends on the minimum distance dmin of the mobile objects from the anchor plane and the resulting minimum ultrasound coverage zone Zmin According to 3 L is com puted to always guarantee a minimum number of four nodes within an ultrasound coverage zone Z independent from its overlay position within the grid Thus L stays fixed after system installation and obeys to the spatial geometry and hardware constraints during system design On the other hand dmax limits the max
173. ic J B Schmitt RELIABLE MULTICAST IN WIRELESS SENSOR NETWORKS ne 69 G Wagenknecht M Anwander M Brogle T Braun FIRST RESULTS OF A PERFORMANCE COMPARISON OF DYNAMIC SOURCE ROUTING VERSUS GREEDY ROUTING IN REAL WORLD SENSOR NETWORK DEPLOYMENTS 73 H Frey K Pind HOW TO TAKE ADVANTAGE FROM CORRELATED NODE MOVEMENT IN MOBILE SENSOR NETWORKS ccsccccccsssscccccsseccceccussecceccuusecccccuusececscuuececessuececcessuseccessaucescessauceecessaececeeeanns IE A Klein GHOST SOFTWARE AND CONFIGURATION DISTRIBUTION FOR WIRELESS SENSORZACT OR NETWORKS ee 81 M Baunach POSITIONIERUNG UND SENSOR WEB ENABLEMENT IN 85 K Walter A Born SELF ADAPTIVE LOAD BALANCING FOR MANY TO MANY COMMUNICATION IN WIRELESS SENSOR NETWORKS Qu ceccccccccecessssscccesssssssecccecssssseeeeeseaas 89 M Gonzalo K Herrmann K Rothermel FROM ACADEMIA TO THE FIELD WIRELESS SENSOR NETWORKS FOR INDUSTRIAL USE 93 Falk H J Hof U Meyer Ch Niedermeier Sollacher N Vicari VERGLEICHBARKEIT VON ANS TZEN ZUR NETZWERKANALYSE IN DRATTELOSEN SEN NEE ZEN e ee ee 97 J Wilke F Werner M Bestehorn Z Benenson S Kellner E O Bla FRONTS FOUNDATIONS OF ADAPTIVE NETWORKED SOCIETIES OF TINY ARTEFACTS 101 T Baumgartner A Kroller S P Fekete C Becker D Pfisterer VORWORT In dem vorliegenden Tagungsband sind die
174. ied licher Me werte um das Kommunikationsaufkommen zu verringern e Netzdynamik Die Menge der eingesetzten Knoten wird gew hnlich schon zu Beginn festgelegt Ein System kann aber auch so eingerichtet sein dass w hrend des Betriebs ohne gro en Aufwand zus tzliche Knoten hinzugef gt wer den k nnen e Zeitmessung Sensorknoten k nnen entweder ber lokale synchroni sierte Uhren verf gen oder s mtliche Zeitmessungen werden wenn n tig zentral vorgenommen e Lokalisierung In den allermeisten Einsatzszenarios ist man an der geographischen Position der Sensorknoten interessiert Diese kann z B durch dezidierte Hardware auf den Knoten ermittelt werden GPS Receiver Andere M g lichkeiten sind verteilte Lokalisierungsverfahren die auf Signalst rke oder laufzeiten beruhen oder eine manuelle Positionszuordnung e Rekonfigurierbarkeit Das Ausma in dem Knoten w hrend des Betriebs rekonfigurierbar sind kann unterschiedlich gro sein Knoten k nnen in ihrem Verhalten festgelegt sein oder es kann ge ndert werden Die Spannbreite reicht dabei von einzelnen Parametern die ge ndert werden k n nen z B L nge der Mefintervalle bis zu einem fast vollst ndigen Austausch des Programms OTAP Es soll bemerkt werden dass zwei unterschiedliche Imple mentierungen innerhalb desselben Einsatzszenarios auch zu unterschiedlichen Auspr gungen der genannten technischen Merkmale f hren k nnen
175. ient source to multisink routing in wireless sensor networks Signal Processing 2007 v 87 n 12 pp 2934 2948 P H Hsiao A Hwang H T Kung D Vlah Load balancing routing for wireless access networks In Proceedings of IEEE INFOCOM 01 Anchorage Alaska USA 2001 pp 986 995 Hill J Szewczyk R Woo A Hollar S Culler D Pister K System architecture directions for networked sensors In ASPLOS IX Proc of the 9nt Int Conf on Architectural Support for Programming Languages and Operating Systems 2000 93 104 Levis P Lee N Welsh M Culler D TOSSIM accurate and scalable simulation of entire TinyOS applications In Proc of the 5th Symp on Operating Systems Design and Implementation OSDI 2002 131 146 From Academia to the Field Wireless Sensor Networks for Industrial Use Rainer Falk Siemens AG Corporate Technology rainer falk siemens com Christoph Niedermeier Siemens AG Corporate Technology christoph niedermeier siemens com ABSTRACT Fundamental differences exist between academic research on wireless sensor networks and industrial wireless sensor networks as envisaged by the BMBF funded project ZESAN Their re quirements and underlying assumptions are described to bridge the gap between research and industrial application 1 INTRODUCTION Wireless sensor networks have been a hot research topic ever since the vision of smart dust was articulated Nowadays sensor networks are on
176. ill providing a small delay in packet delivery Based on the information of one hop neighbors and the number of hops to the sinks the algorithm attempts to reduce both the delay and the power consumption We present the cost function used to select the next hop s towards the sinks The evaluation of the protocol demonstrates its ability to fulfill its target 1 INTRODUCTION A wireless sensor network WSN is a multi hop ad hoc wireless network composed of hundreds or even thousands of autonomous nodes with disposable batteries and sensors attached These sensor nodes provide physical information about the environment such as acoustic light or tempera ture measurements Due to new capabilities offered by more powerful sensor nodes new scenarios are emerging where data must be de livered to several sinks Consider a wireless sensor network which can provide several kinds of data Independent agents with different kinds of applications running on them can ac cess the network and ask for a certain kind of data each one independent from each other Hence each source has to de liver its information to a certain group of sinks which can be different for each source Another example is the increas ing use of actuators in WSN where each actuator requires different kinds of information from the network These new scenarios require a many to many communication paradigm Unfortunately current protocols used for many to one com muni
177. imited memory and lim ited knowledge of the nodes and cooperation models that cover homogeneneous hierarchical and diversified systems With the aid of the resulting unifying framework we intend to get a consistent working set of design rules of such sys tems These rules will help us to develop adaptation tech niques for which we provide basic laws This includes the effect of adaptation on the system performance the cost of distributed coordination regarding adaptation the commu nication overhead overhead in terms of energy consumption and possible trade offs However these adaptation tech niques will help designing a dynamic and adaptable network infrastructure to cover nodes that adapt to each other and to changing needs This includes both the communication structure between nodes and general data access even when the data moves dynamically in the network Another objective is the combination and interaction of het erogeneous nodes to obtain a useful global behavior A net work of nodes that have different capabilities should combine the various strengths resulting in a net that is much more powerful than a network of homogeneous nodes can ever be A crucial issue here are strategies for role assignment that act dynamically according to the current needs of the 102 system Different capabilities of nodes lead also to new ap proaches for security For example work can be outsourced to more powerful nodes in the network
178. imulation runs the results obtained from the individual runs varied more strongly as ta depends strongly on the startup time of the different nodes Eq 2 makes clear that the energy a node consumes 63 200 150 100 t a n sec 100 80 55 30 Node ID n A 2 Figure 3 ta for varying activity and strategy 1 associating to the network is proportional to the number of association attempts na n it has to make Our studies showed that in most cases node 9 has to make the most association attempts the energy consumptions of node 9 are thus suitable as benchmark metric We illustrate the asso ciation energy consumptions of node 9 4 9 for a trans mission output power of 15 dBm which corresponds to a node being able to communicate with its next hop neighbor in Fig 4 Estimations for the energy consumptions were obtained from Eq 2 under strategy 2 U 0 x T are presented The most straightforward observation is that energy consumptions obtained for 4 m are smaller than for g 5m This is in accordance with Fig 2 as an in creased network density reduces na n Next the curves representing results for the same g intersect indicating that the optimal length of a depends on the activity ratio pw Moreover the curves increase strongly for pw lt 5 lead ing to association energy consumptions in the range of 10 J and make it hard to identify the most advantageous
179. imum number of anchors within the largest possible ultrasound coverage zone Zmaz Unfor tunately no analytical solution for the exact number of grid points within a circle is yet known However approxima tions for the so called lattice point problem are presented in e g 8 9 Anyway the reference points must obviously be placed at distance dmazx from the anchor plane Then their final number decreases along with increasing difference dmax Amin Still the optimal alignment remains to be found The idea so far is to initially place four non collinear ref erence points in a way to maximize the intersection area of all resulting coverage zones Fig 2d shows an exam ple This intersection produces the so called explicit cov erage area ECA Then the just created basic layout is re peated with a certain distance to implicitly cover the anchors in between the explicit areas implicit coverage area ICA too Fig 3d shows a corresponding alignment of reference points Indeed this special basic layout is rather adverse since several reference points are close to each other and the implicitly covered areas are very small If we increase the space between the basic layouts to enlarge the implicit coverage areas we might require some additional reference points to keep fourfold coverage e g the bold circle in Fig 3a During our analysis we found that the proportion between explicitly and implicitly covered areas shou
180. in de Michael Baar Abstract Sensor network hardware designs consist of a central micro controller to which sensors and communication peripherals are connected Resource arbitration and concurrency man agement must be implemented in software Existing hardware arbitration mechanisms use explicit locking to protect against resource conflicts Explicit locking may lead to deadlock which must be avoided for long term sensor network deploy ments We present a power saving resource arbitration archi tecture that is deadlock free portable and resource efficient The architecture explicitly manages inter device dependencies to know what devices to power down Categories and Subject Descriptors D 4 7 Operating Systems Organization and Design Real time systems and embedded systems Keywords hardware abstraction resource arbitration fault tolerance operating systems wireless sensor networks 1 INTRODUCTION Sensor network hardware designs consist of a central mi crocontroller to which sensors and communication peripher als such as radio transceivers are connected Because peri pherals are used both by applications and by the operating system access to the peripherals must be controlled with an arbitration mechanism We present a power saving resource arbitration architec ture and hardware abstraction layer HAL that provides portability of applications and drivers between different hardware platforms Unlike existing res
181. in the tree a network node is the higher its delay becomes due to the data be ing multiplexed with flows from other nodes higher up in the tree and because of forwarding delays This leads to a very uneven distribution of experienced delays both worst and average case however this is rarely necessary or de sirable behavior Often all data from a network is equally important and a high penalty for outliers endangers the network s utility To remedy this and to provide a worst case behavior closer to the average case we propose a multiplexing scheme giving flows with a longer traveled distance priority over flows that have passed fewer hops We will show that this scheme improves fairness in sensor networks while adding little complexity to their analysis 1 INTRODUCTION When looking at a sink tree from the perspective of a sin gle flow of interest the resulting characteristics are as shown in Figure 1 As can be seen the flow is multiplexed with more and more flows as it gets closer to the sink Under multiplexing schemes like FIFO or when using arbitrary multiplexing this results in potentially high delays favoring flows with a shorter route to the sink However in most sen sor network applications where all data is equally important that kind of behavior opposes the requirements This becomes even more of an issue when dimensioning a network based on its worst case behavior e g by using Sensor Netwo
182. in their energy budgets Once the entire battery charge has been consumed the nodes stop operating and need to be replaced Assuming a constant battery charge there is a linear relation between power consumption and node lifetime wherein the lifetime decreases with rising cur rent consumption Even solar powered nodes such as the Trio platform 4 cannot completely tackle this challenge as they rely on deployment in areas regularly exposed to direct sunlight It follows from the correlation between current consumption and node lifetime that maximizing a node s lifetime can only be achieved by minimizing its overall energy requirements As available sensor node platforms commonly employ dedi cated radio transceivers sensors and peripherals selectively disabling these components allows to measure the node s en ergy consumption in different configurations Exemplarily the current consumption figures for the widely used tmote sky platform have been taken from the datasheet 9 and were plotted against each other in Figure 1 It is clearly visible that the radio transceiver exhibits a power consump tion that is about one order of magnitude higher than the corresponding values for the used microcontroller As the employed CC2420 radio transceiver does not pro vide low power modes a common solution to the problem of its comparably high energy consumption is limiting the time of its operation and putting it into sleep mode for the
183. ink estimation 6 extends this WMEWMA estimator into a hy brid link estimator that accumulates information from all layers of the sensor node networking stack However es timation techniques based on WMEWMA performs poorly on medium quality links These links often offer the high est routing progress 3 suggesting the need for more precise estimation methods for medium quality links The assumption underlying the majority of existing link estimation concepts 15 that packet losses inside one measure ment period occur independently of each other i e they fol low a Bernoulli distribution This assumption has been chal lenged before in research 4 14 The analysis of our data in Section 4 supports the hypothesis that the assumption of in dependent packet losses is not appropriate at the fine grained time scales dealt with in this paper In addition to online link estimators there has been signif icant research in link modelling and link measurements for WSN 1 13 15 17 21 23 For example Koksal et al 9 de velop metrics to model long term link quality and short term link dynamics Additionally Cerpa et al 3 provide statisti cal models of radio links in WSN including short term and long term temporal characteristics In contrast to these ap proaches STLE does not aim to provide link models but to identify phases of reliable or unreliable connectivity at run time 21 0 8 Good links quali
184. into account 6 The results are also taken from a limited set of topologies It is evident that even small changes in parameters can have a large in fluence on the results of a Network Calculus analysis This calls for a closer examination of the conditions under which LFF has a significant advantage over arbitrary multiplex ing It would of course also be interesting to analyze the impact of LFF on a real network where the worst case is usually rarely encountered We hope to present results of the average case impact in the near future Overall the exhibited behavior is overcompensation The longest flows benefit the most while the shortest flows do not show any improvement This is of course different from the intended result which is to balance the delay bounds for all network participants as well as possible A few strategies to reach the goal could be e Data scaling 6 By taking data transformations like compression or aggregation into account the nodes close to the sink may experience a vastly lower load than in a model that just forwards data unaltered e Topology adaptation If a node becomes backlogged for too long it may decide to skip a hop in the original network topology e g by using a higher transmission power This may serve to avoid bottlenecks but of course there may be issues with medium access due to larger collision domains e Priority Injection When a node is backlogged with upstream data for a cert
185. ios long term observa tion may be limited Experiments revealed that the limited memory of the DSN Nodes restricts backbone traffic because only a small amount of packets can be buffered on each node at a time To avoid debug message loss at backbone level in a setup with 25 DNS Nodes each DNS Node could send no more than one packet every two seconds Thus scalabil ity problems might arise The TKN Wireless Indoor Sen sor network Testbed TWIST at the TU Berlin 10 pro vides a large scale testbed for wireless sensor devices So far about 90 sensor nodes have been deployed Network Attached Storage NAS devices called super nodes form the backbone of the sensor network by providing a power source and functioning as gateway to wired infrastructure Multiple xeyesIFX and Telos off the shelf sensor nodes are attached to a super node via USB hubs and cables An important feature is the power supply control which ac cording to the USB Hub Specification 2 0 enables software side port power switching With this feature the simulation of dynamic topologies common in sensor networks caused by unresponsive moving or newly deployed nodes becomes very convenient While the super nodes run a customized Linux TinyOS is in use for the sensor nodes with extensions to offer remote procedure function calls RPC The Korea Advanced Institute of Science and Technology KAIST 12 reports that their wireless mesh network called WiSEMesh is used as backhau
186. is so called wiring is completely static It is not possible to change callbacks or load new functionality at run time whe reas we also support dynamic modification loading of new virtual devices at run time The Integrated Concurrency and Energy Management ICEM proposed by Klues et al in 7 shifts responsibility from the application to the core which reduces application complexity In ICEM shared resources are explicitly acquired and released by clients which cooperatively use locked re sources For reliable operation a deadlock recovery strategy needs to be added to this approach Power management is done by special devices which are selected as default devices I O request queues are used to optimize switching Our archi tecture has a very small overhead between consecutive opera tions on the same shared virtual device so we can do with out the additional complexity and administrative overhead introduced by queuing requests Completely transparent use of our HAL removes complexity from applications while still being efficient 4 CONCLUSIONS We present a novel deadlock free resource arbitration for sensor nodes that integrates configuration and power man agement completely transparent to the application By design ing a deadlock free architecture we remove a hazard for sys tem failure We are currently implementing this HAL on the Contiki operating system 8 for the ScatterWeb MSB Av2 platform REFERENCES 1 M Baar
187. is below PRRout the node is removed from the list and added to the blacklist If the neighbor list is full and in the preparation list a link with PRR gt 0 86 exists an entry from the neighbor list must be evicted to provide space for the new potential neighbor Therefore a screening process exists to choose the most ap propriate neighbor that is evicted from the neighbor list The following rules are applied to the list containing all neighbors and the candidates from the preparation list The rules are processed one by one until a single node remains If the new potential neighbor is selected the neighbor list is kept unchanged and the node is inserted in the blacklist and removed from the preparation list Rule 1 Select asymmetry links Rule 2 Select nodes with neighbors Rule 3 Select highest overlapping Rule 4 Select less unknown neighbors Rule 5 Select lowest bidirectional PRR Rule 6 Select newest neighbors The first rule tries to evict nodes that are asymmetric from the list If several nodes are found or none of the entries are asymmetrical the next rule is applied This rule pro tects those nodes from being removed that have no other neighbor The next rule selects those nodes that have the highest number of overlapping neighbors since those nodes are connected via one of the common neighbors with a high probability If there are still several nodes having the same properties those nodes are selected which provide the s
188. is woken up it can receive an image via radio and store it into a memory slot By another command the module can afterwards invoke the bootloader that loads the image from the slot overwrites the old application including module in the program memory and starts the new application contained inside the image see Figure 2 12 Ext Memory AirFlash module Slot 0 Slot 1 Slot 2 Bootloader Goldenlmg Figure 2 AirFlash components inside node The implementation of the bootloader and the AirFlash module will be discussed in the following 4 1 Bootloader The bootloader has the main purpose to read an image from the external memory process an error check and then write the image file into the upper part of the microcontroller s program memory The bootloader is invoked from the main application by setting the program counter to the bootloader s start address The information from which slot to load an image from the external memory is passed to the bootloader via a command flag in the microcontroller s EEPROM These flags are also used to signal the application if the image was successfully loaded This information has to be passed through the EEPROM since the RAM of the microcontroller will be reinitialized when changing between main application and bootloader Before starting to write the image into the internal program memory the bootloader checks a fl
189. ithout LB mean with LB 200 1 1 1 1 2 4 6 8 Number of sinks Figure 1 Standard deviation of the node energy consumption in terms of sent messages experiments have been repeated at least five times and the presented results represent their average Several configurations have been tested but due to lack of space only some significant results are presented in this pa per In Fig 1 we compare an approach where the load balancing is not considered i e in cost function w3 w4 ws we 0 with another approach that minimizes contention and bal ances the load in the network i e w 80 w2 100 w3 8 wa 1 w5 1 we 40 We can see how the standard devia tion improves with our approach For the case of 6 sinks the percentage of delivered packets to the sinks even increases 10 while the total energy consumption only increases 4 This promising results show how our algorithm is able to balance the load in the network while maintaining the en ergy consumption and delay low 5 CONCLUSIONS AND FUTURE WORK In this paper we presented a routing algorithm tailored to scenarios with multi sources and multi sinks in WSN which balances the load in the network We proposed a cost func tion used by nodes to select the forwarding node s towards the sinks This function uses local information to balance the load among the neighboring nodes while not increasing the delay significantly As the total energy consum
190. iver driven centralized approach joins are sent to the source which then acts as in the source driven cen tralized approach In the de centralized receiver driven ap proach joins cause intermediate nodes to react as in the Overlay Multicast case They either become forwarding nodes for that group if they are not handling that group yet or become branching nodes if applicable Acknowledg ments are handled the same way as described above in the de centralized source driven approach 2 2 3 Protocol Stack Figure 2 shows a possible protocol stack for reliable multi cast in WSNs End to end reliability for reliable IP based Multicast is ensured by REMC Reliable Multicast and for Overlay Multicast by SNOMC Sensor Network Overlay Multicast To avoid unnecessary end to end retransmis sion or caching in branching we use a hop to hop reliable network protocol realized by H2HR Hop To Hop Reliabil ity in combination with the MAC protocol This allows to directly delete cached packets after successful transmis sion to the next hop To optimize the mentioned protocols information is exchanged across different layers using the cross layer interface For example the MAC layer informs H2HR about the successful or unsuccessful transmission of a packet H2HR is caching the packet until it has been transmitted successfully Additional neighborhood informa tion for deciding how to forward multicast packets and the involved paths is also
191. ize with their neigh bors Allowing the sensor nodes to sleep a fraction 1 py of a fixed epoch length T this algorithm operates between the 802 15 4 MAC and a ZigBee routing layer and promises acceptable packet delivery ratios even at low duty cycle for the price of an increased end to end delay 7 The question how such a duty cycle network can success fully start up both autonomously and efficiently has not yet been considered Each sensor node wanting to join a 802 15 4 network also called personal area network PAN has to associate i e exchange a sequence of organizational messages with the PAN coordinator or with an other node which already has associated with the PAN This procedure is mandatory for 802 15 4 networks and has not been de signed under consideration of duty cycled or lossy networks A closer look on the energy optimization potentials of the association procedure will therefore be presented in the fol lowing In Section 2 we review related work In Section 3 we formalize the association procedure using an analytical framework In Section 4 we present some early simulation results before we conclude in Section 5 and outline our fu ture research 2 RELATED WORK One of the first performance evaluations of 802 15 4 in ns 2 9 is reported in 11 This code is still the base of the actual 802 15 4 WPAN ns 2 33 simulation framework we adapted for our purposes Among other mechanisms
192. k a toolkit for the development of wsns In Proceedings of the 4th European Conference on Wireless Sensor Networks EWSN 07 pages 195 211 January 2007 5 Freie Universit t Berlin Scatterweb project 6 Fuxjager D Valerio and F Ricciato The myth of non overlapping channels interference measurements in ieee 802 11 In Proc Fourth Annual Conference on Wireless on Demand Network Systems and Services WONS 07 pages 1 8 2007 G ne Blywis Juraschek and Schmidt Concept and design of the hybrid distributed embedded systems testbed Technical report Freie Universit t Berlin 2008 8 G nes Blywis F Juraschek and Schmidt Practical issues of implementing a hybrid multi nic wireless mesh network Technical report Freie Universit t Berlin 2008 9 G ne B Blywis and J Schiller A hybrid testbed for long term wireless sensor network studies In Int Workshop on Sensor Network Engineering IWSNE 2008 10 V Handziski A K pke A Willig and A Wolisz Twist a scalable and reconfigurable testbed for wireless indoor experiments with sensor networks In REALMAN 06 Proceedings of the 2nd international workshop on Multi hop ad hoc networks from theory to reality pages 63 70 New York NY USA 2006 ACM 11 P J Marr n Gauger A Lachenmann D Minder O Saukh and K Rothermel Flexcup A flexible and efficient code update mechanism for sensor netw
193. kground process Writes received images to external memory jump to bootloader after successful image reception Provides AirFlash status for host This leads us to the last part of the AirFlash application namely the host software This software runs on a PC and controls the image transfers to all nodes 4 3 Host Software The implementation of the host software can be divided into three main parts The first part 15 an interface to the serial port of the PC that transmits packets to the radio Gateway The second component processes the AirFlash protocol The last component realizes a user interface In this paper only the basic functionality of this application is presented For further details please refer to 7 In order to transfer an image to a node the image file 15 loaded from the PC s hard drive into the application The integrity of the image is verified by checksums It is also checked if the image would fit into a slot In a further check the so called AirFlash signature 10 bytes is read from the image file which 15 only present in AirFlash capable images If this signature is missing the image will not be transferred since the node will not be reachable via AirFlash after being reprogrammed with a non AirFlash image A second signature contains information about the AirFlash version the image was compiled with This ensures that also the host software is compatible to the image Additionally the AirFlash ver
194. l network for a WSN but no publiction with further information does exist 5 CONCLUSION In this paper we described the integration of our wireless sensor network into an existing IEEE 802 11 mesh network infrastructure The work in progress hybrid testbed offers advanced features once the initial setup phase has been fin ished We presented the concept of architectural integra tion of the wireless sensor networks with our wireless sensor testbed integration WSTI approach We discussed the ben efits for the mesh network gained through the kernel inter face WSKI Our concept of network check points NPCs will simplify long termed and disrupted experiments and help to understand and evaluate networks and protocols The extensive data gathered by our testbed shall provide more realistic simulation input Finally an XML based domain specific language to specify execute and evaluate experi ments in both parts of the testbed is being developed 6 ACKNOWLEDGEMENT This work presented has been partially sponsored and sup ported by the European Union OPNEX project 32 7 REFERENCES 1 OPNEX Optimization driven Multi Hop Network Design and Experimentation http www opnex eu 2 Ralink technology Ralink RT2501U chipset product specification 3 WISEBED Wireless Sensor Network Testbeds http wisebed eu 4 Dyer J Beutel L Thiele Kalt P Oehen K Martin and P Blum Deployment support networ
195. ld be roughly bal anced to omit extra reference points Then implicitly and explicitly covered anchor sets with about the same size al ternate Fig 3a c shows some more basic patterns with balanced proportion Regarding the metric describe above table 1 shows that the legend coverage zone add coverage zone ICA 4 PARN eo 2 Las gt Figure 3 Slightly different alignment strategies when using basic layouts a b and d from Fig 2 Table 1 Quality Q of various alignment strategies for Tmax 9L quality Q is exceptionally good for these balanced align ments since few reference points are required for calibrating many anchors Hence pattern 3a offers the best quality of the analyzed strategies although it requires additionally reference points Still the achievable precision for each anchor s position esti mation needs careful consideration Indeed the calibration accounts significantly to the accuracy of the localization sys tem s regular operation Thus we recommend the following method for checking the anchors and to perform partial re calibration of the anchors by additional measurements where necessary 1 Calibrate the system as described above 2 Operate the system to localize objects at some addi tional well known test points by using each anchor at least once 3 Compare the correct known position to the estimated
196. lisierung in Ad Hoc Geosensornetzwerken mittels geod tischer Ausgleichungsrechnung In GIS Zeitschrift f r Geoinformatik 2008 Nr 1 S 4 16 7 Botts M 2007 OGC White Paper Sensor Web Enablement Overview and High Level Architecture Version 3 URL http www opengeospatial org pressroom papers rev 28 12 2007 OGC 07 165 8 Delin K A 2002 The Sensor Web A macro instrument for coordinated sensing Sensors 2002 2 S 275 9 T Huang C Blum B Stankovic J Abdelzaher T Range free localization schemes for large scale sensor networks 2003 81 95 San Diego CA USA 10 Heunecke O 2008 Geosensornetzwerke im Umfeld der Ingenieurvermessung In Forum Zeitschrift des Bundes der ffentlich bestellten Vermessungsingenieure 02 2008 S 357 364 11 Nebert D D et al 2004 Developing Spatial Data Infrastructures The SDI Cookbook Version 2 0 URL http www gsdi org publications asp rev 25 01 2004 Letzter Zugriff am 25 02 2008 12 OGC 1998 OGC White Paper The Digital Earth Understanding our planet in the 21st Century URL http www opengeospatial org pressroom papers 31 01 1998 13 Reichenbach F 2007 Ressourcensparende Algorithmen zur exakten Lokalisierung in drahtlosen Sensornetzwerken Dissertationsschrift Universit t Rostock 14 Reichenbach F Born A Timmermann D Bill R Splitting the linear least squares problem for pr
197. lled flow sensor 0 8 1 5 4 0 Figure 2 Temporary monitoring in process automation used to find source of mass loss 2 3 Decentralized Process Control Intelligent field devices and an industrial sensor network can be used to implement decentralized control mechanisms To do so sensors measure actual values and send them to intelligent field devices acting as local controllers These field devices check if actual values match specified values and react if necessary Addi tionally the sensor nodes report measured values to the control station so that the plant operator can react depending on the state of the overall process Figure 3 shows an example in which four local controllers and the control station receive measurements from sensors over the sensor network Control Figure 3 Decentralized control in process automation 94 3 Academic Assumptions vs Industrial Ap plications Significant differences exist between assumptions made for wire less sensor networks in academic research and wireless sensor network applications envisaged for process and factory automa tion and other industrial applications The differences of the un derlying assumptions made regarding deployment use of infra structure operation self organization integration and number of nodes are described 3 1 Deployment The first difference arises regarding the deployment of sensor nodes Plants are carefully planned and the employed networks a
198. lt werden wel che einzelnen Mechanismen f r verschiedene Klassen ber haupt zur Verf gung stehen Diese sollen dem Anwender beim Entwurf einer konkreten Sicherheitsarchitektur unter st tzen 5 ZUSAMMENFASSUNG In drahtlosen Sensornetzen kann eine gro e Spannbreite an unterschiedlichen Techniken und Architekturen zum Einsatz kommen Bei der Auswahl und dem Entwurf von Sicherheits mechanismen ist dies zu ber cksichtigen Mit dem vorgestell ten Klassifikationsansatz hoffen die Autoren Informations sicherheit in Sensornetzen beherrschbar zu machen 6 LITERATUR 1 Radiasense Inc http www radiasense com 2 A Arora Ramnath Ertin and et al ExScal Elements of an extreme scale wireless sensor network In 11th IEEE International Conferenceon Embedded and Real Time Computing Systems and Applications pages 102 108 IEEE 2005 3 M Hinkelmann R Reischuk D Pfisterer and S Ransom Ein weiterer Vorschlag zur Einteilung der Merkmale von Sensornetzen 2008 Internes Dokument 4 Krau Schneider and Eckert Defending against false endorsement based DoS attacks in wireless sensor networks In First ACM Conference on Wireless Security WiSec 2008 pages 13 23 2008 5 5 Lee and Y Choi A secure alternate path routing Computer Communications 30 1 153 165 2006 6 J Lynch and Loh A summary review of wireless sensors and sensor networks for structural health monitoring The Sho
199. ly Dynamics Other than in today s networks different kinds of dynamics will happen at the same time in a net work Some nodes may enter and leave the network due to off duty cycling Others are moving with some external carrier like vehicles animals or water flows Claudia Becker and Dennis Pfisterer Institute of Telematics University of Lubeck Lubeck Germany becker pfisterer itm uni luebeck de Finally some nodes may actively move around to in vestigate events help bridging communication gaps or populate unexplored areas with passive nodes An other source of dynamics is the environment In long lasting networks there may be huge changes in the surroundings and therefore in the constraints and ob jectives of the network Massive Sizes When issues about information exchange between neighboring sensor networks are solved huge sensor networks arise instantly Also continuously de creasing hardware costs and ongoing research on real sensor nodes lead to the feasibility of obtaining a great amount of nodes Hence current algorithms that were tested on mostly a few hundred nodes must also scale up to hundreds of thousands of nodes The FRONTS project 2 is part of the European Seventh Research Framework Programme 1 It is a collabo ration of 11 European working groups aiming at investigat ing the algorithmic foundations of future WSNs and provid ing answers for questions that arise now The
200. mall est number of unknown neighbors Those nodes are not that important for insuring connectivity These last two rules are protected by a hysteresis in order to prevent an oscillation ef fect In rule 5 the nodes with the lowest bidirectional PRR are selected to be removed from the neighbor table This PRR is the product of the estimated inbound and received outbound PRR If still more than one node is selected the newest neighbors are preferred since the older neighbors in sure more stability If still several nodes are in the list after applying the last rule a node is picked randomly The ns 2 simulator is used for validating Mahalle To com pare the connectivity awareness of Mahalle a simple neigh borhood management protocol is also implemented which uses only the inbound and outbound PRR values provided by ALE Additional the LEEP 1 protocol integrated in TinyOS 2 x see Section 2 is ported to ns 2 Several dif ferent densities with 200 randomly placed nodes are inves tigated For each density 50 different topologies are used The used propagation model provides around 15 of asym metrical links which means that the inbound and outbound PRR differs more than 0 15 In Figure 3 the results for a density of 24 is shown 180 Nodes are started at round 0 and 20 nodes where added after 1750 rounds The connectivity 26 N UI 100 7 Mahalle f mo N 79 50 connectivity number
201. measurements is one objec tive of SomSeD This network is intended to be a platform to integrate all results from research The TUHH campus is well suited to sustain a WSN due to its continuous area A permanent installation will support practical demonstra tions for various applications and technological concepts of WSNs This encourages the interest in SomSeD 2 INSTITUTE COOPERATION Developing WSNs demands contributions from all experts in the field of electrical engineering and computer science because all components in both hardware and software have to be aligned to each other While Quality of Service cri teria such as goodput and latency of classical networks can be loosened energy efficiency reliability in adverse condi tions low complexity and the size of single sensor nodes are the most important challenges Classical models such as the OSlI reference model are given up to allow a crosslayer design The application has a very strong influence on all layers of abstraction e g routing and media access control MAC must be considered together when energy efficiency shall be achieved 2 To fulfill these demands a strong cooperation between dif ferent faculties is necessary to define interfaces and redesign all components Although the individual nodes are very small the overall complexity of WSNs and the bandwidth of research topics is still high beginning with hardware aspects such as sensor design and autonomous e
202. might even be possible 3 4 Tracing and Simulation The ethernet backbone connection provides virtually unlim ited storage capacity which enables to trace network traffic over long periods of time As one of the important aspects of the WISEBED project the collected data will be used as input for simulative experiments and as comparative results Data gathered from real world experiments is vital to evalu ate and improve simulation models The testbed server runs the data gathering and management DGM service and will be responsible to configure the granularity of traces and data logging The mentioned DSL in section 2 1 supports the specification of desired trace data si 4 RELATED WORK The Deployment Support Network DSN 4 developed at the ETH Z rich consists of an additional wirless backbone network formed by DSN Nodes and allows monitoring and managing of a deployed WSN The wireless setup provides the possibility to attach a DSN to sensor networks in lo cations which lack a wired infrastructure A DSN Server provides a client interface to communicate with the DSN Each DSN Node has exactly one wired connection to one sensor node of the attached WSN The wired connection is realized using a hardware UART interface mounted on the DSN Nodes To avoid interference with the monitored WSN backbone traffic uses Bluetooth radios Since the DSN Nodes are battery powered and considering the high power consumption of Bluetooth rad
203. multiple groups and multicast trees requires memory and processing power which is limited on sensor nodes Also the default implementations of IP Mul ticast are designed to scale on large network groups with multiple receivers and senders In practical WSNs typically the amount of nodes is rather low Also the amount of ac tive trees and general management communication should be kept to a minimum Existing Overlay Multicast 3 so lutions such as Scribe Pastry CHORD Bayeux are nor mally not taking the wireless nature and limited capabilities of sensor nodes into account In general it is also a bad idea to have overlay connections established all the time which would lead to higher energy consumption and therefore re duces the lifetime of WSNs Several other issues concerning liveliness wireless communication and collisions exist Also reliability for a WSN multicast solution would also be de sirable because code updates and other critical tasks could then be solved efficiently using multicast 2 2 Designing Multicast in WSNs Multicasting in WSNs can be designed in different ways We will look at two approaches reliable IP Multicast and Over lay Multicast For both approaches we will look at source driven and receiver driven designs both centrally managed as well as de centrally organized Generally we will dis tinguish between two node types Branching nodes have to duplicate packets and store state information about re c
204. n Weltweit existiert jedoch eine gro e Anzahl verschiedener Hard und Softwarestandards f r den Umgang mit Sensorik und Sensordaten begr ndet in der breiten Menge von Sensortypen Nutzergemeinschaften und teilweise propriet ren Technologien und Softwareanwendungen Seit dem Jahr 2000 besch ftigt sich das Open Geospatial Consortium OGC im Rahmen der Sensor Web Enablement Initiative SWE mit der Erstellung von standardisierten Schnittstellenprotokollen und Datenmodellen siehe Tabelle 1 um alle Arten von verf gbaren Sensoren und Instrumenten aber auch Archive von Sensordaten ber das WWW auffindbar zugreifbar und wenn m glich auch kontrollierbar zu machen Botts 2007 Tabelle 1 Spezifikationen der OGC SWE Reihe Spezifikation Anwendung Observations amp Beschreibung von Measurements Scheme amp O amp M Sensorbeobachtungen Sensor Model Language Modellierung von Sensorik SensorML und Sensorprozessen Transducer Markup Modellierung von Language TML Messwertgebern Sensor Observations Service Lieferung von SOS Sensorbeobachtungsdaten Sensor Planning Service Anforderung und Planung von SPS nutzerbasierten Datenanfragen Sensor Alert Service SAS Web Notification Services WNS Ubermittlung sensorbezogener Meldungen Ziel der SWE Spezifikationen ist jedoch nicht der Ersatz notwendiger Spezialanwendungen zum Betrieb von Sensorik sondern deren Kapselung um den einheitlich
205. n nor com plete knowledge of wireless parameters can help an attacker to successfully attack the network As a result this work demon strates how the chaotic nature of radio communication which is often considered a disadvantage in regard to security objectives can be exploited to enhance protection and support implementation of lightweight security mechanisms Keywords Security Wireless sensor networks Radio propagation Authenti cation Implementation 1 INTRODUCTION Consisting of sensors for measuring temperature pressure air hu midity and other environmental conditions wireless sensor net works WSN provide means for various applications to increase comfort and security within private and public residences By ac cumulating data of many sensors a sophisticated system capable of triggering complex application specific tasks can promptly re act to environmental changes for example fire spreading water or gas leakage and other origins of catastrophes can be detected and countermeasures promptly initiated e g the room airing if gas leakage is detected closing the water valve to prevent further flooding or calling emergency in case of human injuries Hence residential monitoring is among the most important emerging ap plications for WSN However shifting such critical decisions to an automated system increases the importance of its security By abus ing the broadcast nature of wireless communication an attacker c
206. n Werkzeugen Weiterhin beschr nkt sich diese Arbeit nicht auf reine Simulatoren sondern greift mit TOSSIM auch auf einen Emulator und mit der theoretischen Ana lyse formale Ans tze auf 3 VERSUCHSBESCHREIBUNG Untersucht werden im Folgenden verschiedene Parame ter die sich in einem Netzwerk durch einfaches proba http www zeus bw fit de bilistisches Fluten einer Nachricht N bestimmen lassen der Energieverbrauch die Anzahl versendeter bzw emp fangener Nachrichten die Wahrscheinlichkeit dass ein zelne Knoten die Nachricht erhalten haben und beson dere Eigenschaften Messgr en die das betrachtete Pro gramm unterst tzt abh ngig von der Simulationsumge bung siehe Abschnitt 4 Die Topologien wurden so ge w hlt dass die Auswirkung von probabilistischem Flu ten auf symmetrische und unsymmetrische Sensoranord nungen Abbildung Ic untersucht werden kann In allen betrachteten Topologien bekommt Knoten A von der Basisstation die Nachricht N und leitet diese mit Wahrscheinlichkeit p 1 weiter Bei allen brigen Kno ten wird beim Empfang eines Paketes gew rfelt d h eine Zufallszahl 0 lt pz lt 1 wird gezogen und ent sprechend dem Ausgang pz lt p das Paket mit der Wahrscheinlichkeit p an alle Nachbarknoten weiterge leitet Um pr zise Daten und eine gute Vergleichbarkeit zu er halten soll eine m glichst geringe relative Standardab weichung in den einzelnen Ergebnissen erzielt werden Dazu s
207. n academic research In the described industrial scenar ios the wireless sensor networks are connected to the plant s control system 3 6 Number of Nodes Another objective addressed in current research on wireless sen sor networks is scalability to tens of thousands of sensor nodes For industrial applications wireless sensor networks however are likely to comprise only some tens to at most hundreds of sen sor nodes Industrial wireless sensor networks are usually quite Application area public spaces access restricted areas Self organization Number of nodes 10 000 It is necessary to carefully consider these differences when de signing a solution for industrial wireless sensor networks to de termine to what degree research results are suitable for envisaged industrial usage scenarios of wireless sensor networks small For example n condition based monitoring the network can be expected to have only some tens of sensor nodes at most It is also likely that a factory complex does not house one single large sensor network but instead several independent sensor net works in parallel The factory network control system takes care of distributing data to where it is needed Hence solutions are needed that are suitable for small to medium size sensor networks consisting of tens or hundreds of nodes 3 7 Summary Table 1 summarizes main differences between academic wireless sensor networks and industrial wireless sensor
208. n with Hardware Abstraction Operating System and Applications Compass Driver Terminal HAL Low level Drivers Resource Arbitration Power Management Hardware Platform Figure 1 System architecture without and with hardware abstraction layer 15 operations to ensure deadlock free operation Every access to a hardware resource such as reading a sensor or sending a packet over the radio is atomic and cannot be interrupted by another operation The resource arbitration architecture does power man agement of all hardware peripherals by switching off peri pherals when they are not used The architecture has know ledge of when devices can be powered off because it keeps track of device dependencies and all access to peripherals goes through the resource arbitration architecture 2 1 Hardware Devices versus Virtual Devices Our hardware abstraction layer can implement device drivers for devices that do not have a hardware equivalent This is useful for simulation or application development with out access to the target hardware We refer to all software devices that may or may not have their representation in hardware as virtual devices Virtual devices can be implemented using four techniques emulation virtualization or layering Emulation is used when no device of the desired class exists The device functionality is either emulated with an existing hardware device or simu lated in software Virtualization is used t
209. nce between ceiling and mobile client is not necessarily known 3 LOCALIZATION SYSTEM This section introduces the underlying localization system SNoW Bat It operates based on the thunderstorm principle where radio packets correspond to lightning and ultrasound signals to thunder According to the requirements from sec tion 2 all sensor nodes are divided into beacons and clients After deployment each beacon knows its position and is able to detect ultrasound signals whereas no client knows any co ordinates a priori However each client is able to generate ultrasound signals and to initiate the following localization procedure 1 If a client wants to know its current position it broad casts a radio packet containing its ID and the time difference between this radio message and the subse quent ultrasound chirp 2 Each anchor receiving this synchronizing radio packet activates its ultrasound receiver and thus expects the client s ultrasound signal after the specified waiting time If the anchors are adjusted to an exact grid each anchor can already determine its position if only one single anchor knows its position and if the anchors have additional knowl edge about their placement orientation etc 41 max movement space Figure 1 Schematic system design anchors are at the ceiling along a grid pattern with grid constant L clients are thereunder The ultrasound signal is emitted as spherical sec
210. nderen Herausfor derungen verbunden Die aus Kostengr nden vergleichwei se leistungsschwache Hardware und der begrenzte Energie vorrat auf Sensorknoten machen die Umsetzung von Sicher heitsmechanismen neben der eigentlichen Funktionalit t relativ schwierig Es existiert zwar eine Vielzahl von Publikationen zu Sicher heitsmechanismen in Sensornetzen h ufig wird jedoch nicht klar f r welchen Einsatzzweck die vorgeschlagenen Mecha Markus Ullmann Bundesamt f r Sicherheit in der Informationstechnik Bonn markus ullmann bsi bund de nismen konzipiert sind oder ob sich die technischen Voraus setzungen f r einen Mechanismus erf llen lassen Dies gilt umso mehr angesichts der Tatsache dass die unterschiedli chen Einsatzszenarios f r Sensornetze zu ganz unterschiedli chen technischen L sungen Realisierungen f hren Konkrete technische Merkmale k nnen jedoch dazu f hren dass be stimmte vorgeschlagene Sicherheitsmechanismen nicht mehr anwendbar sind Ziel eines geplanten Projekts des Bundesamts f r Sicherheit in der Informationstechnik ist es eine Bestandsaufnahme ber praxisrelevante Einsatzszenarios von drahtlosen Sen sornetzen zu gewinnen diese geeignet zu klassifizieren den Schutzbedarf f r unterschiedliche Anwendungsklassen fest zustellen und schlie lich geeignete Ma nahmen zur Umset zung der Sicherheitsziele zu empfehlen Im Januar 2008 hat das BSI hierzu ein Workshop organisiert an dem einige In
211. nding but efficient compression algorithms and sensor node capabilites and have thus designed algorithms specif ically adapted to sensor networks such as S LZW 8 and SBZIP 10 However Tsiftes et al have mainly regarded software updates that were compressed before deployment in 10 Software updates are deployed less frequently than data packets and even when a higher number of packets are required to transmit an application image the necessity to compress packet payloads still persists 4 CONCLUSION Limited energy budgets of sensor network nodes make data compression a necessity when energy savings are required However it is very likely that there is no universally optimal compression algorithm for sensor network data Instead the need to adapt to the payload contents and dynamically choose the appropriate compression algorithm that yields best compression rates whilst preserving energy is arising to keep the energy consumption low and thus enhance node lifetime Compression algorithms implemented for desktop computers are generally not suited for sensor nodes as they have high requirements towards CPU and memory size However a lot of existing approaches and concepts can be applied in sensors networks in modified forms We propose the use of a modular compression framework that can be dynamically updated during runtime while automatically detecting the locally optimal compression scheme for application specific data
212. ned with its long term reliability enables to determine if an intermediate quality link is tem porarily available for transmission Consequently adapting the neighbor table of a node and offering more forwarding choices to routing protocols 1 Introduction Instability in low power wireless sensor networks WSN connectivity has so far been regarded as a difficult problem that existing routing algorithms try their utmost to avoid In doing so they forego a large class of potentially valuable communication links An understanding of the patterns un derlying the seemingly irregular variations of the wireless channel would enable algorithms to make use of this previ ously disregarded class of links To achieve better connectivity and reliable packet com munication today s link estimators restrict communications only to the neighbors with constantly high quality links These high quality links are identified based on the long term success rate of a link typically collected over a time frame on the order of minutes However this approach has certain pitfalls First neighbors with intermittent connectiv ity might reach farther into the network Their use would therefore reduce the number of hops reduce energy usage in the network and increase its lifetime Second in a sparse network with a low density of nodes a node might have no high quality neighbor in its communication range requiring a mechanism to deal with unstable connectivity
213. nent code checking during the application development process Efficiency It is fast for everyday use 6 CONCLUSION AND FUTURE WORK In this paper we presented our technique and prototype implementation for automatic testing of sensor network ap plications before deployment It is important to fully test taking all possible inputs into account embedded appli cations such as sensor networks where the cost of occur ring undetected errors after deployment could be fatal We have demonstrated that it is possible to close the gap be tween the testing and development community by providing a user friendly automated bug finding tool which is strongly integrated in the system development life cycle We gave an overview of our system design and of the preliminary evalu ation results achieved Strenuous deployment requirements and resource constrained nature of sensor hardware e g inadequate power supply and limited computational power demand even more rigorous testing of sensor network applications Incorporating fur ther useful checks in Klee such as runtime monitoring of long loops and computationally intensive tasks by adding time annotations is future work Similarly verifying the dis tributed behavior of sensor network protocols such as cor rect state transitions remains to be addressed Moreover apart from TinyOS we will apply our solution to other sen sor network operating systems and development platforms
214. nergy supply and reaching to software methods like data gathering and visu alization The following institutes participating in SomSeD represent a unique aggregation of competence in WSN de velopment The Institute of Telematics has been active in research of WSNs since 2004 3 It is obvious that Telematics is the best choice to supervise the research field and to build the bridge between communication and information engineers The Institute of Telecommunications has its traditional re search activities in radar development and broadband com munications While the radar technique already performed 31 Routing Communication Networks Microsystems Technology Control Systems Measurement Technology Telecom munication Sensors Nanoelectronics Figure 1 Hardware Overview Integration Institute the transformation from large rotating antennas with huge transmit power into small and cheap automotive equipment the communications section is also gaining more and more interest in wireless sensor networks Sensors are a major component of WSNs Hence it is natu ral to include the Institute of Measurement Technology into the research field Besides a profound knowledge about var ious sensors the Institute offers a great experience in field testing Another crucial point is Networking The Institute of Com munication Networks plays an important role in developing suitable routing and networkin
215. network used in industry is administered and if problems arise mainte nance staff has access to every single sensor node 3 4 Self Organization Self organization is a big issue in academic research For indus trial applications of wireless sensor networks a reliable control lable and predictable operation is of high importance The plant engineers plant operator and maintenance staff need a way to plan and supervise the sensor network Self organization has to be designed so that reliable and predictable comprehensible opera tion is ensured Even a self organized industrial WSN needs to report its state to allow for maintenance if required In some situa tions self organized behavior is not feasible rather the system must be able to identify such situations and ask for external inter vention 3 5 Integration with Infrastructure Systems Industrial wireless sensor networks are likely to be integrated into other sometimes already existing systems Usually these sys tems base on standards such as a certain fieldbus protocols or Profinet which compared to office IP networks have differing requirements regarding determinism Often the higher level sys tems follow a cyclic polling approach which is in contrast to the event based model used in research networks Care has to be taken that the plant operator can access all necessary data at all times In contrast the integration with other networks is often not covered i
216. netzten Sensorknoten und den Beacons Hier wird der Lokalisierungsalgorithmus entweder auf allen Sensorknoten und oder zus tzlich auf den Beacons durchgef hrt Das bringt den Vorteil mit sich dass jeder Knoten seine Position mit dem geringsten Kommunikationsaufwand im Netz selbst bestimmt Im Gegensatz dazu senden im zweiten Fall Abbildung Ib alle Sensorknoten ihre Daten zur Senke z B einem Server PC im Netzwerk Die Senke bernimmt dabei alle Berechnungsschritte der komplexen Lokalisierungsalgorithmen Die Nachteile dieser Vorgehensweise sind der extrem hohe Kommunikationsaufwand im Netzwerk sowie die Anf lligkeit bei Unerreichbarkeit der Senke durch deren Ausfall oder Blockade der Weiterleitung Vorteil ist allerdings dass die Senke Teil einer Infrastruktur ist die die Bestimmung eines geod tischen Datums Ursprung Rotation Translation und Ma stab eines Referenzkoordinatensystems erm glicht Damit kann die Position eines einzelnen Sensorknotens in einem weltweiten Referenzsystem z B WGS84 wiedergegeben werden Beide Techniken haben den Nachteil dass sie entweder einen hohen Kommunikations oder Berechnungsaufwand mit sich bringen Im dritten dem g nstigsten Fall Abbildung Ic ergibt sich eine Mischtopologie in der der weniger komplexe Teil der Lokalisierung auf den Sensorknoten durchgef hrt und der energieaufw ndige Teil im Hinblick auf m glichst geringe Kommunikationskosten auf die Senke ausgelagert wird Dieser hy
217. ngenaue Eingangsgr en wie z B Distanzmessungen und die sehr eingeschr nkten und begrenzten Energie und Rechenressourcen eines jeden Knotens sowie die hoch miniaturisierte Hardware und geringe Batteriekapazit t erfordern die Entwicklung robuster energieeffizienter und pr ziser Lokalisierungsalgorithmen Dieses Kapitel gibt einen berblick ber die aktuellen Forschungen an der Universit t Rostock auf dem Gebiet der Lokalisierung in Geosensornetzwerken GSN und beschreibt Methoden zur 85 Standortbestimmung von Sensorknoten sowie der Verbesserung ihrer Position durch Ausrei erdetektion 3 1 1 Netzwerktopologien Der Lokalisierungsprozess in einem drahtlosen GSN basiert auf unterschiedlichen Netzwerktopologien Abbildung 1 welche im Folgenden kurz beschrieben werden sollen Allen gemein ist dabei dass einige wenige energie und rechenst rkere Knoten mit existierenden Positionierungssystemen Global Navigation Satellite System GNSS Global System for Mobile Communications GSM ausger stet sind Diese Knoten werden im Folgenden als Beacons bezeichnet Nachdem diese Beacons ihre Position ermittelt haben bestimmen die restlichen Knoten ihre Position z B durch Streckenmessungen eigenst ndig Abbildung 6 Ad hoc vernetzte Sensorknoten a mit Beacons b mit Beacons und einer festen Senke freies Netz c ohne Beacons und vier Senken Infrastrukturfall In dem ersten Fall Abbildung la besteht das Netz aus ad hoc ver
218. ngle aggregated data packet is forwarded by each node Although node distributions in real deployments might con siderably differ from this example it should be remarked that the number of packets cannot increase by applying data aggregation 2 2 Data Compression Opposed to the previously described mechanisms of data aggregation data compression targets to locally reduce the packet payload size by increasing the information density within a packet In general compression algorithms can be classified as either lossless or lossy While some applications such as software updates require a bitwise correct trans mission of data intrinsic offsets in sensor hardware might reduce the achievable accuracy and thus provide an applica tion area for lossy compression A variety of different approaches exist and will be briefly introduced in Section 3 1 While generic compression al 20 gorithms often achieve good average compression ratios on various types of input data algorithms adapted to specific applications often provide better performance however at the cost of being limited to a narrow application area A common metric to evaluate the savings achieved by data compression is the energy cost per bit Data compression is only feasible in terms of energy efficiency if the computa tional efforts to reduce the payload size by one bit are lower than the energy required for transmitting this additional bit Depending on the
219. nk to this new branching node Receivers that want to leave a group send a leave message towards the source Forwarders on the path update their status for that group and forward the leave message further Branching nodes receiving a leave message update their status remove the overlay link to the leaving node and discard the leave message If the branching node has just one overlay link left it has to change its status to a simple forwarding node for the remaining receiver and removes the affected overlay link Further it sends a notification towards the source and all intermediate nodes update their states accordingly They forwarding the message until it reaches a branching node which then establishes the overlay link to the remaining re ceiver To support end to end reliability in overlay multicast the receivers have to acknowledge the receipt of each multicast message or acknowledge the receipt accumulated after a se ries of messages Branching nodes aggregate and forward the acknowledgments In case of missing acknowledgments they send negative acknowledgments further towards the source Branching nodes also take care of retransmission of lost packets and therefore need to cache the multicast data up to a certain degree Hop to hop reliability is supported by underlaying protocols as described in Section 2 2 3 2 2 2 Reliable IP based Multicast Contrary to Overlay Multicast which uses TCP we do n
220. nks and the power consumption in the network If the delay from sources to sinks is to be minimized the number of hops between source and sinks must also be min imized The simplest solution consists of finding the shortest individual paths to each sink For each one of the sinks the neighbor which offers less hops to each sink is chosen as the next hop towards this sink This approach offers a mini mum delay although the number of intermediate nodes can be very high In case the power consumption is to be minimized the main strategy is to combine multiple transmission paths so that they are split as close to the sinks as possible in order to avoid the replication of packets Although the total number of relay nodes is minimized this can highly increase the transmission delay Both approaches explained above only take into account the tree created by a single source Nevertheless when there are more sources sending their readings the trees are created independently 3 2 Motivation The algorithm that we present takes into account several pa rameters in order to maintain a trade off between the energy consumption and the delay These parameters are motivated below If both metrics are to be optimized the number of hops must be close to the minimum Because when the number of in 90 termediate nodes increases the probability of losing a packet in such an unreliable medium increases accordingly If paths are split
221. nsor Networking Simu lator KSNSim wurde mit dem Ziel entwickelt Mecha nismen zur relationalen Anfrageverarbeitung in Sensor netzen schnell und ohne Spezifikation von Netzwerk Energie oder Betriebssystemparametern auf ihre Funk tionstiichtigkeit zu testen Ein weiteres Entwurfsziel bei der Entwicklung des Simulators ist die Schnittstellen Kompatibilit t mit dem auf den Sun SPOT 10 verwen deten Java SDK so dass entwickelte Programme oh ne Portierungsaufwand auch in einem realen Sensornetz aus Sun SPOT Knoten genutzt werden konnten Ziel war es also nicht drahtlose Kommunikation m glichst reali t tsgetreu nachzubilden oder besonders ausgefeilte Kom munikationsmechanismen bereitzustellen Das Netzwerkmodell zur Simulation von drahtloser Kom munikation ist daher so einfach wie m glich Sensoren werden an explizit definierten Punkten in einem virtuel len Raum platziert und haben eine Sendereichweichte r Als Kommunikationsmodell dient ein Unit Disk Graph mit Radius r Alle Knoten die einen Abstand gr er als r zum Absender einer Nachricht haben werden von der Kommunikation nicht beeinflusst Auf Grund der extrem einfachen Modellierung der Kom munikation ist es daher vor allem zu erwarten dass die Anzahl der empfangenen Nachrichten bersch tzt wird da keinerlei Paketverlust simuliert wird Diese erwarte te bersch tzung der empfangenen Nachrichten bewirkt eine bersch tzung der Anzahl erreichter Knoten und des En
222. nsor node The serial terminal interface also provided by the USB con nection will be used to monitor the state of the wireless sensor network The hosting mesh router sends commands to the sensor node in order to trigger a desired behavior in the context of an ongoing experiment or to retrieve the current information about its state As mentioned in sec tion 2 1 we implement these features with a combination of SSH remote command execution and the SNMP service An SNMP agent responsible for the sensor node will enable real time configuration and debugging Information regarding the sensor network state will be gathered periodically while additional requests can be issued on demand These sen sor node log files will be stored locally on the mesh routers The logged data is then transferred to the testbed server periodically as well as after the completion of experiments Therefore the testbed server functions as a central data stor age point for the sensor node log files With our management console hosted at the testbed server we will have a powerful tool to maintain and configure the locally distributed sensor nodes centrally in a very conve nient way minimizing the need of physical access Sensor nodes can be selected and configured on demand to satisfy the means of a given experiment Therefore it will also be possible to run concurrent experiments with a subset of sen sor nodes Further we extend t
223. nsorintervalle unterteilt die z B durch Luft oder Satellitenbilder gewonnen wurde Die gemessenen Werte werden dann mit den Positionen verglichen die mit durchgef hrten Trilaterationen erreicht auf die Footprintkarte abgebildet und somit den a priori definierten Sensorintervallen zugeordnet wurden Stimmen die Sensormesswerte mit den erwarteten berein werden die Beobachtungen zur weiteren Lokalisierung verwendet Im Falle einer Diskrepanz wird die Trilateration als Ausrei er verworfen Allerdings ist es auch m glich dass ein Sensorintervall auf mehrere r umlich getrennte Fl chen aufgeteilt wird um verschiedene Gebiete mit dem gleichen Sensorprofil auszustatten Diese getrennten Fl chen verhalten sich aber im Wesentlichen wie getrennte Sensorintervalle mit dem Unterschied dass Positionen die in einer Teilfl che liegen jedoch zur anderen geh ren nicht als Ausrei er erkannt werden 4 Sensor Web Ein Schwerpunkt bei der Untersuchung der Anwendungsbereiche von Geosensornetzwerken ist die Adaption des Grundkonzepts des Sensor Web Im Rahmen von Forschungsprojekten der NASA pr gte sich der Begriff Sensor Web f r die Zusammenarbeit heterogener und r umlich verteilter Sensorsysteme und Sensorplattformen zur Nutzung als Makro Instrument Delin 2002 Vor allem im Bereich der Beobachtung und Vorhersage von Naturkatastrophen zeichnet sich dadurch ein hohes Potential ab bestehende Systeme und Konzepte ma geblich zu verbesser
224. nt node is the best one the message will be dropped 2 2 2 Implementation In greedy routing each node is supposed to know the loca tions of the nodes in its vicinity In this work we focused on the so called beacon less greedy routing variant 5 However we implemented the following active neighbor selection vari ant As long as a node has not to forward anything it is not supposed to know its neighbors As soon as a message gets in the neighbors have to be determined first The node will broadcast a neighbor discovery message NDSC first Any neighbor node receiving an NDSC message will send a uni cast reply message NREP which includes the node s ID and location To avoid collision at the receiver an NREP will be sent out after a short random period between 0 and 16 msec In a static topology neighbor discovery needs to be per formed only once As soon as the first message was handled the other messages coming in can use the already determined set of one hop neighbor nodes In a highly dynamic scenario however this set will be outdated and a new neighbor dis covery gets necessary To support any topology properties between those two extreme cases we implemented a timeout based scheme where the neighbor set is invalidated period ically Thus message handling within one period requires 74 neighbor discovery only once The timeout period can be set as a protocol parameter The behavior of a greedy routing schem
225. nt records for the future research and automatic medication purposes In this paper we discuss the suitability of an emerging IEEE 802 15 4 protocol in the medical field especially its timeli ness and bandwidth related features We explore the limita tions of the protocol and conclude that the standard compli ant IEEE 802 15 4 protocol needs enhancements before it can be applied to urgent medical applications We propose an efficient solution to overcome those limitations and derive optimal parameter settings to enhance the IEEE 802 15 4 network performance in the medical field The remainder of the paper is organized as follows In Sec tion 2 we discuss IEEE 802 15 4 with its suitability to med ical applications and limitations concerning latency band width and energy issues Afterwards in Section 3 we elabo rate the state of the art related to our work Subsequently in Section 4 we define our system model and propose solution to those limitations We discuss performance evaluation in Section 5 and finally conclude the paper in Section 6 gt i 2 IEEE 802 15 4 OVERVIEW The IEEE 8021 5 4 protocol defines the PHY and MAC lay ers for LR WPAN It supports data rates of 20 40 100 and 250 Kbps within the range of around 10 meters and uses 16 channels in the 2 4 GHz band 30 channels in the 915 MHz band and 3 channels in the 868 MHz band It sup ports both star and peer to peer operation with fully and reduced function devic
226. nten oder Kranken hauspersonal schnell zu lokalisieren Notf lle rechtzeitig zu erkennen und dadurch die Versorgung von Kranken und Ver letzten zu verbessern Systeme zur Patientenlokalisierung sind bereits kommerziell verf gbar 1 In 7 wird eine kon krete Architektur f r ein derartiges System vorgestellt MICA2 Sensorknoten werden dort mit einem Pulsmesser und Oxi meter ausgestattet Die Vitaldaten werden entweder per Ad Hoc Routing an eine Basis oder direkt an PDAs die rzte mit sich f hren bertragen Fest installierte Lokalisierungs knoten erlauben die ungef hre Ortung eines Knotens Das Routingprotokoll muss den hohen Grad der Mobilit t der Knoten ber cksichtigen 2 3 Grenz berwachung Die Eignung von drahtlosen Sensornetzen f r Grenz ber wachungsszenarios wurde in unterschiedlichen Projekten un tersucht Im Rahmen des Projekts ExScal 2 wird ein ent sprechender Protoyp beschrieben der aus mehr als 1200 Sen sorknoten besteht Das System soll Personen und Fahrzeuge in einem 1 260 x 288 m gro en Areal zuverl ssig detektieren klassifizieren und nachverfolgen Zu diesem Zweck sind die Sensorknoten mit Passiv Infrarot Sensoren Magnetometern und Mikrofonen ausgestattet Die Gesamtarchitektur ent spricht einer dreischichtigen Hierarchie und besteht aus den eigentlichen Sensorknoten Schicht 1 Aggregationsknoten Schicht 2 und der Basisstation Schicht 3 Die Knoten werden per Hand in einem Raster in be
227. ntiert zen mit Hilfe des Closest Pattern Matching CPM Ver fahrens 8 CPM braucht als Eingabe sogenannte noi se traces die zum Teil mit TinyOS mitgeliefert wer den Die von uns verwendete mitgelieferte Datei bein haltet die noise traces der Meyer Library an der Stan ford University 11 Das Rauschen in dieser Umgebung isthoch wegen vieler WLAN Access Points Somit wird der Empfang der Packete durch den gain zwischen dem Sender und dem Empf nger die Interferenzen mit an deren Nachrichten und durch das Umgebungsrauschen bestimmt Da sich das obige Modell zum Empfang von Datenpake ten erheblich von den Modellen in allen anderen von uns verwendeten Simulatoren unterscheidet ist nicht klar wie und ob in TOSSIM die Modellparameter so gew hlt werden k nnen dass sie in etwa zum Beispiel denen von GloMoSim entsprechen Au erdem kann man die Sen dereichweite in TOSSIM im Gegensatz zu anderen Si mulatoren nicht ndern da diese Eigenschaft noch nicht implementiert wurde In unseren Experimenten lag die Sendereichweite zwischen 10 und 18 Metern 4 3 Formale Betrachtung Die theoretische Analyse wird mit Markov Ketten durch gef hrt wodurch sich die auftretenden Wahrscheinlich keiten mit hoher Pr zision ohne Standardabweichung berechnen lassen Die Modellierung und Evaluation er folgt mit Prism 6 f r jede Topologie getrennt Die be rechneten Gr en lassen sich als untere Schranke f r die zu messenden Werte verstehen
228. ntroduction of the participating institutes the de ployment of aWSN on the TUHH campus will be outlined 1 SOMSED Since the foundation of the Hamburg University of Technol ogy a unique organization structure has been applied On the one hand teaching is distinguished in classical deaneries Research on the other hand is performed over boundaries of faculties in distinct research fields Thus teaching and research are related to each other in a matrix organization structure One of these research fields is Self organized mobile Sensor and Data networks SomSeD Its benefits goals and chal lenges will be described in this paper Besides the coordina tion on the level of the university teachers the cooperation of undergraduate and Ph D students is encouraged Within SomSeD the so called Junior group has been created in which Ph D candidates design and develop a WSN The de gree of freedom is high all decisions regarding the system design are made by the participants themselves Thus the of Telecommunications Institute of Telematics Institute of Automation Institute of Control Systems Institute of Measurement Technology Institute for Security in Distributed Applications Institute of Communication Networks Institute of Microsystem Technology of Nanoelectronics launch of SomSeD Junior leads to a great motivation and an appreciable progress Deploying a WSN for long term
229. ntry is updated 3 3 Neighbor Expiration The neighbor expiration time has to be chosen carefully since it depends on the traffic pattern and the routing pro tocol Many protocols keep track of the surrounding nodes by periodically transmitting hello messages To minimize changes in the neighbor list if some nodes are temporarily unavailable the neighbor expiration time is set to four times the duration of the hello message interval of the routing pro tocol The value was empirically determined from a large set of scenarios 3 4 Observation Interval The observation interval has a large impact on the ap proach since its duration mainly affects the neighborhood change and thus the forwarding delay of the routing mes sages The duration has to be chosen in respect to the topol ogy change and link break rate Duration of twice the hello message interval of the routing protocol provided the best performance in the set of test scenarios It has to be kept in mind that the duration affects the reaction time of the approach to changes in the neighborhood 3 5 Forwarding Delay The time a node defers the forwarding of a routing mes sage is in the following referred to as forwarding delay The delay 6 is calculated from the neighborhood change metric e of the last observation interval Equation 2 is used to calculate the forwarding delay in the simulated scenarios in Section 4 B Ah 1 5 2 The quotient of the h
230. o increase the num ber of devices of a specific type A larger number can be pro vided by software multiplexing Multiplexing is done transpa rently to the application Layering is used when one or more hardware devices provide functionality that can be combined or extended to a more powerful high level device 2 1 1 Structural Design Elements Hardware devices are shared resources that usually cor respond to physical devices Each hardware device is represented by a unique device entity in software Different method implementations are used to access a hardware de vice This is useful for devices that can handle more than one protocol Each implementation can be used by multiple virtual devices to minimize platform specific code We use device entities to model inter device dependencies and to provide per device power control Dependencies are modeled as trees Each device entity keeps a list of devices it depends on The owner marks device entities when the device entities are in use Root entities are owned by virtual device entities and descendants are owned by their root We use virtual device hubs to implement device sharing Different virtual devices may operate on the same shared extends Virtual Device Class 0 1 active device Virtual 0 1 connected to 0 n Methods Events Device State Device Config Virtual Device Hub Device 0 1 default device device Each virtual
231. ocol for network programming at scale In J A Stankovic A Arora and R Govindan editors SenSys pages 81 94 ACM 2004 7 C J M Liang R Musaloiu Elefteri and A Terzis Typhoon A Reliable Data Dissemination Protocol for Wireless Sensor Networks In Verdone 11 pages 268 285 8 L Mottola Picco and A A Sheikh FiGaRo Fine Grained Software Reconfiguration for Wireless Sensor Networks In Verdone 11 pages 286 304 9 Texas Instruments Inc Dallas USA CC1100 Single Chip Low Cost Low Power RF Transceiver 2006 10 Texas Instruments Inc Dallas USA MSP430x161x Mixed Signal Microcontroller Aug 2006 11 Verdone editor Wireless Sensor Networks 5th European Conference EWSN 2008 Bologna Italy January 30 February 1 2008 Proceedings volume 4913 of Lecture Notes in Computer Science Springer 2008 12 D J Wheeler and R M Needham TEA a Tiny Encryption Algorithm In B Preneel editor FSE volume 1008 of Lecture Notes in Computer Science pages 363 366 Springer 1994 Positionierung und Sensor Web Enablement in Geosensornetzwerken Kai Walter Professur f r Geod sie und Geoinformatik Universit t Rostock Justus von Liebig Weg 6 18059 Rostock 49 381 498 3213 kai walter uni rostock de Keywords Drahtlose Sensornetzwerke Geosensornetzwerke Positionierung Sensor Web OGC Sensor Web Enablement Fr hwarnsystem Hangrutschung 1 Einleitung Der Begriff Digitale Erd
232. oday s low power nodes are not practicable Thus common keys increase the performance while significantly reducing the network load altogether with ener gy consumption Our current implementation focuses on usage during WSN deve lopment in research and education Thus we prefer performance to security and use either no encryption at all or the TEA algorithm 12 for a minimum of security Then the password is not necessa rily node specific but can depend on the application type versi on and build number This way and despite of a regular password change with each update it is still possible to supply several nodes simultaneously with new software At receiver side the high level task validates the packets regarding encryption and CRC checksum before executing control commands or buffering new images 2 3 Safety and Reliability Safety and reliability are other important factors in remote mainte nance systems This includes the guaranteed and complete recep tion of command and data packets at the desired destination no des Concerning the communication Ghost completely relies on the used network and routing protocol This accounts for compact ness and flexibility of the Ghost subsystem Just the data packets are successively numbered with each new transmission While communication is shifted to a WSN specific protocol Ghost integrates some mechanisms to guarantee smooth completion of the actual update process and to prevent the
233. of ongoing work of visualization and maintaining tools for wireless sensor networks WSN are briefly described An initial graphical user interface GUI will be introduced as well as the foreseen backbone architecture The current implemented GUI can be used to modify rules stored on individual nodes This rule editor can be displayed on several user displays of different client systems and the related concept is therefore called Multi Client Systems MCS Categories and Subject Descriptors H 4 3 Communications Applications General Terms Management Measurement Design Experimentation Human Factors Languages Reliability Keywords Wireless Sensor Networks Visualization Management Rule Management IEEE 802 15 4 Networks 1 INTRODUCTION In the recent years the networking of home appliances like PCs smart phones or multimedia devices and thus the access to WSN or home automation systems for instance climate or lighting control is evolving rapidly The data exchange between various devices of different vendors is not a problem nowadays Several standards have been published in the past years like IEEE 802 15 4 1 ZigBee 2 or ZWave 3 However the rich quantity of proprietary and standard protocols which cannot communicate to each other can decrease acceptance of potential users of WSN systems Especially users of a new system need often to study a user manual first Therefore a more or less autonomous architecture for
234. ohl in der Austausch und Skalierbarkeit der Dienstkomponenten als auch in der Entkopplung von anf lligkeitsbehafteten hierarchischen Informationsstrukturen solcher monolithischer Systeme zu sehen Die Anpassung des Konzepts an die Ziele der Sensor Web Initiative verspricht die Erschlie ung neuer Anwendungsdimensionen f r das GSN und erm glicht den dynamischen Import und Export von Dienst und Datenressourcen von und zu weiteren interoperablen Sensor und Informationssystemen 5 1 1 Echtzeitanforderungen Eine zentrale Problemstellung bei der Realisierung ist die Nutzung als Echtzeitbeobachtungssystem Klassische Echtzeitsysteme erfordern meist eine strikt bidirektionale Prozesskommunikation durch die nutzerbasierte Messbefehle zeitnah von der Sensorik ausgef hrt und Messergebnisse ber ereignisbasierte Push Kommunikation vermittelt werden k nnen Ein selbstorganisiertes ressourcenlimitiertes GSN ist gr tenteils unidirektional ausgelegt Knoten gt Datensenke und erlaubt nur eine begrenzte Fernsteuerbarkeit Zus tzlich handelt es sich bei SWE Diensten wie dem SOS der von zentraler Wichtigkeit f r die Verf gbarkeit von zeitkritischen Messdaten ist um einen passiven Server parameterbasierte Anfragen eines Clientsystems beantwortet Pull Kommunikation Die SWE Spezifikationsreihe bietet mit den Diensten SPS und SAS grundlegende Mittel um diese Probleme zu adressieren jedoch wird weiterhin die Entwicklung spezieller Gateway
235. ollows 1 The reference client is situated at a known position so called reference point and broadcasts a radio packet containing its ID its current position and the time dif ference between this radio message and the subsequent ultrasound chirp 2 Each beacon with yet unknown position receiving this synchronizing radio packet activates its ultrasound re ceiver and thus expects the client s ultrasound signal after the specified waiting time 42 3 The reference client transmits the ultrasound chirp as denoted in the previous radio broadcast 4 Each beacon which detects this ultrasound signal cal culates its distance towards the mobile client from the time difference of arrival between the radio and ultra sound signals If an anchor just receives the initiating radio message but no chirp it aborts measurement and waits again for radio packets of the reference client 5 If a beacon successfully completed at least four dis tance measurements to non collinear reference points it estimates its fixed position e g by using multilat eration From now on this beacon knows its position and is available for locating mobile clients with un known position as described in section 3 Obviously deployment this section and regular operation section 3 are very similar Their only differences are the knowledge of the nodes position and the handling of the distance information In comparison to the proce
236. ollten mindestens 200 Durchl ufe f r jede Konfi guration durchgef hrt werden 3 1 Topologiewahl Das Experiment wurde mit unterschiedlichen Topologi en durchgef hrt F r diese Arbeit wurden drei Topologi en Abb 1 ausgew hlt die deutliche Unterschiede be z glich Aufbau Symmetrie und Vernetzung zeigen Die Kanten innerhalb der Topologien repr sentieren den Konnektivit tsgraph Routing Baum also den Weg den die Nachricht N im Netz zur cklegen soll Die Distanz zwischen den Knoten f r die eine Verbindung Kante im Konnektivit tsgraph existiert soll h chstens 10 Me ter sein Die zu flutende Nachricht N betr gt exakt 56 Byte maximale Paketgr e in TinyOS Mit Beginn der Simulation befinden sich alle Knoten im eingeschalteten Zustand und warten auf Nachrichten Die Position der einzelnen Sensorknoten werden f r alle Topologien in einer Szenariobeschreibung definiert so dass diese allen Experimenten als Ausgangsbasis zur Verf gung stehen Neben den Topologien wird auch die Wahrscheinlich keit einer Weiterleitung variiert wobei 0 2 lt lt 1 in Schritten von je 0 2 simuliert wird Abh ngig von den M glichkeiten der verwendeten Werkzeuge werden wei tere Parameter spezifiziert Alle wichtigen Simulations parameter sind in Tabelle 1 zusammengefasst 97 98 N NUN IN oe VNVN 20 sy IAN an Ny Bu N a Topologie 1 b Topologie 3 Topologie 5 Figure 1 Verwendete Topolo
237. omprises the clients connected to a backbone router which may or may not provide a routing service themselves depending on the test run configuration The usual mesh client is regarded as mobile Tier 3 encompasses the sensor nodes which are connected by wire to a mesh router or client A subset of sensor nodes may be operated independently of the mesh network infrastructure with alternative power supply creat ing an even more dynamic topology 2 1 Testbed Server and Network Management The testbed server is connected by ethernet to the mesh routers and provides their root filesystem over the network using NFS Log files and evaluation data for mesh and sen sor network test runs are stored on the testbed server in a PostgreSQL database The testbed server hosts a man agement console implemented in the Java programming lan guage which provides the user interface to both networks The management console enables the configuration and main tenance of the network components realized with a combi nation of SSH remote command execution and the SNMP service SNMP agents running on the mesh routers offer a convenient way to configure and debug the network state The testbed server is also responsible of the task to sched ule experiments and their execution In order to have a uniform and easy to read way of defining reproducible ex periments we started to develop a domain specific language DSL based on XML The DSL comprises meta inform
238. on Grundlagen zuverl ssiger Kom munikation in drahtlosen Sensornetzen werden an den beteiligten Instituten Simulationen und Evaluierung mit verschiedensten Werkzeugen durchgef hrt Neben einer Netzwerksimulation mit GloMoSim kommen auch die Eigenentwicklung KSNSim sowie der Emulator des Ti nyOS Betriebsystems TOSSIM zum Einsatz Zus tzlich wurde in einer formalen theoretischen Analyse der Kon nektivit tsgraph verschiedener untersuchter Topologien durch Markovketten modelliert und mittels Prism 6 ana lysiert Um die Vergleichbarkeit der so auf unterschiedli che Weise gewonnenen Ergebnisse auch ber die einzel nen Teilprojekte hinaus zu gew hrleisten wurde ein ge meinsames Experiment durchgef hrt Der folgende Be richt dokumentiert dieses Experiment und stellt eine Aus wertung der Ergebnisse und der daraus gewonnenen kenntnisse dar Nach einem berblick ber verwandte Forschungsarbei ten in Kapitel 2 wird der Versuchsaufbau des durchge f hrten Experiments in Abschnitt 3 beschrieben Es folgt mit Abschnitt 4 eine Diskussion der Besonderheiten der einzelnen eingesetzten Werkzeuge Eine Analyse und Zu sammenfassung der Ergebnisse in Kapitel 5 schlie en diese Arbeit ab 2 VERWANDTE ARBEITEN Cavin et al haben 2002 bereits einige MANET Simula toren verglichen 4 Die dort verwendeten Simulatoren OPNET Modeler NS 2 und GloMoSim berschneiden sich jedoch nur in einem Fall mit den bei uns verwende te
239. onal Conference on Mobile Computing and Networking MobiCom 2000 B H Liu N Bulusu H Pham and S Jha CSMAC A Novel DS CDMA Based MAC Protocol for Wireless Sensor Networks In Proceedings of the 2004 IEEE Global Telecommunications Conference Workshops GLOBECOM 2004 J Polastre J Hill and D Culler Versatile Low Power Media Access for Wireless Sensor Networks In Proceedings of the 2nd International Conference on Embedded Networked Sensor Systems SenSys 2004 C M Sadler and M Martonosi Data Compression Algorithms for Energy Constrained Devices in Delay Tolerant Networks In Proceedings of the 4th International Conference on Embedded Networked Sensor Systems SenSys 2006 Sentilla Corp Tmote Sky Datasheet Online http www sentilla com moteiv endoflife html 2007 N Tsiftes A Dunkels and T Voigt Efficient Sensor Network Reprogramming through Compression of Executable Modules In Proceedings of the 5th Annual IEEE Communications Society Conference on Sensor Mesh and Ad Hoc Communications and Networks SECON 2008 W Ye J Heidemann and D Estrin An Energy Efficient MAC protocol for Wireless Sensor Networks Technical Report ISI TR 543 Information Science Institute ISI University of Southern California USC 2001 21 Connectivity aware Neighborhood Management Protocol in Wireless Sensor Networks Christoph Weyer Stefan Untersch tz and Volker Turau Institute of Telematics Hamburg Unive
240. one and recalibrate these anchors which were involved in deviations larger than an acceptable threshold 6 CONCLUSION AND FUTURE WORK In this paper we presented an efficient strategy for the de ployment of an ultrasound based localization system There fore some basic system requirements were defined E g the static anchor nodes must be mounted along a grid like pat tern above the observed mobile clients One benefit of our strategy is the usage of identical hardware for calibration during deployment as well as for localization during regular operation Just the roles of anchors and mobile clients will be reversed Since the anchors initially do not know their coordinates the basic idea of our approach is to use a reference node at well known positions for measuring some distances to these an chors which then calculate their static position The prob lem is to find a minimal set of reference points and to still 43 provide a sufficient number of distances for complete anchor coverage We showed that this problem is not trivial and does not only depend on the system accuracy and param eters grid constant etc but also on the reference node s positions during the calibration process So far we just analyzed a few different reference point lay outs and strategies The next step is to integrate the deploy ment procedure as described above into the existing localiza tion system SNoW Bat Analyzing systems with ar
241. onitoring or analyzing the network Application code that has been developed with the iSense framework is ready to run on any platform that supports an implementation of the isense API Primarily it is avail able for the iSense hardware platform as well as for the simulation environment Shawn Thus system designers and solution developers are able to test their implemented appli cations in the simulator before transferring them onto the iSense hardware That way time will be saved for the overall development process Thus there are three options given to practically evaluate implemented protocols or algorithms 1 Simulating with Shawn 2 Testing on real hardware with iSense 3 Testing on a common centralized sensor node testbed When implementing algorithms protocols or applications for either of these three options it is trivial to switch to one of the others afterwards This is because of the compatibility between Shawn and iSense This expands the possibilities for evaluating and validating and reduces the overall time of development processes in addition to all the other tech nologies features and advantages mentioned above 4 CONCLUSION We introduced the European project FRONTS The main objectives of the project are to get a fundamental under standing of the properties of wireless sensor networks to develop algorithms design rules for algorithms and applica tions and to form a unifying framework for
242. op on Sensor Network Engineering IWSNE 08 Santorini Island Greece Jun 2008 3 M Hosseini D T Ahmed S Shirmohammadi N D Georganas A Survey of Application Layer Multicast Protocols Communications Surveys amp Tutorials 9 3 58 74 2007 4 Lua J Crowcroft Pias Sharma S Lim A survey and comparison of peer to peer overlay network schemes Communications Surveys amp Tutorials 7 2 72 98 2005 5 A Okura Ihara and A Miura BAM Branch Aggregation Multicast for Wireless Sensor Networks IEEE International Conference on Mobile Adhoc and Sensor Systems Conference MASS 05 Washington DC USA Nov 2005 6 A Sheth B Shucker and Han VLM2 A Very Lightweight Mobile Multicast System For Wireless Sensor Networks JEEE Wireless Communications and Networking Conference WC NC 03 New Orleans LA USA Mar 2003 R Flury R Wattenhofer Routing Anycast and Multicast for Mesh and Sensor Networks EEE International Conference on Computer Communications INFOCOM 07 Anchorage Alaska USA May 2007 8 B Chen K Muniswamy Reddy and M Welsh Ad Hoc Multicast Routing on Resource Limited Sensor Nodes International Workshop on Multi hop Ad Hoc Networks REALMAN 06 Florence Italy May 2006 9 J S Silva T Camilo P Pinto R Ruivo A Rodrigues F Gaudncio F Boavida Multicast and IP Multicast Support in Wireless Sensor Networks Journal of Networks 3 3 1
243. operation is executed atomically and thus can produce a sig nificant overhead 3 Additionally the function is hardware depended and must be implemented for every platform Other event driven sensor node operating systems like SOS 4 and Retos 5 leave the power management to the programmer and do not implement any deepest sleep mode computation Contiki 6 can not take advantages from any sleep mode because of its polling methodology for interrupt handlers 3 Reflex REFLEX Real time Event FLow EXecutive is an oper ating system implemented in C targeting deeply embed ded systems and wireless sensor nodes It is based on the so called event flow model presented in 7 Activities are the schedulable entities which are triggered if something was posted to their associated event buffers Initial source of all activity in the system are interrupts Figure 1 shows a simple example containing the elements of a common REFLEX application The interrupt handler in the timer component writes data into the event buffer of the AD Converter component using an event channel which forces the activity of the converter to sample a value When the value is available the ADC interrupt arises and is han dled by writing the value to the event buffer of the applica tion logic component The associated activity implements the threshold logic which sends the data over the serial inter face component if necessary 4 Power Management in Reflex The powe
244. operations that may only be supported by some devices Functionality not supported by the hardware needs to be emulated in software A selection of these sets needs to be defined as methods of the virtual device Implementations for all these functions need to be polymorphic and implemented for each hardware platform Multiple virtual devices can use the same implementation Upon this interface a second set the API functions which is independent of the hardware platform can be defined This API can explicitly export virtual device methods or define high level functionality upon these 2 2 Behavioral Design We have shown the structural composition of our hard ware abstraction layer architecture To bring these structures to life and enforce the requirements implicitly given by the structural design we introduce a central actor for device management This actor takes part in all API calls to ensure proper device configuration and power management com pletely transparent to the application Fault tolerance is given by the fact that layers above the HAL do not need to care about switching power configuring busses and devices but can simply use them Using them in an unstructured way will generate more overhead but the HAL core will ensure that all virtual device operations are possible Based on our experience with numerous hardware plat forms and several operating systems for wireless sensor net works we designed the HAL core for non preem
245. orks In Proceedings of the Third European Workshop on Wireless Sensor Networks EWSN 2006 pages 212 227 February 2006 12 S L Shrestha J Lee A Lee Lee J Lee and S Chong An open wireless mesh testbed architecture with data collection and software distribution platform In J Lee editor Proc 8rd International Conference on Testbeds and Research Infrastructure for the Development of Networks and Communities TridentCom 2007 pages 1 10 2007 13 Zhang Sadler 5 A Lyon and Martonosi Hardware design experiences in zebranet In SenSys 04 Proceedings of the 2nd international conference on Embedded networked sensor systems pages 227 238 New York NY USA 2004 ACM Klassifikation von sicherheitsrelevanten Einsatzszenarios f r drahtlose Sensornetze Christian Wieschebrink Bundesamt f r Sicherheit in der Informationstechnik Bonn christian wieschebrink bsi bund de ABSTRACT Mit der wachsenden Bedeutung von drahtlosen Sensornet zen in der Praxis wird auch das Thema Sicherheit in Sen sornetzen wichtiger In der Vergangenheit wurden eine F lle von Sicherheitsmechanismen f r drahtlose Sensornetze vor geschlagen Die Umsetzung von solchen Sicherheitsmecha nismen in der Praxis ist aufgrund abweichender technischer Voraussetzungen h ufig mit Schwierigkeiten verbunden Ein geplantes BSI Projekt soll wesentliche Einsatzszenarios f r Sensornetze identifizieren und Sicherheitsmechanismen a
246. ot have a reliable end to end transport protocol Instead we are using UDP End to end reliability is realized using acknowl edgment messages as described above Branching nodes know only that their one hop neighbors are forwarding the packets on their behalf Acknowledgments be handled on one neighbor basis hop to hop and not between branch ing nodes as for Overlay Multicast For the source driven de centralized approach the source sends the join list to its direct neighbors that should act as forwarders The next forwarder is determined on a hop to hop basis If a node has to become a branching node it re members from which neighbors it expects acknowledgments Joins and leaves are handled by appropriate messages that could cause forwarders to become branching nodes and vice versa triggering a modification of the expected acknowledg ments state for a node In the centralized source driven approach the source sends the list of all branching nodes to the closest branching node which processes and forwards them further to the nearest branching nodes on the path Intermediate nodes become forwarders and store the status for the involved multicast group Acknowledgments are handled directly between the branching nodes Additional joining nodes trigger an update of the affected branching nodes all initiated directly by the source Leaves are handled accordingly triggering updates of the branching and forwarding nodes involved In the rece
247. ould easily impersonate sensor nodes and inject fake sensor data into a WSN without even being physically present inside the res idence Continuous aggregation of fake data may lead the system to wrong decisions and provide vectors for more sophisticated at tacks e g sending fake packets to emulate gas leakage and initiate room airing may be exploited for physical intrusion Traditionally security in computer networks has multiple objec tives and attempts to equally satisfy authentication confidentiality and integrity goals of transmitted data When following such one size fits all approach the cost of key exchange identity verifica Reflection N a _ Diffraction 1 mm EEE mm attenuated Receiver Figure 1 Indoor radio propagation and different propagation phenomena reflection diffraction and scattering result in un predictability of signal propagation within real world environ ments tion and data encryption puts high demands on protocol complex ity its implementation and computational power While such re quirements can usually be met by the relatively high performance devices in common computer networks in a WSN the security costs become more tangible through decreased battery life and var ious vulnerabilities from depletion of computation or memory re sources Operating within an indoor environment a WSN may take advan tage of access cont
248. ource arbitration architectures 3 7 our architecture is deadlock free by design As illustrated in Figure 1 introduction of a hardware ab straction layer HAL increases portability of operating system and application code through common well defined device interfaces High level device drivers can reuse low level driv ers and do not need to provide their own implementation To further reduce operating system complexity we integrate resource arbitration and power management into the HAL as part of the operating system Most operating systems for wire less sensor nodes provide no hardware abstraction at all Even those who do require cooperative interaction of user applica tions which need to take care of resource arbitration This increases development overhead and vulnerability to uncoo perative or erroneous applications Energy is a scarce resource on wireless devices To reduce the energy consumption every circuit and peripheral has to be powered off when they are not used An optimal power saving strategy can only be implemented when the dependen Adam Dunkels Swedish Institute of Computer Science Box 1263 SE 164 29 Kista Sweden adam sics se cies between devices are known to switch off a shared bus all devices that use the bus must be switched off first Our archi tecture explicitly encodes device dependencies to determine when to switch off devices and shared busses 2 A DEADLOCK FREE RESOURCE ARBITRATION AR
249. overhearing first few packets can identify intermediate quality links tem porarily available for transmission For example consider a sensor network based fence mon itoring system 19 During normal conditions 1 when there is no intruder breaking into the fence the network will generate very limited or no traffic However as soon as an in truder is detected by the system large bursts of packets will be generated by the distributed event detection algorithm In this situation STLE recognizes intermediate quality links currently stable for transmission and informs routing algo rithm of the availability of such links by online adaptation of neighbor tables We believe that this technique will signifi cantly reduce the hop count a packet has to traverse from its source to destination Thereby minimizing the energy con sumption and increasing network life time 4 Evaluation To evaluate the concept of short term link estimation we executed a number of experiments in our indoor testbed The testbed consists of a regular 6 x 6 grid of Telos B motes 12 with a spacing of approximately 2 80 m inside a 20m x 20m indoor auditorium Every node transmitted a burst of 20 se quentially numbered packets with a length of 15 bytes at 25 dBm We ran this experiment for 5 500 seconds To calibrate STLE we need to identify a threshold when an intermediate or bad link should be considered temporary reliable Figure 2 a depicts the probabili
250. participants have a great deal of experiences with the fundamental the ory of wireless sensor networks Together we will contribute research to the following issues e Development of validated mathematical models for large scale wireless sensor networks beyond oversimplifica tions such as Unit Disk Graphs infinite capacity links random waypoints movements e Design of dynamic and adaptable network infrastruc tures including enabling security by developing appro priate algorithms protocols and strategies e Design of algorithms for solving global tasks like collec tive exploration of unknown areas or target tracking in dynamically changing environments The paper is structured as follows The project FRONTS itself and its aims are described in more detail in the follow ing Section 2 The IBR at the Braunschweig University of Technology and the ITM at the University of Liibeck par ticipate in FRONTS The practical parts of the project are our responsibility It is our aim to get valuable feedback from implemented algorithms protocols and applications that are developed within the project Feedback will be 101 achieved by evaluations and validations Therefore we will build up the Experiments Repository which will consist of an appropriate simulation package for wireless sensor net works as well as at least one physical testbed It will be remotely accessible via the Internet for all project partners The requir
251. pplications For home automation and industrial control purposes the ZigBee Alliance 2 and the HART Communication Foundation 1 respectively specified higher layer protocols based on the 802 15 4 PHY and MAC These protocols enhanced 802 15 4 and made it very popu lar for commerical applications with limited battery power and small throughput requirements In the academic community the interest for 802 15 4 has also grown and numerous studies focused on the perfor mance evaluation of the WPAN standard made its benefits and deficiencies well known If however 802 15 4 WNS run ning on batteries shall become reality one major problem re sides If the network is running in the beacon mode a central coordinator broadcasts beacons i e special command mes sages for synchronization purposes to all nodes in the net work This enables all devices to operate with superframes consisting of an active and a passive phase where devices can be put to sleep mode In the absence of a central coor dinator this fixed structure is not existing in the nonbeacon mode which enables larger and more flexible topologies and is hence a good choice for 802 15 4 WSNs If the network shall enable a distributed routing solution no energy saving options for the nonbeacon mode are given in 5 In earlier works we therefore investigated the effects of a simple sleep scheduling solution where the sensor nodes duty cycle at a regular schedule and loosely synchron
252. pproach is to send only fractions of the PRR values in a round robin schedule Doing so all packets are equally sized and every PRR value is broadcasted with a fixed schedule Mahalle uses the last approach to exchange link information between neighboring nodes One additional benefit of this is that additional information can be included into the broadcast packets e g the node state needed by self stabilizing algorithms 3 The node table consists of entries to store the neighbor list and the preparation list neighbor list contains the nodes that are treated as neighbors All other entries in the node table are used by the preparation list where po tential neighbors are stored in order to be able to estimate their link quality At startup the whole node table is occu pied by the preparation list If a node is evicted from the node table it is inserted into a blacklist for some time in order to prevent them to reenter immediately The num ber of entries in the node table N and the ratio between the maximum number of entries in the neighbor list Nn and the preparation list Np are compile time parameters of Mahalle and correlate to the expected average network density In every round each node broadcasts a neighborhood packet This packet contains the identifier of the node a sequence number and one entry of the node table This information contains of the identifier of the neighbor and the estimated PRR T
253. provement for the whole network owing to the sim plistic example To demonstrate the potential for overall improvement we will now show results from larger network topologies 3 RESULTS We compared the results of LFF vs arbitrary multi plexing by implementing the scheduling discipline in the DISCO Network Calculator tool 4 5 using the concepts from above For scenario generation we used the generator as used for 6 to generate random sink tree networks with a single sink a tree depth of 5 levels with each node having one to five child nodes Arrival and service curves for each node are a 70 01KB s ikB and B 215K B s 0 5s We then calculate 46 60 1 Arbitrary LFF 50 40 Frequency 30 20 10 ad 7 8 8 2 8 8 9 2 9 8 10 11 11 12 Delay bound s Figure 3 Frequency of delay bounds in an example network Maximum delay s 6 O Arbitrary aH E LFF Number of hops Figure 4 Comparison of the maximum delay bounds per flow length in an example network the delay bound for each flow using the Pay Multiplexing Only Once Analysis PMOOA 7 8 both under arbitrary and LFF multiplexing An example of the change in the delay bounds result ing from LFF scheduling is shown in Figure 3 Many flows have an upper delay bound of over 11s with a maximum of 11 528 while LFF scheduling results in a maximum delay bound of 10 425 This makes for a 9 5 lo
254. ption in the network only rises slightly the network lifetime is max imized Our future work includes further experimentation with sev eral weights as well as the inclusion of some other param eters in the cost function such as the number of collisions registered by the MAC layer 6 REFERENCES 1 Pietro Ciciriello Luca Mottola and Gian Pietro Picco Efficient Routing from Multiple Sources to Multiple Sinks in Wireless Sensor Networks In Proceedings of the 4th European Conference on Wireless Sensor Networks EWSNO07 Delft The Netherlands January 2007 2 Feilong Tang Minyi Guo Minglu Li Yanqin Yang Daqiang Zhang Yi Wang Wireless Mesh Sensor Networks in Pervasive Environment a Reliable Architecture and Routing Protocol In Proceedings of ICPP Workshops 2007 92 3 Anna F rster and Amy L Murphy FROMS Feedback Routing for Optimizing Multiple Sinks in WSN with Reinforcement Learning 3rd International Conference on Intelligent Sensors Sensor Networks and Information Processing ISSNIP Melbourne 3 6 Dec 07 C Intanagonwiwat R Govindan D Estrin J Heidemann and F Silva Directed Diffusion for Wireless Sensor Networking IEEE Trans on Networking February 2003 Samuel R Madden Michael J Franklin Joseph M Hellerstein and Wei Hong TAG a Tiny AGgregation Service for Ad Hoc Sensor Networks OSDI December 2002 Shun Yu Chuang Chien Chen Chang Jie Jiang Minimum delay energy effic
255. ptive systems although the structural architecture can support these as well Small energy efficient microcontrollers used in sensor net works do not provide virtual memory controllers for true preemption The overhead necessary for fault tolerant dead lock handling and device access scheduling can be saved For energy efficiency operating systems are built event based and driven by hardware interrupts Data processing can be syn chronized by means of a processing loop Interrupt code is executable from lower power modes and kept short to allow for other interrupts A processing loop is running in full power mode and switches the system to low power when processing is done In this programming model interrupts are used for notification e g timers completion of an operation or for receiving data serial connection and to trigger actions which are executed outside the interrupt Only for strict real time requirements it may be necessary to perform actions inside of interrupts Our architecture can handle all processing transpa rently when used from outside interrupts Some extra care is needed for special cases where interrupt code must access virtual devices 2 2 1 Defining Atomic Operations We implement the basic set of operations on a virtual de vice as atomic Most of them directly access the underlying hardware Depending on the specific hardware and its proto col interrupting an atomic operation and reconfiguration ofa shared reso
256. r and are able to store the position informa tion These algorithms are often based on hidden markov models 3 or on large databases 6 Another strategy to reduce the number of topology chan ges is to select a forwarding link according to different met 77 rics like reliability 7 node speed or link duration 8 It has to be kept in mind that the most reliable or stable path does not necessarily result in the shortest possible path in respect to the number of hops The method that is presented in this paper prefers slow moving nodes over fast moving nodes It is inspired by sev eral observations made by 8 and 9 which analyze differ ent mobility models The papers show that the link dura tion and the number of topology changes strongly depend on the used mobility model and its configuration Further more the presented node speed histograms reveal that the link duration is correlated to the speed of the nodes Thus slow moving nodes should be selected as forwarding nodes since their links tend to be more stable than the links of fast moving nodes Section 3 introduces the functionality of the approach and shows its capability to take advantage from the fact that the node speed is not uniform distributed in many scenarios due to restricted path selection 3 DELAY BASED APPROACH The end to end reliability in wireless networks strongly depends on the mobility of the network The faster the movement speeds of
257. r different tasks such as event detection localization tracking and monitoring Such applications should be configured and updated during the life time of the sensor nodes and over the network 1 An update with many unicast connections to the nodes is very inefficient and consumes resources such as bandwidth and energy Thus it is obvious that multicast communication the management of WSNs may benefit by reducing the number of transmit ted packets and by saving energy To access WSNs via the Internet a IP based communication is required 2 Thus multicast communication should be IP based as well 1 2 Multicast Multicast is an efficient way to disseminate data to a group of receivers that are interested in the same content Con trary to unicast where the sender has to transmit the data This work has been supported by the Hasler Foundation under grant number ManCom 2060 for each receiver individually multicast requires the sender to transmit the data only once Thereafter the network or other hosts interested in the data will replicate when re quired and forward the data to the receiving group members In the Internet the multicast paradigm has been implement in the form of IP Multicast Interested receivers send an IGMP group join message the routers process these mes sages according the IP Multicast protocol used RSVP PIM etc and build the distribution tree among them A sender now only sends a UDP Multicast packet
258. r management in REFLEX has two views a user view and a system view The user view allows the programmer do define modes of operation utilizing differ ent parts of the hardware and even software components This makes it possible to divide the execution of an appli cation into different phases The system provides the class EnergyManageAble from which every part of the system 37 AD Converter handle C gt interrupt handler Application logic activity Q threshold Packet control reception C component Serial interface O event buffer modul output tx lt gt interrupt E am gt event channel Figure 1 REFLEX application example association that should be included into power management must be de rived Manageable objects can be interrupt handlers as well as event channels The programmer is responsible for chang ing the modes E g a timer driven module can be used for mode changes At startup each manageable object is registered with the power management and assigned to one or more groups During operation groups can be independently activated and deactivated If a manageable object is member of multiple groups it is only deactivated when all of these groups are deactivated The system view is used for the determination of the deep est possible sleep mode Every instance of a component has a variable which specifies the deepest possible sleep mode that may be used when it is active
259. r such a system and which minimal number of calibration steps are required during deployment 1 INTRODUCTION Lots of application scenarios for wireless sensor networks WSNs are in need of accurate knowledge about current positions of several objects Because this can be such an im portant basic requirement for the system s successful opera tion not only different approaches were proposed but also based on these solutions various localization systems were built and successfully proven c f 1 4 5 6 7 10 11 12 Indeed the sometimes costly deployment process of such a WSN was hardly analyzed so far Yet deployment can be very time consuming and cost intensive especially when numerous beacons have to be mounted and calibrated quite precisely That s why this paper presents a self deployment strategy for beacon based localization systems to save time and costs Therefore section 2 defines some basic require ments for the self deployment procedure whereas section 3 introduces the localization system SNoW Bat 4 to form the basis of our further research Section 4 specifies our de ployment approach Tab WoNS which is quite similar to the localization stage of SNoW Bat Our thoughts and results for the calibration during deployment stage are presented in section 5 and finally a conclusion in section 6 closes this paper 2 SYSTEM REQUIREMENTS The self deployment strategy presented in this paper re quires the desi
260. ration of this critical process Section 4 shows the ener gy requirements of a concrete Ghost implementation Yet underestimated energy resources as well as other unpredictable problems can still lead to node reset or hanging during the update Thus the Ghost low level functions configure an available watch dog timer to force an reset in case of a hanging node A reset always leads to a well defined system state since the Ghost low level part will never be erased from the ROM In particular it contains an emergency function Fig 4 which is automatically executed at node start up to recover from awkward situations First it tests a flag indicating the state of the last update On success the regular application is started and the operating system takes over control On failure the power supply is again checked repeatedly until there is sufficient energy to restart the update 3 IMPLEMENTATION As useful and indispensable remote maintenance is for large scale or frequently updated sensor networks as complex are the resulting requirements and challenges for a concrete implementation There by some problems are rooted in the underlying network hardware and protocols in the node s architecture and in the applied opera ting system Our implementation of Ghost uses the SNOWS 2 5 sensor node and the SmartOS 3 operating system Within the test application presented below we used the SmartNet wireless com munication protocol
261. rcuits to safe power 3 RELATED WORK A fully grown operation system as Linux already provides broad set of services including device identification enumera tion and hot plugging 3 It needs to use explicit locking to support hardware multi tasking and preemption and still has to be tolerant against faulty applications Its hardware ab straction and driver model is too large to focus on the energy and efficiency aspects that are crucial to small embedded devices The TinyOS 2 0 operating system provides both resource arbitration and hardware abstraction suitable for sensor nodes 4 5 The combination of both roughly covers the same endpoints as our solution Using the nesC language two layers of device interfaces are defined 6 A Hardware Presen tation Layer HPL provides a device specific set of functions that encapsulates the immediate hardware access such as register names On top of the HPL a HAL is defined as a device specific abstraction Finally a high level Hardware Interface Layer HIL provides a device independent interface The abstraction level of the HIL corresponds to our virtual device methods abstraction While defining interfaces using nesC which cannot be ex pressed in ANSI C helps structuring the API by adding seman tics it does not provide additional functionality The addition al coding layer with its compiler allows semantic checks dur 18 ing compilation that are not performed without However th
262. re work 5 REFERENCES 1 R L Cruz A Calculus for Network Delay Part I II IEEE Transactions on Information Theory 37 1 114 141 1991 2 Jens Schmitt and Utz Roedig Sensor Network Calculus A Framework for Worst Case Analysis In Proceedings of IEEE ACM International Conference on Distributed Computing in Sensor Systems DCOSS 05 Marina del Rey USA pages 141 154 Springer LNCS 3560 June 2005 ISBN 3 540 26422 1 3 Jens B Schmitt Frank A Zdarsky and Markus Fidler Delay bounds under arbitrary multiplexing When network calculus leaves you in the lurch In 27th IEEE International Conference on Computer Communications INFOCOM 2008 Phoenix AZ USA April 2008 4 Jens Schmitt and Frank A Zdarsky The DISCO Network Calculator A Toolbox for Worst Case Analysis In Proceedings of the First International Conference on Performance Evaluation Methodologies and Tools VALUETOOLS 06 Pisa Italy ACM November 2006 5 Nicos Gollan Frank A Zdarsky Ivan Martinovic and Jens B Schmitt The disco network calculator demonstration In 14th GI ITG Conference on Measurement Modeling and Evaluation of Computer and Communication Systems MMB 2008 Dortmund Germany March 2008 GI ITG 6 Jens B Schmitt Frank A Zdarsky and Lothar Thiele A Comprehensive Worst Case Calculus for Wireless Sensor Networks with In Network Processing 47 In IEEE Real Time Systems Symposium RTSS 07 Tucson AZ U
263. re engineered A well planned network is cost efficient reliable and predictable Hence a planned deployment of industrial sensor networks is often a must have In the scenarios described above sensor nodes are carefully placed at distinct locations e g in vicinity of a machine near the expected source of mass loss at a point of measurement However current research often as sumes a random deployment of sensor nodes e g sensor nodes are assumed to be air dropped While this assumption is true for applications such as disaster management military intelligence and environmental monitoring a random deployment cannot be expected e g for process and factory automation 3 2 Infrastructure Most sensor network research assumes that the sensor network is decentralized and uses no infrastructure for its operation besides a central data sink In an industrial environment often infrastruc ture components as e g gateways or range extenders are available that can be used for controlling and optimizing the operation of the sensor network e g to minimize communication path length to minimize energy use of battery powered devices to support security In fact as wireless sensor networks used in industrial applications are planned networks the infrastructure components can be placed in an optimized way In all three scenarios de scribed above the function of the sensor network includes report ing to the plant operator via a gatew
264. receive image data and to write it into the external memory The module goes to sleep again if the host sends a sleep packet or if a timeout occurs In the following the AirFlash protocol that handles the image transfer between host and node will be explained First the host transmits a wakeup packet in order to put the node into the Wide awake state The awoken node answers to the host Then the transmission of the image file can start This process is depicted in Figure 4 Since the image is up to 120kB in size it must be fragmented Each fragment contains 256 Bytes which is also the size of a page in the memory Each AirFlash packet see Figure 5 contains 16 bytes of image data a so called subfragment Thus to transmit one fragment of 256 bytes 17 AirFlash packets necessary where the last packet contains a CRC checksum to verify the fragment Page complete Write page buffer to external memory Receive Packet Write packet to page buffer Success Page incomplete Send Figure 4 Image transfer from the node s perspective Storage page bias Data CRC16 Figure 5 AirFlash packet format All received subfragments are buffered in the RAM of the node The CRC from the 17 packet is then used to verify the whole fragment If verification fails a NACK packet is sent to the host and the whole fragment is retransmitted If verification
265. red localization system to comply with the following requirements First of all the sensor nodes of the localization system are classified as beacons i e anchor nodes with known and constant position and clients i e mobile nodes with unknown and variable position To de rive basic conditions for efficient calibration during the self deployment stage our approach presumes that the beacons Marcel Baunach Chair of Computer Science V Am Hubland Wurzburg Germany baunach informatik uni wuerzburg de are aligned in a regular pattern For this paper a roughly grid like layout is analyzed The underlying localization system uses ultrasound signals so called chirps for distance measurement But it is suffi cient even for deployment stage if beacons are just able to receive ultrasound signals whereas only clients are ca pable of transmitting ultrasound for distance measurement There will be no need for the anchors to generate ultrasound signals which saves energy and reduces environmental pol lution by ultrasound emitted from the quite large number of beacons in such systems Besides we are just interested in three dimensional localization the two dimensional equiva lent would be much easier but less realistic However for simplification we will assume that the anchors are fixed at the common ceiling plane and a mobile client moves freely thereunder within a certain distance range cf Fig 1 Ad ditionally the dista
266. rk Calculus 1 2 a short introduction can be found in the appendix In that case the whole network design depends on the nodes experiencing the worst service which may lead to a vastly overdimensioned network As was shown in earlier work 3 the worst case behavior of a network may deteriorate rapidly with the network s diame ter This can be avoided by giving priority to certain flows in a way that balances delay times for all nodes reducing the variance of that metric However this requires keeping and analyzing the state of all flows In environments with restricted resources like sensor networks this may impose a significant additional load to the network nodes reducing the network s lifetime and utility thus a balancing scheme needs to be as simple as possible Simplicity is especially 03 OG On flow MORO 3 interest Figure 1 Typical characteristics of a sink tree net work with feed forward characteristics As a flow of interest traverses servers on the way to the sink it gets multiplexed with other flows and remains so until reaching the sink called for when trying to improve a worst case bound since especially with low network load that bound may be en countered only rarely if ever So spending a large amount of resources would be a waste in the average case 2 LONGEST FLOW FIRST SCHEDULING The proposed multiplexing scheme is very simple When flows are multiplexed the flow that has pa
267. roach the source sends the list of all receivers of the multicast data to its one hop neighbors The neighbor nodes then check if all receivers in the list can be reached through a single of their next hop neighbors In this case such a node just forwards the list to its next hop and remembers that it acts as a forwarder for that multicast group If nodes from the list can be reached via different neighbors the list is split accordingly and the partial lists with the node s own address as source of the message are forwarded to the respective neighbors The node splitting the list becomes a branching node and opens a TCP connection for the overlay link to the sender of the original list when the corresponding tree is activated Fi nally if such a list message arrives at the receiver group member then that node prepares the overlay link to the source of the message normally the last branching point Therefore TCP connections for the overlay network links are only established between the source branching nodes and the receivers see also Figure 1 when the correspond ing multicast tree becomes active New nodes are added to the multicast tree by sending a list message including the new nodes as described before Only when a tree is inac tive new nodes can be added Therefore no connections are established directly but potential new overlay links are just prepared and only established upon activation of the tree If a branching node
268. rocessed Finally after processing the last event TinyOS scheduler is stopped Figure 1 shows an overview of KleeNet s build process First a user can optionally specify in a configuration file which variables in his code should be marked as symbolic For this purpose we parse the C code after NesC compilation and insert calls to klee_make_symbolic function Second all incoming packet buffers are also marked symbolic automat ically Third the instrumented code is then passed to Klee which builds the C object file Finally Klee interprets this object file and terminates when no bug is detected Other wise a test case with real input values is automatically gen erated Running this concrete test case with the unmodified version of the code will cause the deployed sensor network application follow the same path and hit the same bug Please note that as we apply code instrumentation to C source code therefore this process is neither bound to a certain hardware platform nor to TinyOS and NesC Hence Virtual platform make kleenet Annotaded C Code Object file Compilation Termination Bug detected make kleenet test test case Figure 1 Integration of KleeNet into TinyOS plat form our approach can easily be integrated into any other sensor network development platform and operating system 5 EVALUATION We first checked the BlinkFail application from the TinyOS source repository It is used to
269. rogrammer only needs to specify which memory loca tions in his code are input derived e g an incoming network packet During code execution all paths and operations on symbolic variables are tracked If a bug is detected Klee au tomatically generates a test case with concrete values caus ing the bug For example consider a simple application code in listing 1 and the associated KleeNet output in listing 2 At the current state of its implementation Klee reports only memory reference and division by zero errors It has been developed to be scalable and to support all sorts of unsafe type operations including pointer casts and pointer arith metics call Timerl startPeriodic 500 int 10 unsigned i klee_make_symbolic_name amp i sizeof i i event void 1 fired call Leds led1Toggle here we violate memory safety on the 11th signal of this event if i lt 11 a i 1 Listing 1 Array index out of bounds bug make kleenet test KLEE ERROR memory error out of bound pointer make kleenet display BlinkFailC i 10 Listing 2 KleeNet detects the memory error and generates a concrete test case 3 2 Automatic Code Instrumentation As discussed earlier one of the major limitations associated with most software testing tools is the lack of user friendly interfaces Most of the available tools are either not properly integrated into software development process or the
270. rol offered by the physical world where walls locked doors or other barriers for preventing physical intrusion es tablish the border between an outdoor attacker and the legitimate sensors Although radio signals are able to trespass such barriers the radio signal received from outside is strongly biased by various propagation effects such as reflection absorption diffusion and scattering and therefore tainted with a significant amount of ran domness see Figure 1 Other than the attacker legitimate senders and receivers within the WSN are positioned in an indoor area and their radio transmission has to penetrate fewer obstructions By a planned placement it is even possible to achieve line of sight LOS between legitimate WSN nodes Consequently signal prop agation within an indoor WSN is more controllable and predictable than a signal arriving from outside 2 SECURITY BY WIRELESS In this section we deploy a real world indoor WSN network in order to empirically evaluate the diversity of signal propagation Our testbed is a network installed in our university lab using 8 65 gt WSN scenario Received Signal Strength dBm Distance m b Measured RSS on indoor sensors Figure 2 Residential monitoring scenario Fig 2 a a WSN with 8 sensors is deployed within a residence indoor while an attacker A attempts to inject fake data from outside RSS unpredict
271. ronics Engineers Inc February 2007 http www ieee802 org 15 pub TG4 html IEEE 802 15 2 Standard 2 ZigBee Alliance The Official ZigBee Alliance Site WWW ZigBee Alliance http www zigbee org 2008 3 ZWave Homepage Zensys Inc http www zen sys com July 2007 4 E home Webpage Spelsberg Geb udeautomation GmbH http www ehome online de July 2008 5 Automation System FS20 Conrad Electronic SE http www 1 conrad de fs20 July 2008 6 Optimized Link State Routing Protocol OLSR The Internet Society http www rfc net rfc3626 txt RFC October 2003 7 OpenWrt Wireless Freedom http www openwrt org WWW 2008 OpenWrt org 8 ZigBee Alliance ZigBee Specification ZigBee Document 053474r13 Version 1 0 http www zigbee org en index asp December 1 2006 9 OLSR org Homepage the ad hoc wireless mesh routing daemon http www olsr org WWW April 2007 10 Olaf Koglin Homepage Freifunk http www freifunk de WWW April 2008 11 IEEE 802 11TM WIRELESS LOCAL AREA NETWORKS Homepage Institute of Electrical and Electronics Engineers Inc http grouper ieee org groups 802 11 2007 12 Denis Martin Homepage Implementation Download of OLSR daemon for PPC http www delta my de devel olsrd 0 5 0 wince tar gz July 2008 35 Implicit Sleep Mode Determination in Power Management of Event driven Deeply Embedded Systems Andre Sieber Karsten Walther Reinhardt Karnapke Andrea
272. ropagation A real world radio propagation will be subject to any combination of the aforementioned phenomena resulting in a transfer function depending on many variables By traveling along several paths over different distances and consequently arriving with different phases a signal exhibits multipath propagation properties result ing 1n constructive or destructive interferences the received signal strength may be weakened or amplified compared to LOS propa gation Considering the topology of the testbed depicted in Fig ure 2 an idealistic free space model would provide the strongest RSS for the sensor which is in the attacker s closest proximity M while the most distant sensor 5 would measure the weak est RSS from the attacker s transmission However when look ing at real world measurements the results are significantly dif ferent The outdoor transmission which passes through different obstacles and is altered by various signal effects arrives highly ran domized in an indoor environment Hence the order of RSS re ceived at indoor sensors is completely shuffled Figure 2 b shows measured RSS on every sensor using the topology from Figure 2 a averaged over 100 transmission The sensor with the strongest RSS is 55dBm and the one with the weakest RSS is Mg 75dBm In a free space model Mg would have measured the second strongest RSS moreover this RSS relationship between sensors and attacker is ch
273. rsity of Technology Germany c weyer tu harburg de ABSTRACT Neighborhood relations are changing over time in wireless sensor networks due to different hardware or environmental effects These effects and memory limitations require a bal anced neighborhood management to ensure agility stability symmetry and connectivity The proposed neighborhood management protocol Mahalle is optimized with regard to these four criteria Agility and stability are achieved by ALE a new adaptive link estimator 1 INTRODUCTION Many applications or algorithms for wireless sensor networks depend on a priori knowledge of neighborhood information This encompasses information about neighborhood relations 1 which nodes can communicate with each other and information about the link quality and reliability between those nodes Self stabilizing algorithms are a prominent ex ample of algorithms that highly depend on neighborhood information 3 The quality of neighborhood information can be described by four criteria agility stability symme try and connectivity Before accounting a node as a neighbor the quality and relia bility of the communication link between the nodes must be assessed Many parameters of the environment e g multi path effects lead to a variation of these link properties over time Due to the characteristics of the wireless channel neighborhood relations are not known a priori and can only be gathered after the deploymen
274. rt SLEWS gef rdert 8 REFERENZEN 1 Akyildiz I F Su W Sankarasubramaniam Y und Cayirci E 2002 Wireless Sensor Networks A Survey Computer Networks 38 3 S 393 422 2 Arnhardt C Asch K Azzam R Bill R Fernandez Steeger T M Homfeld S D Kallash A Niemeyer F Ritter H Toloczyki M Walter K 2007 Sensor based Landslide Early Warning System SLEWS Development of a geoservice infrastructure as basis for early warning systems for landslides by integration of real time sensors In Koordinationsb ro GEOTECHNOLOGIEN Science Report Early Warning Systems in Earth Management Potsdam Vol 10 S 75 88 3 Bill R Niemeyer F Walter K 2008 Konzeption einer Geodaten und Geodiensteinfrastruktur als Fr hwarnsystem f r Hangrutschungen unter Einbeziehung von Echtzeit Sensorik In GIS Zeitschrift f r Geoinformatik 2008 Nr 1 S 26 35 4 Blumenthal J Reichenbach F Timmermann D Precise positioning with a low complexity algorithm in ad hoc wireless sensor networks PIK Praxis der Informationsverarbeitung und Kommunikation 28 2005 80 85 88 5 Born A Reichenbach F Bill R und Timmermann D 2006 Bestimmung Optimaler Startwerte zur Exakten Lokalisierung mittels Geod tischer Ausgleichung In 5 GI ITG KuVS Fachgespr ch Drahtlose Sensornetzwerke Technischer Bericht 2006 7 S 93 98 6 Born A Reichenbach F Bill R Timmermann D Loka
275. s not expected Reasons for this could be that the percental influence of the analog parts to the power consumption rises with deeper sleep modes These do not fall under the CMOS power consumption formula Pstatic HC f and thus can behave different At last the formula is idealized no power management simple power management Imp0 full power management power consumption in mW ar Figure 3 Power consumption of Reflex power manage ment schemes for at IMHz and 2 2V For comparison purposes TinyOS 2 0 2 was used The same application was implemented and utilized only implicit power management The results are shown in figure 4 and table 2 0 9 TinyOS power management Reflex power management power consumption in mW Figure 4 Power consumption of TinyOS and Reflex at 1MHz and 2 2V For the given application the results show that the power management of REFLEX is more effective than the one from EEE TMoteSky 2 2V IMHz 0 195 0 324 TMoteSky 3V 1MHz 0 258 0 588 Table 2 Average power consumption of TinyOS and Re flex TinyOS The REFLEX power management consumes about 41 respectively 56 less energy depending on the volt age The main reason is that the power consumption during sleep is higher in TinyOS than in REFLEX Due to the driver module for the serial connection it is not possible to deter mine if the receiving direction is
276. s Lagemann J rg Nolte Distributed Systems Operating Systems group BTU Cottbus Cottbus Germany as kwalther karnapke ae jon informatik tu cottbus de Abstract Energy consumption is a crucial factor for the lifetime of many embedded systems especially wireless sensor net works In this paper we present an approach for power man agement in embedded systems based on the event driven operating system REFLEX In contrast to other systems the implicit power management is mostly hardware inde pendent and efficiently chooses the optimal power saving mode of the microprocessor automatically First experi ments yield promising results comparable to existing implicit power management concepts 1 Introduction Typical sensor net applications such as environmental monitoring mostly demand that nodes should work for years with a single battery Thus saving energy is essential to attain this goal Even in embedded systems with external power supply saved energy helps to reduce costs For min imal power consumption the appropriate hardware must be chosen But the hardware must be appropriately utilized too This is a complex task that should not be the application pro grammer s burden There is a wide range of low power mi crocontrollers available for example the Texas Instruments MSP 430 series 1 These processors provide the program mer with fine grained control over components and sleep modes To utilize these features the opera
277. s mi crocontroller which has special hardware support for low power applications e g detailed sleep modes and fast wakeup times We used a simple experimental application which sends data over the serial connection every second This applica tion was chosen because it shows the characteristics of a typ ical sensor net application such as a low duty cycle and pe riodically sending data over an interface 5 2 Results In figure 3 and table 1 the power consumption for different versions of the power management schemes in REFLEX are compared With deactivated power management there is no difference in power consumption regardless of whether the serial port is active or not In contrast deactivating the idle core the highest possible sleep mode called lpm0 makes the send activity clearly identifiable and shows a significant amount of energy savings This simple power management reduces the energy consumption to about 16 of unman aged version Using the power management approach pre sented in section 4 consequently to utilize the deepest possi ble sleep mode reduces the energy consumption to 7 Reflex w o Ipm0 full TMoteSky 2 2V IMHz 2 825 0 451 0 195 TMoteSky IMHz 7 445 1 053 0 258 Table 1 Average power consumption of Reflex Due to the quadratic influence of voltage the overall power consumption at 3V is higher but as table 1 shows the effect reduces with better power management which wa
278. s representing the link behavior is that they exclu sively describe a snapshot of the state involving only the recent past They can not be used as a reliable estimate of the near future which is needed to provide stability Taking the considerations above into account estimating the PRR by evaluating the packet reception rate of periodical broadcasts is the most appropriate way for a solution inde pendent from hardware and environment In 4 several link 23 estimation methods are evaluated and compared The ex ponentially weighted moving average EWMA is one of the most appropriate methods related to fast response mean error and memory usage PRR is estimated by PRR a 1 a if a packet was successfully received and PRR a if a packet loss occurs A single or consec utive packet loss is detected by a timeout or on the basis of sequence numbers The behavior of EWMA is controlled by the parameter a ranging from 0 to 1 For a 0 99 EWMA is called stable and has a high accuracy with a low mean square error but with a high crossing time metric crossing time is de fined by the time the estimator needs after detecting a new link to come up with a PRR value having an error of at most e In the following e is assumed to be 0 05 For a 0 915 EWMA is called agile This value leads to a much shorter crossing time and a smaller reaction time to link changes but suffers from a high variation with a
279. s the core engine of KleeNet interprets a virtual instruction set it cannot detect hardware platform depen dent assembly level bugs nor enforce runtime memory safety Thus KleeNet complements the beneficial features of Safe TinyOS allowing altogether even more rigorous application testing see Table 1 Overall we make following contributions First we integrate an effective bug finding tool into the event driven TinyOS programming model with usability as a primary goal Sec ond we show that apart from the general checks already available Klee can easily be extended to incorporate further checks useful for testing sensor network applications And third we practically demonstrate that sound testing tech niques can be used throughout the application development process with minimum manual effort 3 SYSTEM OVERVIEW In this section we present an overview of our system First we introduce Klee which we use for symbolically executing C applications We continue by discussing how we automat ically instrument source code without the help of applica tion developers We conclude this section by presenting our extension of Klee to enable struct type checking in sensor network applications 3 1 Klee Klee is a symbolic execution tool for C programs based on LLVM 11 In contrast to common runtime testing where the program input is manually generated Klee runs the code on symbolic input initially allowed to be anything The p
280. s the pointer to this data is casted to a known struc ture type based on particular bit fields During code execu tion further casts on this memory location are executed We have extended the functionality of Klee to check the struct type equality during casting operations This check is op tional but nevertheless the warnings as shown in listing 4 are useful and facilitate program comprehension NewRoutex msg NewRoutex payload message processing call Queue enqueue msg RouteUpdatex msg RouteUpdatex call Queue dequeue further message processing Listing 3 Casting between pointers to different structure types make kleenet test struct KLEE WARNING Struct types don t match KLEE struct NewRoutex gt struct RouteUpdatex Listing 4 KleeNet warnings 4 INTEGRATION INTO TINYOS We decided to integrate Klee into TinyOS by adding a vir tual KleeNet platform based on the TinyOS null platform This approach allows us to easily add different modules to a platform that automatically marks sensor value input and incoming packets as symbolic Since TinyOS applications are event driven parts of the code are executed only eventually after particular events are fired In order to cover all possible program control flow paths during testing we have extended our virtual platform with an automatic event signaling mechanism Once an ap plication is booted all implemented events are signaled and p
281. sagtem Areal an geordnet Me daten der Sensoren auf Schicht 1 werden auf den Knoten vorverarbeitet um Fehlalarme m glichst auszu schlie en Die Daten werden ber Schicht 2 Knoten an die Basisstation weitergeleitet wo eine entsprechende Aggrega tion und Klassifikation der Daten stattfindet ber die Ba sisstation kann das Verhalten s mtlicher Knoten konfiguriert werden 2 4 Designmerkmale Betrachtet man obige und weitere Einsatzszenarios genauer stellt sich heraus dass sich die dort realisierten Systeme in 54 vielen Designmerkmalen unterscheiden 3 9 Nachfolgend sind einige der wesentlichen technischen Merkmale aufge f hrt e Kommunikationstopologie Die Kommunikationstopologie beschreibt welche Kno ten mit welchen anderen Knoten direkte Funkverbin dungen aufbauen und auf welchen Pfaden Nachrichten pakete zu anderen Knoten gelangen M gliche Topolo gien sind z B die Sterntopologie S mtliche Sensor knoten kommunizieren direkt mit einer Basisstation oder die Baumtopologie Pakete werden ber mehrere Zwischenknoten an eine Basis geleitet e Heterogenit t der Knotenhardware Eine Applikation kann aus gleichf rmigen Sensorkno ten bestehen oder es existieren Knoten mit unter schiedlicher Hardware die verschiedene Aufgaben wahr nehmen e Kommunikationsrichtung Werden Nachrichtenpakete stets von den Sensorknoten an die Basis geschickt oder auch in anderer Richtung von der Basis zu den Knot
282. self organizing 802 15 4 networks 61 in simulations in terms of packet loss delay and energy consumptions This is mainly due to routing overhead as HERA gets the initial routes for free by exploiting the associ ation messages The authors do not mention whether HERA can be extended to duty cycling networks and do also not analyze the association procedure Their interest is just on the operational phase of the network lifetime demonstrat ing the benefits of the association procedure Together with an optimized energy efficient association procedure HERA could be a promising approach for low power networks 3 ANALYTICAL FRAMEWORK 3 1 Problem Description In the last section we presented some examples out of the numerous energy efficiency studies of 802 15 4 algorithms They are good example for all studies we know of as they mainly concentrate on the operational phase of the network For this purpose the network is in general assumed to be associated and routes stable A reasonable approach if e g the average delay of a packet routed in a WSN is of inter est As the association procedure of 802 15 4 networks is a vital MAC layer functionality 5 we will concentrate on the initial phase of a 802 15 4 WSN in the following A good description of the association procedure can e g be found in 11 we only recall the most important facts Sensor node n activated in a nonbeacon enabled multihop PAN will start an active
283. sents the lower two layers physical and data link This feature will offer us additional channels in the ISM frequency band that are or thogonal to IEEE 802 11 Recent reports 6 and our own measurements 8 have identified this aspect as one of major concerns for multi transceiver based research Interference does exist even between orthogonal channels Some opera tion system related changes and extentions are required both on the sensor node as well as the mesh router side We will use a stripped down version of our ScatterWeb OS Most interrupts have to be switched off for low latency ex cept the ones used for transceiver and serial line handling One USART interface of the microcontroller is connected to a USB to serial converter chip Data sent from the host to the sensor node has to be relayed with minimal handling to the transceiver and vice versa Media access and queuing of frames shall remain in the firmware but everything else has to be handled by the Linux kernel We call this integra tion wireless sensor kernel interface WSKI There are two ways to implement the WSKI Right now the sensor node is represented as a type writer device e g ttyUSBO With a custom tailored serial communication protocol we send ar bitrary data via the transceiver This solution is simple and requires no code changes but complicates a seamless in tegration While sending data between sensor nodes each packet has to be extracted from the Lin
284. simulations revealed that a range of 0 06 is a good choice between oscillation and size of the hysteresis This leads to PRRin 0 86 and PRRout 0 74 4 NEIGHBORHOOD MANAGEMENT The neighborhood management protocol Mahalle uses the information about link qualities between nodes provided by the adaptive link estimator described in the previous sec tion ALE delivers only information about one direction of a communication link If node A receives packets from node B it can calculate the PRR of node B at node A by using ALE But this information is of no significance for the PRR of node A at node B To ensure symmetry of the commu nication links between neighbors the PRR values of both directions of the link must be exchanged The PRR is normally exchanged between nearby nodes by using periodical broadcasts In general three different ap proaches exist The simplest way is to put the estimated PRR values of all known nodes in each packet The main drawback of this approach is the required size of such a packet Due to the high bit error rate the probability of packet loss is increased which influences the PRR estima tion The second approach is to send only the PRR values that have changed lately This decreases the amount of in formation sent in one packet but the size varies If a packet gets lost the information must be resent even if the PRR is unchanged The complexity of this approach is thereby increased Another a
285. sion of the image must be the same as the version of the Golden Image on the node The only possibility to upgrade the AirFlash version on the node is to exchange its Golden Image via AirFlash Before starting the image transfer the necessary number of pages inside the node s memory is calculated as well as CRC checksums for each page Then a command to erase the selected slot on the node is sent After an ACK is received the image transfer starts This process is analogous to the one described in chapter 4 2 If an error occurs that cannot be handled by the protocol the process is aborted and must be restarted via the user interface 5 PERFORMANCE Finally the transfer performance of AirFlash is evaluated To test the upload speed images of 16kB 32kB 64kB and 120kB were transferred multiple times to a node The elapsed times were collected and averaged If the radio channel is not frequently used by other applications the transfer durations are almost constant 14 for a fixed image size In average a transfer rate of 1 1 kB is achieved 16 32 64 0 20 40 60 80 100 120 Image transfer time sec Image size kB Figure 6 Transfer times for various image sizes 6 CONCLUSION The discussed AirFlash application is well suited for experimental networks since it allows the programming of heterogeneous networks The use of a Golden Image makes the application very reliable if
286. six points out an image should not be flooded to all nodes in the network It should be possible to address nodes individually so that various applications can be 11 present inside the WSN And finally since AirFlash is only used for reprogramming while the node normally executes its usual measurement application AirFlash should stay in the background if not used and claim as few resources as possible All these features named above are highly useful for running and maintaining a WSN They are all included in the AirFlash application discussed in this paper This feature combination is unique to the presented AirFlash and is not found in other OTAP implementations like Crossbow OTAP 2 or Deluge T2 3 For example neither Crossbow nor Deluge support feature three and five Deluge additionally does not support feature six And Crossbow has the disadvantage that it consists of proprietary code These are some of the reasons why AirFlash was created The implementation of all the listed features into AirFlash will be introduced in the following sections 3 PLATFORM AirFlash is realized and tested for the IRIS node 4 as an example This is a wireless node with a 2 4 GHz radio transceiver and an ATmegal281 microcontroller that contains 128kB of internal program memory Flash and 4kB of internal EEPROM 5 Additionally the IRIS node provides 512kB of external flash memory which will be used to store multiple firmware images AirFl
287. so lutions in this section Data aggregation tries to optimize the number of transmissions throughout the sensor network while data compression strives for a local optimization of the amount of transmissions and their payload size 2 1 Data Aggregation If incoming packets and locally generated data are addressed to the same destination combining these data sets into a single packet can effectively reduce the number of transmis sions required 5 For example consider the linear topology depicted in Figure 2 The rightmost node 5 acts as a sink to sensor readings from the remaining nodes A B C MulttHop Forwarding Data A Data A io Data Data A Data A B Data A B C Data Aggregation Figure 2 A linear sensor network topology If only one hop connectivity is given a total of six trans missions are required to forward all measured data to the sink depicted as the multi hop forwarding case Another downside of this issue is that node C located next to the sink must forward a higher count of packets than A or B and will thus most likely deplete its energy budget first Aggregating the incoming sensor data with local readings and forwarding these results to the sink can reduce the re quired number of transmissions from six to three in this basic scenario hence achieving savings of 50 in the number of transmissions This is indicated in the lower part of Figure 2 where only a si
288. ssed how the image packets are received at the node and how they are written into the slots of the external memory This is the task of the AirFlash module which is described in the following 4 2 AirFlash Module The AirFlash module is implemented as a TinyOS component and therefore can be integrated into any TinyOS application The integration of this module needs less then 6kB of additional program memory It uses the node s radio to receive image data from the host PC and to transmit status information The radio device is shared between the AirFlash module and the main TinyOS application The main application possesses the master control over the radio The request and release of the radio by AirFlash is signaled to the main application by events The AirFlash module can assume three main states see Figure 3 In order to save resources the AirFlash module works in the background of the main application and is mostly in the sleep state timer event Sleeping Peeking sleep packet or timeout AirFlash packets Figure 3 Sleep wake cycle of AirFlash module The module periodically enters the peeking state to listen for AirFlash related packets If nothing is received the module goes to sleep after a while If a so called enumeration packet is received the module sends a status message back to the host If a wakeup packet is received the module steps into the Wide awake state Inside this state it is ready to
289. ssed more hops at the multiplexing node gets priority over the younger flow This has the advantage of being very easy to implement with little overhead since all that is required is a hop counter in each data frame which gets incremented by each server that forwards the frame The scheduler can then be a simple priority queue Since data packets usually carry a header adding that counter should be trivial in most cases Another common scheme would be to prioritize flows that have the most hops left until their destination However that fails in sink tree networks since at any given node each passing flow will have the same distance left to travel to a given sink 21 LFF in Network Calculus Multiplexing in Network Calculus is handled by reducing the minimum service curve offered by a network node by the sum of all maximum arrival curves that can possibly be scheduled before the flow of interest In the following let be the input bound of flow f at node n and be the set of flows interfering with f at node n according to the rules of the analysis method The effective service curve offered to a flow of interest f at node n can then be given as Ons 45 vo 20 Sink nz Sensed input Figure 2 Simple feed forward network To implement LFF it is only necessary to filter Fn so that only flows that have passed at least the same number of hops
290. stitutionen teilgenommen haben die im Bereich Sensornetze t tig sind und auf dem erste Ideen zum Vorgehen zusam mengetragen und diskutiert wurden In diesem Abstract sol len die dort diskutierten Ideen und das geplante Vorgehen im o g Projekt vorgestellt werden Im folgenden Abschnitt 2 sollen kurz beispielhaft einige Einsatzszenarios f r Sensor netze und m gliche technische Merkmale diskutiert werden In Abschnitt 3 soll ein berblick ber unterschiedliche Si cherheitsma nahmen verschafft werden Schlie lich wird in Abschnitt 4 das Vorgehen im Projekt erl utert 2 EINSATZSZENARIOS F R DRAHTLO SE SENSORNETZE Drahtlose Sensornetze werden unter anderem f r den Ein satz in der Umweltbeobachtung im Katastrophenschutz in der Infrastruktur und Geb ude berwachung im Gesund heitwesen in der Logistik oder f r die Grenz berwachung vorgeschlagen Im folgenden sollen beispielhaft einige dieser Szenarios konkretisiert werden 2 1 Stabilit tspr fungen Ein Einsatzszenario das intensiv untersucht wird ist die Stabilit tspr fung von Bauwerken insbesondere Br cken Bauwerke sind in vielen F llen starken mechanischen Bela stungen ausgesetzt sei es durch Benutzung schwere Fahr zeuge die ber Br cken fahren oder ung nstige Umwelt einfl sse Erdbeben St rme etc Dies macht eine st ndige Pr fung der Integrit t solcher Bauwerke erforderlich H u fig werden solche Pr fungen mit Hilfe fest verb
291. strat egy This illustrates that more studies for very low duty cycles and more sophisticated algorithms for the association procedure are necessary ad g 4m x 0 1 A 0 5m x 0 1 9 4m x 0 5 Wy e g 5m x 0 5 ty s 4 g 4m x 1 A o 4 g 5m x 1 Figure 4 9 for strategy 2 5 CONCLUSION In this paper we examined possibilities for enabling the association process in a self organizing low power 802 15 4 nonbeaconed sensor network To our knowledge this prob lem has not been studied before We pointed out directions for an efficient solution by examining three different simple 64 strategies for proceeding after the initial association attempt fails To compare the benefits and trade offs of the individ ual solutions on the performance of the WSN startup phase we used three metrics suitable for characterizing the quali tative and quantitative performance of a specific association solution Our results illustrate the inherent energy saving poten tials of the often neglected start up phase of a 802 15 4 sen sor network In dependence of the strategy and the network connectivity the association retry interval has to be cho sen with care in order to avoid wasting energy We found that under some conditions too short retry intervals lead to unnecessary channel scans and beacon collision thereby de teriorating the association performance Under other condi
292. t of different packet sizes on packet reception rate remains to be investigated In practice nodes do not transmit at the same short in tervals used during our measurements We have yet to in vestigate whether the correlations derived in section 4 hold 29 for longer intervals between reception events Devising other mechanisms such as sending probe packets to evaluate link quality could also be used but it will raise the communica tion overhead significantly 6 Conclusion Intermediate quality links constitute a significant frac tion of wireless links in low power WSN STLE captures short term link dynamics of these links at a high resolution in time and predicts their temporary availability or unavail ability Our measurements indicate that these intermittent links if utilized can significantly improve the performance of routing algorithms by offering further choices for forward ing packets Moreover STLE is more suitable for networks showing bursty traffic patterns such as in typical applications of WSN 7 References 1 Anastasi Conti Gregori A Falchi and A Passarella Perfor mance Measurements of Mote Sensor Networks In 7th ACM Interna tional Symposium on Modeling Analysis and Simulation of Wireless and Mobile Systems MSWiM Venice Italy October 2004 2 A Becher O Landsiedel G Kunz and K Wehrle Towards short term link quality estimation In Proc of The Fith workshop on Em bedded Ne
293. t the position of the current node its immediate neighbors and the message destination It is believed that such localized protocols provide scalable solutions that is solutions for wireless networks with an arbitrary number of nodes Future sensor networks for instance might consist of thousands or ten thousands of nodes Kristen Pind Universtity of Southern Denmark Campusvej 55 DK 5230 Odense M Denmark kristen imada sdu dk Research on localized algorithms has focused mainly on the oretical investigations and simulation While both form an important aspect in getting a thorough understanding on what is going on a critical part has widely been neglected so far It is of paramount importance to evaluate such pro tocols in the domain where they are supposed to be applied Localized algorithms are not supposed to run in Ns2 Om net or Qualnet They are supposed to run on real hardware platforms like MicaZ or Tmote Sky nodes What follows is just a drop in a bucket to provide some numbers of real world measurement for comparing the do main of localized versus topology based data communica tion We focused at first on two protocols greedy routing and dynamic source routing The protocols and how we im plemented them are described briefly in Section 2 Then we describe our experimental setup in Section 3 The results are presented in Section 4 We conclude our work in Section 5 2 PROTOCOL IMPLEMENTATIONS We implemente
294. t when the network is in operation The local view upon the neighborhood of a node forms the topology of the network at the global scope Variations in hardware changes in the environment or node failures lead to dynamic changes of the topology Therefore neighbor hood relations are not fixed during the whole lifetime of the network Frequent topology changes lead to a degradation of the algorithms or applications running on top of the formed topology This aspect of the quality of a neighborhood pro tocol is captured by the concepts of agility and stability Agility measures the speed of adapting to desirable events such as failure of a node respectively a link or deployment of a new nodes Stability is the ability to ignore transient failures Despite variation over time the neighborhood relations must be consistent between nodes This requirement is expressed by the symmetry property In particular if node A is in the neighborhood of B then node B must also be part of the neighborhood of node A Otherwise the resulting topology is not useable in most cases Reason for such an asymmet rical neighborhood relation is the existence of different link qualities caused by different noise levels due to fluctuations in hardware accuracy Another reason lays in the memory restriction of sensor nodes Memory restrictions are an im portant topic in regard to ensuring connectivity If the den sity of the network is higher than the number
295. tOS according to its priority Fig 2 shows the integration of the Ghost subsystem into a SmartOS based software Since SmartOS is fully preemptive it is sufficient to link the Ghost modules into the the WSN application Besides adding the high level task the linker also places the low level mo dule adequately to start the execution of the emergency recovery function at system start prior to the operating system The required memory RAM ROM adds up of the Ghost modules plus the com munication protocol and the flash memory driver However the last two components are often part of the application anyway and thus mean no extra system load Yet a slim and OS independent version of the flash memory driver is required within the low level module to gain read access to this device In general performance losses caused by Ghost are negligible since only the relatively rare Ghost commands trigger the execution of the high level task As already mentioned in section 2 3 an energy harvester is optional but can account significantly to the system stability during memory update processes 4 TESTBED AND RESULTS For evaluating the just described techniques under real world con ditions we used a SNoW Bat 4 indoor localization system com prising 45 nodes In this section we will present some results con cerning update speed resource requirements and energy efficiency 83 20 kB program image SmartNet at 250kbps no encryption 45 nodes
296. tance intervals on other verifying sensors which are successfully broken The results of a thorough empirical sampling of var ous physical positions used to attack the WSN network with different number of verifying sensors are presented in Table 1 as the number of secure wireless channels 1 the frequencies which did not result in an ceptance of any fake frames So for example a WSN with only two verifying sensors one sender one receiver and an interval width of 4dBm enables an injection of fake frames on eight channels while the other eight are not possible to attack using the same con figuration of transmission properties Similarly using three sensors to verify the transmission and the width of acceptance intervals set to 2dBm the attacker could use only three wireless channels for successful injection while its attack is not effective on the other 13 channels Hence there is a simple tradoff provided by this scheme by increasing the number of sensors we can increase the number of secure channels even in environments with a high variance of RSS which results in larger o and wider intervals In the de ployed WSN network using five or more sensors we could not find any position that resulted in a single successful injection even if acceptance intervals were set to 10dBm allowing very high RSS variations for more implementation results see 4 4 CONCLUSION The main purpose of this paper was to briefly demons
297. te data frames having length of 10 25 50 100 or 118 bytes we calculate the required CFP slots with the acknowledgment transmission in Figure 4 a We calculate same for the unacknowledgment transmission in Figure 4 b In both figures we consider the value of SO as 0 1 and 2 It is clear that even for the maximum size of data packet we need only one GTS slot if we use SO of 2 For SO of 1 we need at most three GTS slots Now to validate the improvement given by our approach for bandwidth under utilization problem consider an exam ple where five devices are generating data packets each of 100 75 50 100 and 25 bytes and request for 3 2 1 2 and 2 GTS slots respectively Figure 4 c confirms that our tech nique outperforms the original protocol in terms of band width utilization and thus more devices can take benefit of the guaranteed services 15 aMinCAPLength 60 MaxC F Pavail aBaseSlot Duration 279 3 2 Data IFS aTurnaroundTime Cr Pye aBaseSlot Duration 280 4 where Data MSDU MHR PHR ACK 6 CONCLUSION In this paper we have elaborated the suitability and limita tions of the IEEE 802 15 4 for low latency bandwidth and energy related critical medical applications protocol needs definite enhancements before it could be applied to 60 the medical field and our simple approach deals with almost all of those limitations Application specific optimal param e
298. ted as next hop link layer acknowledgments continuously provide in formation about link availability We expect that the probability of a successful packet transmission depends on the success rate of any recently sent packets i e the more packets were transmitted successfully in the recent history the higher the probability is that an up coming packet is transmitted successfully too 3 2 Integration with Long Term Link Estima tors and Routing Protocols STLE is designed to embed deeply into the routing proto col and to cooperate with long term link estimators Modern sensor network routing protocols such as BVR 7 use the number of expected transmissions to a destination as rout ing metric computed by combining the distance in hops and the number of expected retransmissions Both long term and short term link estimation aim to predict the number of necessary retransmissions on a link each on their respective temporal granularity Consequently when no short term es timation for a link is available the routing protocol will use the long term prediction as fallback probably resulting in conservative link selection As explained above neighboring nodes overhear ongo ing data flows and may suggest themselves to the forward Short Term Link Quality Dee 1 0 8 0 6 0 4 SSS999999990H 0 2 0 4 Long Term Link Quality 0 8 0 6
299. ted in 6 and 5 in which properties of wireless com munication are used to assist in providing various security objec tives adapted to the peculiarities of wireless networks 5 REFERENCES 1 Y Chen W Trappe and R Martin Detecting and localizing wireless spoofing attacks In Proceedings of the Fourth Annual IEEE Communications Society Conference on Sensor Mesh and Ad Hoc Communications and Networks pages 193 202 2007 2 M Demirbas and Y Song An RSSI based Scheme for Sybil Attack Detection in Wireless Sensor Networks In WOWMOM 06 Proceedings of the 2006 International Symposium on World of Wireless Mobile and Multimedia Networks pages 564 570 June 2006 3 D Faria and D Cheriton Detecting Identity based Attacks in Wireless Networks using Signalprints In WiSe 06 Proceedings of the 5th ACM workshop on Wireless Security pages 43 52 September 2006 4 I Martinovic L Cappellaro and J Schmitt Lightweight Crypto free Authentication Against Injection Attacks in Wireless Sensor Networks Technical Report 369 08 University of Kaiserslautern Germany July 2008 5 I Martinovic N Gollan and J B Schmitt Firewalling Wireless Sensor Networks Security by Wireless In 3rd IEEE International Workshop on Practical Issues in Building Sensor Network Applications SenseApp 2008 Montreal Canada October 2008 IEEE Accepted for publication 6 I Martinovic F Zdarsky
300. ters setting for the health care scenario is also discussed where we have found that the GTS slots length requested by the devices mostly causes bandwidth under utilization problem The coordinator should assign GTS slots depend ing on the traffic delay and energy constraints to the device Even for the maximum data packet length the device needs only one CFP slot for the SO value of 2 7 REFERENCES 1 IEEE 802 15 4 2006 MAC and PHY specifications for LR WPANs http ieee802 org 15 pub TG4 html 2 Moteiv Tmote http www sentilla com pdf eol tmote sky brochure pdf 3 Tinyos 2 x http www tinyos net tinyos 2 x 4 Y K Huang Pang and H N Hung An adaptive GTS allocation scheme for IEEE 802 15 4 IEEE transactions on parallel and distributed systems August 2007 5 L C Ko and Z T Chou A novel multi beacon superframe structure with greedy GTS allocation for IEEE 802 15 4 wireless pans JEEE Communication Society WCNC 2007 6 A Koubaa Alves and E Tovar i game An implicit GTS allocation mechanism in IEEE 802 15 4 for time sensitive wireless sensor networks in Proc of 18th Euromicro Conference on Real Time Systems 7 A Koubaa Alves and Tovar GTS allocation analysis in IEEE 802 15 4 for real time wireless sensor networks in Proc of the 14th International Workshop on Parallel and Distributed Real Time Systems WPDRTS Apr 2006 8 N Salles N Krommenacker and V L
301. th the inter grid spacing 5 m are shown by dashed lines solid lines are used for 4m monotonic in crease of the curves and the fact that the association success for 4 and the same value of x is always greater or equal then for the same x 5 m illustrates that the association success is increasing with the connectivity of the network While this is an obvious result another observa tion is more surprising The highest association success is guaranteed for a 0 i e if node 9 immediately retries to associate after an association failure Our studies showed that in denser topologies this behavior leads to unnecessary channel contentions choosing a larger but randomized a is more advantageous in these scenarios Fig 3 illustrates how the time required for successful asso ciation varies with the relative position of the node and the duty cycle in a topology where all nodes can only communi cate with their direct neighbors and under strategy 1 with x 0 1 ie a U 0 0 1pu T 1 sec We use a surface plot and study the influence of very small duty cycles more closely Observe that ta n is increasing strongly with the distance from the sink and the sleep time 1 pw T This increase was observed for all chosen parameterizations of a and all topologies but differs in magnitude for varying ra dio range and parametrizations of a Keep in mind that the surface plot shows mean values obtained from 20 s
302. the consequence of the short transmission range and high mobility of the nodes However the num ber of link breaks and topology changes can be reduced in a significant way if the forwarding nodes along a route are chosen carefully In this paper an approach is introduced which decreases the number of topology changes by taking advantage from correlated node movement The approach is based on de ferring routing messages and can be used for many popular routing protocols to increase their performance Its impact on the end to end reliability and the selection of forward ing nodes in high mobile networks is simulated by adding the method to the Ad Hoc On Demand Distance Vector AODV and the Statistic Based Routing SBR protocol Categories and Subject Descriptors D 2 2 Network Protocols Metrics Routing Protocols delay based routing metrics General Terms Algorithms Management Performance Reliability Theory Keywords Mobility Routing Wireless Metric Delay 1 INTRODUCTION The number of link breaks usually corresponds to the movement speed of the nodes in the network Therefore mobility is often considered as a major problem in wireless networks since it is mainly responsible for topology changes Nonetheless in dense mobile networks there is a high prob ability that several nodes move along in the same direction with similar speed Thus the relative movement speed of these nodes is quite small compared to their a
303. the current GTS allo cation is the bandwidth under utilization Most of the time device uses only a small portion of the allocated GTS slots the major portion remains unused resulting in an empty hole in the CFP Moreover the protocol explicitly supports only seven GTS allocations In the medical field where one ill ness usually boost ups other illnesses many devices should be able to reach the coordinator via such guaranteed ser vices Also the current protocol only supports first come first serve based GTS allocation and does not take into ac count the traffic specification delay requirements and the energy resources In medical scenario many critical events may occur at a time and some of them are more critical and need most urgent response With the current proto col the device can request for all seven GTS slots even if it is not really needed Such unbalanced slot distribution can block other needful devices to take such timeliness ad vantages Moreover the protocol uses GTS expiration on the basis of some constant factors and the assigned GTS slots are broadcast for the constant number of times in the superframes Such restrictions also cause unnecessary en ergy consumption and CFP slots blockage for longer time Even if CFP is not present in the superframe the beacons transmitted by the coordinator always use unnecessarily one byte of the CFP resulting in energy inefficiency Last but not least the current superframe s
304. the sink Due to the sink tree structure of the network all flows that join a flow of interest remain multiplexed until the sink making it possible to calculate the total informa tion transfer delay D for a given sensor node i by using a flow specific end to end service curve PMOOA can be shown to deliver a tight bound for sink trees 7 of homoge neous nodes When compared to the addition of the nodal delay bounds as done by the TFA this results in consider ably less pessimistic bounds as each interfering flow s burst has to be taken into consideration only once Design Concepts of a persistent Wireless Sensor Testbed Bastian Blywis Felix Juraschek Mesut G nes Jochen Schiller Freie Universit t Berlin Takustra e 9 14195 Berlin Germany blywis jurasch guenes schiller inf fu berlin de ABSTRACT In this paper we introduce our work in progress wireless testbed The testbed consists of hybrid nodes Each node contains a sensor node and a mesh router We present con cepts which exploit the features of an existing wireless mesh network for the wireless sensor network The concepts in clude the wireless sensor testbed integration WSTI which simplifies the management and provides a basic infrastruc ture for the sensor network and experiments The integra tion into the Linux kernel is realized by the wireless sen sor kernel interface WSKI Network check points NCPs enable long term experiments that may span several
305. ting system should at least be able to provide the application with power saving mechanisms It would be even better to do this implicitly without any need of control from the programmer Event driven operating systems can potentially go into sleep mode if no event is pending and the hardware can wakeup the rest of the system upon the occurrence of an event Since most microcontrollers support a variety of sleep modes with dif ferent energy footprints the selection of the mode can have a perceptible effect on the lifetime of battery powered de vices If the decision was wrong the energy savings could be marginal or even worse events can get lost The rest of the paper is structured as follows In sec tion 2 the related work is presented Section 3 gives a short overview over the REFLEX operating system In section 4 the power management mechanisms of REFLEX are described Early evaluation results are presented in section 5 Finally a conclusion is given in section 6 2 Related Work The most common operating system for wireless sensor nodes is TinyOS 2 which is event based The power man agement sends the controller to sleep if the task queue is empty To determinate the deepest possible sleep mode the scheduler evaluates a dirty bit which is set by ev ery driver module by calling the McuPowerState update function This results in a re computation of the lowest pos sible sleep mode by reading all device registers The update
306. tions For 802 15 4 performance evaluations this approach which has been introduced by 3 We extend upon this model by including the results of 10 The authors propose to use a simplified radio control state machine ex tracting values for current consumptions and times required in states or for state transitions from transceiver data sheets and using a typical voltage of U 1 8V By measurements they showed that that this so called Communication Subsys tem Energy Consumption Model CSESM is exact enough for an analysis In Fig 1 we show the corresponding simpli fied state machine for Texas Instruments CC2420 8 which is a widely used 802 15 4 compliant transceiver module The durations and power consumptions of states and transitions are taken from 8 or estimated as the average of the two initial and final state for the transition 10 Power Off PO 0 001 mA Power Down PO I 0 02 mA turn Voltage Regulator t 62 76 msec 0 62 mA crystal oscillator on 0 96 msec 0 426 mA Idle 1 0 426 mA Transmit TX 1 8 5 17 4mA Receive RX 1 19 7 mA t 0 192 msec 14 1 18 55 mA Figure 1 Simplified state machine of 2420 For analyzing we focus on node n and neglect the costs the node which with n associates has for sending the association beacons ton As CSESM is not yet implemented in our simulation framework an estimation for
307. tions in contrast shorter retry intervals are speeding up the association procedure significantly Our results thus demon strated that extensive studies on a wide range of parame ters are necessary as many interacting factors influence the performance of the association process and that an opti mal strategy has to consider a huge range of aspects An extensive factor screening for comparing the benefits and trade offs of different association mechanism will therefore be the topic of future works 6 REFERENCES 1 HART Communication Foundation http www hartcomm2 org 2 ZigBee Alliance http www zigbee org 3 B Bougard F Catthoor D C Daly A Chandrakasan and W Dehaene Energy Efficiency of the IEEE 802 15 4 Standard in Dense Wireless Microsensor Networks Modeling and Improvement Perspectives In DATE 05 Munich Germany March 2005 4 Cuomo S D Luna U Monaco and Melodia Routing in ZigBee benefits from exploiting the IEEE802 15 4 association tree In IEEE ICC 2007 Glasgow Scotland June 2007 5 IEEE Computer Society IEEE Standard 802 15 4 MAC and PHY Specifications for Low Rate Wireless Personal Area Networks September 2006 6 Kohvakka Kuorilehto Hnnikinen and T D Hmlinen Performance analysis of IEEE 802 15 4 and ZigBee for large scale wireless sensor network applications In PE WASUN 06 Terromolinos Spain October 2006 7 Staehle Ho feld N
308. to 20 for 20 simulation runs which were found enough to result in satisfying confidence intervals for all considered mean of the considered metrics For sakes of clearness confidence intervals are thus not shown 0 4 0 a g 5 0 Ber e g 4m x 0 5 0 2 u e g 5m x 0 5 4 g 4m x 1 4 g 5m x 1 45 10 5 0 P dB Figure 2 t 9 for strategy A self organizing PAN initialization is simulated as fol lows The nodes 1 9 are activated in a random order with a time of U 0 1 60 sec between the start times After having started each node tries to associate to a PAN us ing the procedure described in Section 3 1 During most of our simulation runs A was found to be enough for all nodes to associate Only for pw lt 5 a transmission out put power resulting in the smallest connectivity degree and nodes close to the sink starting last it occured that the outmost nodes were not able to associate association success of node 9 s4 9 for the extreme scenario pw 1 is thus a good metric for the quality of an association strat egy In Fig 2 we illustrate the percentage of simulation runs node 9 was able to associate under varying transmis sion powers inter node spacings and parameters for setting a according to strategy 3 i e x Different colors and markers show different values for x Sparser topologies i e wi
309. tor which defines a coverage zone Z with radius r 3 The mobile client transmits the ultrasound chirp as denoted in the previous radio broadcast 4 Each beacon which detects this ultrasound signal cal culates its distance towards the mobile client from the time difference of arrival between the radio and ultra sound signal If an anchor just receives the initiating radio message but no chirp it aborts measurement and waits again for further clients to be localized 5 Each beacon which successfully completed its distance measurement passes this result together with its known position to the requesting client as unicast This is done by the self organizing communication protocol HashSlot 2 3 6 By using for example multilateration the client can estimate its three dimensional position autonomously from at least four such distance measurements 4 DEPLOYMENT STRATEGY Inspired by the localization procedure we head for the de ployment stage The main idea of our deployment approach is reversing the roles of anchors and clients during deploy ment That is just during deployment a reference client below the anchor plane knows its position whereas every beacon doesn t yet know its coordinates The remaining re quirements from section 2 are retained only the reference client generates ultrasound signals which can be detected by the grid aligned beacons at the anchor plane The de ployment procedure will than be as f
310. trate how the chaotic nature of wireless communication can be used as an advantage and how it can assist in increasing security of WSNs Rather than implementing additional protocols necessary for tradi tional security mechanisms we designed practical and lightweight protection without requiring more features than the communication itself provides To successfully inject fake frames the attacker is forced to search for an appropriate configuration of a physical po sition antenna orientation and transmission power However even if the configuration for a successful attack is found it is promptly invalidated by a dynamic change of acceptance intervals Addition 68 ally the design allows for a simple yet effective tradeoff depending on the security objectives of a particular scenario increasing the number of sensors results in increasing the security of a WSN and its resilience against injection attacks Our solution neither requires specialized hardware nor introduces complex message exchange Probably the closest research to our work is tackled by Demirbas et al in 2 While in their work the authors attempt to minimize the randomness of signal strength by comparing the ratios of RSSI values in this work we take a different approach and show that protection can be based on sup porting the randomness and using it against an attacker Further instantiations of the paradigm security by wireless as we call it is demonstra
311. tructure contains at least aMinCAPLength a constant size CAP For most urgent scenarios we may need flexible size CAP rather than the fixed one 3 RELATED WORK Most of the IEEE 802 15 4 related research has been sub jected to the CSMA CA and general performance evalu ation Only a small amount of the literature deals with the timeliness and bandwidth issues An implicit GTS al location scheme i GAME for the time sensitive WSN is proposed in 6 where the coordinator uses an admission control algorithm to decide whether to accept a new GTS request or not based on the traffic specification of the flows and their delay requirements Using Network Calculus for malism in 7 two accurate models for the service curve are proposed and the delay bounds guaranteed for GTS allo cation is provided An expression of the duty cycle as a function of the delay is also presented Based on those results the impact of the BO and SO on the maximum throughput and delay bound is analyzed In 4 adaptive GTS allocation AGA scheme is proposed which consid ers low latency and fairness There are two phases for the proposed scheme In the classification phase devices are assigned priorities in a dynamic fashion based on recent GTS usage feedbacks Devices that need more attention from the coordinator are given higher priorities In the GTS scheduling phase GTSs are given to devices in a non decreasing order of their priorities A starvation
312. tworked Sensors Emnets 2008 June 2008 3 A Cerpa J L Wong L Kuang M Potkonjak and D Estrin Sta tistical model of lossy links in wireless sensor networks In Proc of the 4th International Symposium on Information processing in Sensor Networks IPSN April 2005 4 A Cerpa J L Wong M Potkonjak and D Estrin Temporal proper ties of low power wireless links Modeling and implications on multi hop routing In Proc of the 6th ACM Interational Symposium on Mo bile Ad Hoc Networking and Computing MobiHoc May 2005 5 A Cerpa J L Wong Potkonjak and D Estrin Temporal proper ties of low power wireless links modeling and implications on multi hop routing In MobiHoc 05 Proceedings of the 6th ACM interna tional symposium on Mobile ad hoc networking and computing 2005 6 R Fonseca O Gnawali K Jamieson and P Levis Four bit wire less link estimation In Proc of the Sixth Workshop on Hot Topics in Networks HotNets November 2007 7 R Fonseca S Ratnasamy J Zhao C T Ee D E Culler S Shenker and I Stoica Beacon vector routing Scalable point to point routing in wireless sensornets In 2nd Symposium on Networked Systems Design and Implementation NSDI May 2005 8 T He S Krishnamurthy J A Stankovic T Abdelzaher L Luo R Stoleru T Yan L Gu J Hui and B Krogh Energy efficient surveillance system using wireless sensor networks In MobiSys 04 Proc of th
313. ty gt 0 9 08 en 0 6 Intermediate links _ 06 n X 0 1 lt quality lt 0 9 ee ee 3 x 0 4 E 04 Bad links quality lt 0 1 0 2 0 2 F 0 0 0 01 02 03 04 05 06 0 7 08 09 1 2 4 6 8 Long Term Link Quality a Empirical distribution of long term link qual ity in our testbed Intermediate quality links com prise roughly one third of all useful links 10 Distance metres quality gt 0 9 quality gt 0 1 Temporal Standard Deviation 01 02 03 04 05 06 0 7 08 09 1 Long Term Link Quality 0 12 14 16 18 0 b Probability of finding a high quality or c Temporal variation of link quality Each point medium to high quality neighbor depending on represents a directional node pair physical distance in our testbed Figure 1 Low power radio links in sensor networks exhibit inevitable fluctuations in their quality Approaches such as Solicitation based forwarding SOFA 10 remove the need for long term link estimation and test link availability by sending a short hand shake packet as a probe before sending any data packets However our evalua tion of STLE in Section 4 shows that a successful hand shake should not be taken as a success guarantee for subsequent data transmissions and indicate a need for more sophisticated models 3 Short term Link Estimation In this section we introduce short term link estimation in detail putting a special focus on its integration into
314. ty of a success ful packet transmission based on the average long term link quality and a short term history of consecutively transmitted packets The figure indicates that for a link with 10 long term link quality the transmission success probability for the next packet exceeds 80 when the four preceding packets were transmitted successfully We consider links of such instantaneous quality useful for routing thus STLE sug gests such a 10 quality link for routing In the same way STLE considers a 60 quality link to be short term reliable after just one successful transmission Figure 2 b depicts the probability of a successful packet transmission based on the average long term link quality and a Short term history of consecutively failed packet transmis sions It indicates that for two or more consecutive losses any link should be temporarily considered broken and be re moved from the routing table 5 Future Work and Challenges Although our evaluation shows that STLE can reliably identify when unstable links have become temporary stable and vice versa only an integration can fully evaluate the ben efits of STLE Thus we are currently implementing STLE extensions to the link estimator in the Beacon Vector Rout ing BVR protocol to analyze performance improvements of STLE Cerpa et al 5 discuss the impact of packet size on packet reception rate In our experiment all packets sent were of the same length The effec
315. tzen mu bei Sicherheits mechanismen insbesondere die Ressourcenbeschr nkung der Hardware ber cksichtigt werden In der wissenschaftlichen Literatur wurden in den letzten Jahren zahlreiche Sicherheitsmechanismen und protokolle f r drahtlose Sensornetze vorgeschlagen Hierzu geh ren Ver fahren zur Schl sselvereinbarung und verteilung zum au thentischem Broadcast zum sicherem Routing und zur si cheren Aggregation zur sicheren Lokalisierung zur Erken nung von DoS Angriffen und zur Detektion von geklonten Knoten Diese werden meistens jedoch unabh ngig von ei nem konkreten Einsatzszenario vorgeschlagen Ebenso wie sich die technischen Merkmale in obigen Applikationen un terscheiden gehen die vorgeschlagenen Sicherheitsmechanis men von ganz unterschiedlichen technischen Voraussetzun gen aus Einige Beispiele e TESLA 8 ein Verfahren zur Authentisierung von Broadcasts setzt synchronisierte Uhren auf den Sen sorknoten und einer Basisstation voraus e Das in 5 vorgeschlagene Verfahren zum sicheren Rou ting und zur Sperrung kompromittierter Knoten setzt unver nderliche Nachbarschaftsbeziehungen der Kno ten voraus d h Knoten m ssen statisch sein und es k nnen keine Knoten in sp teren Phasen hinzugef gt werden e Verfahren wie 4 die auf gegenseitiger Kontrolle der Sensorknoten beruhen setzen f r ein akzeptables Si cherheitsniveau voraus dass das Netz ber eine gro e Anzahl an
316. uch ihrem Leiter Prof Jochen Schiller Berlin im September 2008 Hartmut Ritter Bug Hunting in Sensor Network Applications Raimondas Sasnauskas J Agila Bitsch Link Muhammad Hamad Alizai and Klaus Wehrle Distributed Systems Group RWTH Aachen University Germany lastname cs rwth aachen de ABSTRACT Testing sensor network applications is an essential and a dif ficult task Due to their distributed and faulty nature severe resource constraints unobservable interactions and limited human interaction sensor networks make monitoring and debugging of applications strenuous and more challenging In this paper we present KleeNet a Klee based platform independent bug hunting tool for sensor network applica tions before deployment which can automatically test ap plications for all possible inputs and hence ensures memory safety for TinyOS based applications Upon finding a bug KleeNet generates a concrete test case with real input values identifying a specific error path in a program Additionally we show that KleeNet integrates well into TinyOS appli cation development life cycle with minimum manual effort making it easy for developers to test their applications 1 INTRODUCTION As with any software testing of wireless sensor network ap plications is an essential part of development life cycle The main goal of this engineering task remains the same finding and fixing program bugs as early as possibl
317. uf Eignung in diesen Szenarios pr fen Der Ansatz dieses Pro jekts soll hier pr sentiert und zur Diskussion gestellt wer den 1 EINF HRUNG Das Thema drahtlose Sensornetze erlebt seit einigen Jah ren einen rasant wachsendes Interesse in der Informatik Ur spr nglich wurden Sensornetze f r milit rische Zwecke kon zipiert heute werden Sie auch und vor allem f r zivile Ein satzzwecke betrachtet In der Vergangenheit waren tats ch lich realisierte Sensornetz Systeme eher prototypischer Na tur was darauf hindeutet dass noch viele praktische Proble me f r einen robusten Einsatz von solchen Netzen zu l sen sind Es ist jedoch abzusehen dass drahtlose Sensornetze in der Praxis zunehmend an Bedeutung gewinnen werden Ihre Flexibilit t macht sie attraktiv f r eine Vielzahl von Ein satzszenarios insbesondere dort wo Umgebungsdaten auf unkomplizierte und kosteng nstige Art und Weise erhoben werden m ssen Mit der zunehmenden Verbreitung von Sensornetz Appli kationen gewinnt das Thema Sicherheit in Sensornetzen an Bedeutung Das Stichwort Sicherheit umfasst zwei Aus pr gungen Zum einen ist damit die Ausfallsicherheit d h die Verf gbarkeit gemeint zum anderen die Informationssi cherheit also die Gew hrleistung von Integrit t Authenti zit t und Vertraulichkeit von bermittelten Informationen bzw der Schutz vor b sartigen Angriffen Die Umsetzung von Sicherheit ist in Sensornetzen mit beso
318. ulate output bounds for each node Using those bounds it is possible to compute the effective input for each node Thereafter the local per node delay bounds D for each sen sor node 2 can be calculated according to a basic network calculus result given in 9 D h a Bi gt 0 amp s lt 5 To compute the total information transfer delay D for a given sensor node 2 the per node delay bounds on the path P i to the sink need to be added D D jEP t Clearly the maximum information transfer delay in the sensor network can then be calculated as D maxj 1 n The whole procedure is called total flow analysis TFA be cause all of the traffic arriving at a given node is treated in an aggregate fashion Examples for the use of this calculus can be found e g in 10 11 12 A 1 Advanced Sensor Network Calculus While the TFA is a straightforward method for applying network calculus in the domain of wireless sensor networks there is room for improvement with respect to the quality of the calculated performance bounds This is due to the fact that the concatenation result for consecutive nodes offering service curves is not exploited by the TFA In particular we can exploit and even extend the concatenation result towards a so called Pay Multiplexing Only Once analysis PMOOA described in 13 to compute an end to end ser vice curve for a specific flow of interest from one sensor node to
319. und betrachten den optimalen Fall in dem Kollisionen Sendewiederholun gen und andere st rende Effekte ausgeschlossen werden Auf die Bit bertragungs physical und die Sicherungs Schicht data link wird bei der Modellierung verzichtet Jeder Knoten wird als Modul Abbildung 2 mit jeweils einer lokalen Variable modelliert die den jeweiligen Zu stand f r dle Forward und Sleep repr sentiert Um die Modellkomplexit t zu reduzieren wird auf eine explizi te Darstellung des Sender Empf ngermodells verzich tet Dennoch konnte f r Topologie 5 keine Erreichbar keit auf einem Rechner mit 32GB RAM wegen Spei cher berlauf berechnet werden Die Modellierung sieht f r die Kommunikation zwischen den Knoten das Set zen eines globalen Nachrichten Bits vor um so den Pa ketempfang zu betrachten Das so konstruierte Modell entspricht dem Konnektions graphen f r die Topologien wie sie Abbildung 1 zeigt bei einer Netzgr e von 14 Knoten w chst das Modell auf 124 10 Zust nde und 1 5 10 Transitionen Hierzu wird keine Sendereichweite definiert sondern die Konnektion aus der Topologievorgabe verwendet Ein Paketverlustmodell wird nicht betrachtet denn wie durch Experimente 2 gezeigt wurde l sst sich bei einer Sen deleistung von 3 99 10 1 mW 163 98 dBm eine Bit Error Ratio von 10 messen die f r die vorliegen den Experimente vernachl ssigbar klein ist 4 4 KSNSim Der JAVA basierte Karlsruhe Se
320. urce may cause errors or leave the hardware de vice in an undefined state To ensure uninterrupted function each atomic operation has to indicate this condition to the system kernel The system kernel needs to deactivate interrupt handling and preemptive scheduling accordingly When developing the device driver the developer has to decide which measures are appropriate for each operation 222 Managing the State of Virtual Devices This management of virtual devices and the state of their associated device entities provides fault tolerant resource arbitration and allows the application developer to use a com plex hardware platform without caring about shared resource or individual power management The HAL core functionality is completely hidden behind the virtual device API and thus transparent to the application We define the HAL core device management upon a state machine graph see Figure 3 of a virtual device with the fol lowing states and transitions a Active the device is powered on configured and ready to use A Inactive The device is still powered on and confi gured It is no longer in use and can be powered off or reactivated instantly Powered off The device hardware is powered off if supported Before a virtual device method is accessed by the API the device must be activated If it is inactive it can be reactivated immediately Otherwise the HAL ensures that all requirements are met and activates the device
321. ure 1 and Figure 2 present the normalized histogram of the absolute speed of the nodes when forwarding a packet The results of both figures show that slow moving nodes forward more traffic than faster nodes This behavior is independent of the protocol and the used metric The shape of the dis tribution results from the node speed distribution which is caused by the random waypoint mobility model 8 Furthermore the figures point out that the node speed distribution of the forwarding nodes can be modified by de ferring the routing information However a lower absolute node speed does not necessarily result in a higher end to end reliability If a cluster of nodes move in the same direction with similar the speed their absolute speed can be neglected in contrast to their relative speed 4 2 End To End Reliability Routing protocols are often rated by their achieved end to end reliability and delay Nonetheless the focus of this subsection lies on the reliability due to the fact that our testing scenarios revealed that the deference of the routing messages has no significant impact on the average end to end delay Figure 3 and 4 show the reliability of the different metrics and protocols depending on the maximum speed of 79 0 8 one Speed Estimation B Absolute Speed ar 0 6 Figure 1 AODV Node Speed Histogram the random waypoint model The results of Figure 3 in dicate that the RSE and th
322. usage among sensor nodes In 1 an algorithm for sending information from multi sources to multi sinks is presented It considers that sources address the same sinks protocol merges the paths that lead to the same sinks so as to avoid the replication of mes sages Nodes aggregate their readings with the ones they receive until the maximum capacity is reached Similarly to us they use the notion of paths and sinks Nevertheless in our scenario a source can potentially route its readings to an independent group of sinks Hence aggregation is not possible Additionally our approach tries to separate the trees created for each source in order to balance the load There exist some approaches 3 6 that address one to many communication in WSN but they do not take into account the existence of several trees In 3 its authors propose a feedback learning approach where nodes explore different possible routes and provide feedback to previous nodes It identifies lowest cost paths and alternates the use of these discovered paths in order to save energy It needs a setup phase In 6 an algorithm to trade off the power consumption and the delay created from sources to sinks is proposed It uses a hop count vector to support the routing decision 3 LOAD BALANCING ALGORITHM 3 1 Metrics When creating a routing protocol for a WSN scenario two different metrics can be optimized the delay transmission from sources to si
323. ux kernel protocol stack and channeled into the type writer device After trans mitting packets as payload of frames the process has to be done in reverse order at the destination host Each received frame has to be injected into the stack In contrast our pre ferred solution aims at the integration of the sensor node into the Linux kernel as a common network device By provid ing a new networking interface no kernel modifications are required and route discovery and packet forwarding features can be exploited Furthermore this approach is on par with kernel modules that drive USB host to host network cables or USB ethernet converters Adaptation of the Address Resolution Protocol ARP to our host to network layer and the deviant Maximum Trans mission Unit MTU of frames using the sensor node as net work device are remaining problems In addition we need to provide configuration programs alike the wireless tools in up to date Linux distributions These shall enable users e g to switch channels configure power settings set the data rate and scan for networks using a unified well known in terface In the long term the integration also offers the potential to run the upper layer sensor network protocol implementa tions via emulation or as native compiled binaries on the mesh router on top of WSKI This will simplify debugging and monitoring of the entire WSN As an outlook a port of the ScatterWeb OS as kernel module
324. vices MobiSys 2005 5 H Cha S Choi I Jung H Kim Shin J Yoo and Yoon Retos resilient expandable and threaded operating system for wireless sensor net works In Proc of the 6th intl conf on Information processing in sensor net works IPSN 07 2007 6 A Dunkels B Gronvall and T Voigt Contiki a lightweight and exible operating system for tiny networked sensors In Proceedings of the First IEEE Workshop on Embedded Networked Sensors 2004 7 K Walther and J Nolte A flexible scheduling framework for deeply embed ded systems In n Proc of 4th IEEE International Symposium on Embedded Computing 2007 8 TMote Sky Datasheet Moteiv Corperation http www sentilla com moteivtransition html 2006 39 Tab WoNS Calibration Approach for WSN based Ultrasound Localization Systems Extended Abstract Clemens M hlberger Chair of Computer Science V Am Hubland W rzburg Germany muehlberger informatik uni wuerzburg de ABSTRACT Accurate localization of objects is often very important thus various methods and systems exist Unfortunately some support easy deployment but are less accurate while others are more accurate but require a complex and time consuming deployment stage When the localization system meets some demands a widely self organizing deployment becomes possible This paper describes which pre conditions and abilities are required how self configuration could be realized fo
325. warnsysteme unter Verwendung eines drahtlosen Sensornetzwerks am Beispiel von Hangrutschungen Im Zuge des DFG Projektes GeoSens wurde ein Lokalisierungsalgorithmus f r drahtlose Sensornetzwerke entwickelt der auf geod tischen Ausgleichungsans tzen beruht und den Berechnungsaufwand sowie den Energieverbrauch auf ressourcenlimitierten Sensorknoten bei gleichzeitig signifikanter Erh hung der Positionierungsgenauigkeit deutlich reduziert 2 Drahtlose Geosensornetzwerke Reichenbach 2007 beschreibt ein Sensornetzwerk als eine Menge von Knoten die ber eine bestimmte Fl che platziert werden und physikalische Daten eines Ph nomens von Interesse Alexander Born Professur f r Geod sie und Geoinformatik Universit t Rostock Justus von Liebig Weg 6 18059 Rostock 49 381 498 3210 alexander born uni rostock de messen Auf einem Knoten sind Detektoren installiert deren Signale stellvertretend f r eine Messgr e stehen Jeder Knoten ist durch einen ressourcenlimitierten Controller mit beschr nkter Speicherkapazit t Prozessorleistung und Kommunikationseinheit auslesbar Bill et al 2008 Erfolgt die Kommunikation zwischen den Knoten funkbasiert wobei Messdaten ber die direkten Nachbarn bis zu einer Datensenke bertragen werden k nnen spricht man von einem drahtlosen Sensornetzwerk DSN engl Wireless Sensor Network WSN siehe hierzu Akyildiz et al 2002 Ein Geosensornetzwerk GSN als spezielle Auspr gung eines DSN
326. wer delay bound under LFF It is interesting to notice the large amount of flows that got a significant improvement More detailed analysis shows that especially the longest flows get a boost Figure 4 shows a comparison between the worst case bounds per flow length The topology had data flows that passed 1 to 4 hops Under arbitrary multiplexing the maximum de lay bound for a given hop count increases with the number of hops Under LFF however the longest flows have a very low bound This is clearly overcompensation for the origi nal penalty incurred by longer flows This will be further discussed below Calculating the overall upper response time bounds for 100 randomly generated networks with over 15000 flows in total and with a topology as described above resulted in improvements ranging from 0 no improvement at all to Frequency 20 30 40 10 0 00 0 05 0 10 0 15 Relative improvement of the delay bound Figure 5 Improvement of the delay bound by LFF over arbitrary multiplexing 15 7 with a mean of 9 The distribution of the improve ments is shown in Figure 5 4 DISCUSSION We have shown that scheduling long flows before shorter flows often yields an improvement of the worst case response time at the cost of minimal data and processing overhead The observed improvement with an average of 9 may seem small but earlier results showed benefits of over 30 in some cases when scaling was taken
327. wever when the number of sources is close to its max imum load indicator is active the cost due to the messages sent also depends on the high load indicator and we The cost in this case must be much higher because such nodes should not be chosen as next nodes thus we must be much larger than ws The number of paths of a node is also related to the number of sources the node is serving as for each source there is at least one path Since it only gives an insight of how loaded the next nodes could be on their way to the sinks a small weight must be assigned to this parameter The weight of each parameter w1 we can be either pre programmed on the nodes or established by a sink Because of this the routing in the network can be adapted to the needs of the users For instance if the delay is the metric to be minimized w must be the prevailing weight If on the contrary it is fundamental to maintain an even load in the network ws must have a significant weight When trying to avoid contention problems we gains importance 4 EVALUATION Network lifetime is the most important criterion in sensor networks We define network lifetime as the time until some percentage of the nodes die possibly causing the partition of the network In order to maximize the network lifetime two conditions must be fulfilled the total amount of energy spent in the network must be minimized and the difference between the individual node en
328. y establish TCP connections when data has to be transmitted This results in an Overlay Network established on demand On the other hand reliable IP based multicast using UDP could be even more efficient and energy consumption friendly but requires more cross layer interactions and higher implementation effort 4 OUTLOOK There are still many open questions regarding the design of multicast support in WSNs When using reliability on end to end basis problems such as acknowledgment implosion handling of negative acknowledgments etc have to be ana lyzed in more detail Generally also congestion control and the resulting limitations have to be considered Multicast ing on the link layer could also improve the performance in combination with our presented approaches determine efficiency energy consumption and overall performance we plan to simulate the different scenarios and solutions in the OMNET simulator 12 Both approaches should be im plemented on real sensor nodes using Contiki 13 as oper ating system 12 5 REFERENCES 1 G Wagenknecht M Anwander T Braun T Staub J Matheka S Morgenthaler MARWIS A Management Architecture for Heterogeneous Wireless Sensor Networks International Conference on Wired Wireless Internet Communications WWIC 08 Tampere Finland May 2008 2 Anwander Wagenknecht T Braun Management of Wireless Sensor Networks using TCP IP International Worksh
329. y even require manual code modifications for testing the code One of our major design objectives is to provide an easy to use bug finding tool for sensor network applications which is strongly integrated in the software development life cycle For this purpose we use grammar based automatic code instrumentation We extend ANTLR 13 based GNU C grammar to automatically insert symbolic annotations i e to mark the memory locations to be checked by Klee in the C source code The user only needs to provide a high level configuration stating the variable names that has to be checked inside the code However providing a configura tion to insert annotations in the code to perform additional checks is optional as our solution performs some built in checks to detect common bugs in sensor network applica tions For example struct type cast checking discussed in section 3 3 and memory checks on received packet buffers 3 3 Struct type checking Type conversions in C using type casts is a very common practice and definitely not an error But since type safety is not guaranteed programmers can interpret each memory region to be of any type Especially the casts between point ers to different structure types make the code maintenance difficult 3 14 One of the main objective of sensor network applications is to collect and process data from the sensors The received data is at first available only as an untyped bit stream Af terward

URL - of /pub - Freie Universität Berlin

Contents

Download Pdf Manuals

Related Search

Related Contents

URL - of /pub - Freie Universit&auml;t Berlin

Contents

Download Pdf Manuals

Related Search

Related Contents

URL - of /pub - Freie Universität Berlin