Home

Evaluation of Accelerometers Mounted on Wireless Sensor Motes

1. iel 0 99 0 96 0 97 0 98 1 01 1 02 1 03 1 04 x axis 0 8 0 6 0 4 0 2 po i 0 99 1 y axis i i i i i i i 0 96 0 97 0 98 1 01 102 1 03 1 04 0 8 206 0 4 0 2 pe 1 z axis 0 98 0 985 0 99 0 995 1 005 1 01 1 015 1 02 Figure 7 Ratio CDF plot of mote 7 100 minutes Freescale accelerometer page 18 0 6 0 4 Acceleration ADC units 0 8 x axis y axis z axis 0 2 0 4 0 6 0 8 1 Time seconds 1 2 Figure 8 Three straight lines fitted to the 5 hour experiment data All lines have been moved to the common origin to emphasize the drift The RMS errors are 1 29 1 89 and 1 11 for the x y and z axis respectively F x F x 1 01 Empirical CDF 1 T T 0 8 J 0 6 7 0 4 E 0 2 4 0 l 1 I L L I L L L 0 99 0 992 0 994 0 996 0 998 1 1 002 1 004 1 006 1 008 x axis Empirical CDF 1 T T 0 8 7 0 67 al 0 4r 4 0 2 4 1 L L L 1 1 L L 1 0 99 0 992 0 994 0 996 0 998 1 1 002 1 004 1 006 1 008 y axis 1 01 Figure 9 Ratio CDF plot of 5 hours measuremements MicaZ sensorboard page 19 However if a line is fitted to the measured data as seen in Figure 10 the drift is higher than that observed with the Freescale accelerometer Calculating the RMS error for the two fits one get 0 4 and 0 3 for the
2. e z t M oM Md bk bl obl h 8J82soo1j inis Snnt i im s Figure 2 Freescale MC13192SARD wireless sensor mote 1 One of the devices used in this project is the Freescale MC13192SARD 1 mote shown in Figure 2 which is based upon the Freescale EVB13192 2 The core of the device is the Freescale MC9S08GT60 3 Micro Controller Unit which has the following key features e 0 40 MHz HCS08 8 bit CPU e 4K RAM e 60K FLASH e 2 two channel 16 bit timers e 8 channel 10 bit ADC e 2 Serial Communications Interface modules SCI e 1 Serial Peripheral Interface module SPI Furthermore the MCU has several power saving modes which make it possible to reduce power consumption by shutting down internal subsystems and changing operating frequency e g power consumption when running at 16 MHz can vary from 6 5 mA to 25 675 nA Also the 8 channel ADC is in fact an 8 channel multiplexer equipped with cross talk preventing circuitry with a maximal sampling rate of 14 0 psec Connected to the SPI module is the Freescale MC13192 transceiver 4 which is a low power short range 2 4 GHz ISM band radio with the following key features e IEEE 802 15 4 compliant ZigBee e 16 channels 5 MHz intervals Sensor Application Reference Design Evaluation Board Dedicated circuitry to avoid signals from one channel to interfere with another page 6 e 250 kbps transfer rate Power consumptions range between
3. Figure 15 is a plot of measurement data taken from the short track and by looking at the z axis it is possible to count the number of steps by counting the number of large peaks The increased acceleration is caused by the fictitious force measured by the mote being directed towards the Earth s gravitational field as explained in Section 3 2 It can be seen from the figure that exactly five steps were taken which corresponds to the number of strips of adhesive tape on the track By calculating the time difference between each peak it is also possible to deduce the duration of each step By defining the middle of each step to be the acceleration turning point when the force applied to the mote is exactly equal to and opposite the gravitational acceleration Table 5 shows a summary of each step duration No Middle lt s gt Duration lt s gt 1 3 780 2 4 440 0 660 3 5 069 0 629 4 5 688 0 619 5 6 277 0 589 Table 5 Step duration 19These defects were probably caused by transmissions errors either between the radios or the serial connection since no error control were present page 22 Figure 14 Long track 15 m long and with stickers every 60 cm page 23 Acceleration m s 4c wo 4 2 3 4 5 6 7 8 9 Time Seconds y axis Acceleration m s Time Seconds z axis Acceleration m s Time Seconds Figure 15 Walking experi
4. In comparison the ADC can be sampled every 14 us Depending on the header size and the number of axes measured a typical TinyOS packet payload can contain 3 and 4 samples for the SARD and MicaZ respectively Since the transmission time for a packet payload is at least 6 ms at the network level the maximum theoretical number of samples transmitted within a second would be 498 and 664 respectively Because the transmission speed at the communication level is higher than at the network level it is reasonable to assume that the communication level can sustain a similar sampling rate At the sensor level the maximum sampling rate would depend on the number of sensor motes present and in contest of the communication level In a multi mote environment the overall sampling rate should be divided among the motes since they all share the same communication level Since overhead and busy channels have not been included in the above discussion it is reasonable to believe that the 498 Hz and 664 Hz sampling rate should be regarded as a theoretical upper bound and will not be obtainable in reality 3 4 3 Transmit This program resides on the motes carrying accelerometers and samples each measurement when a timer is fired The timer is controlled by a predefined sampling rate and each axis is sampled one at a time The result is stored in a buffer together with a timestamp which is the lowest 16 bit of the internal counter On the Freescale motes this corres
5. In order for this algorithm to be effective it is necessary to do the timestamping as close to the physical layer as possible or else the time difference t t will be neglible Since the radio transmission bandwidth is limited and mostly saturated with acceleration measurements it would be infeasible to do a sensor mote to gateway RTT calculation for each and every mote This approach would also have the defect of not being scalable to larger networks Instead of calculating the exact RTT for each mote we use an approximation of the RTT as proposed in 19 The principle is called Reference Broadcast and is based on the same fact as the time beacon protocol that broadcasts arrive close to simultaneously at all the sensor motes in communication range By appointing a master mote to initiate a RTT calculation with the gateway transmitted as a broadcast the other motes can use the RTT request and reply to estimate the RTT This approach is different than that proposed in 19 since both messages are transmitted as broadcasts and hence each mote in the network is capable of calculating the RTT Also since the message flow is highly asymmetrical the gateway should have a lower reply latency than the sensor motes because the former only receives messages while the latter transmit them 2Since this is close to the speed of light traveling in the atmosphere the time delay caused by propagation for motes placed within meters of each other will be
6. build evb13192 directory The executable file can then be uploaded with the command hc08s exe com port number com speed file 2 4 3 Communicating over RS 232 As mentioned above the EVB13192 is equipped with both a RS 232 port and a USB port but because the USB connection can serve as a power supply as well all work at DIKU has been done under the assumption that this port would be the preferred communications port The former HPLUARTO module located in the contrib diku evb13192 tos platform hcs08 directory was the primary UART communication module and resided at the lowest level such that all other applications in need of UART communication were build on top of this This module was connected directly to the USB port via the SCI2 module Since the RS 232 port is connected via the SCI1 module the most consistent way to implement RS 232 connectivity is to create two new modules HPLUART1 and HPLUART2 where the HPLUART2 module is simply the old HPLUARTO module renamed and the HPLUART1 module has had all communication diverted to the SCI1 module instead of the SCI2 module Thus it becomes possible to communicate with the RS 232 port on the MC13192SARD by using the HPLUART1 module and to communicate with both the USB and RS 232 port simultaneously on the EVB13192 8The default value for TOS LOCAL ADDRESS is 1 The execution of this command must be combined with a reset of the MC13192SARD The sequential order of this pr
7. 0 7 y down 549 0 5 y down 583 0 5 y down 570 0 5 y down 588 0 3 Table 4 Calibration data for Crossbow sensor boards It can be seen from Table 3 and Table 4 that the measuring range of g g is of the order of 500 and 100 ADC units respectively for the Freescale and Crossbow motes This results in a resolution of 0 0393 ES and 0 196 pe wis respectively There are however several problems with this calibration method First of all it is based on the assumption that all three accelerometer axes have been perfectly aligned with the gravitational acceleration If this is not the case the measured value will be smaller than the actual g since only a projection of the gravitational acceleration is measured and subsequently all absolute accelerations will thus be larger than their real values Second the calibration is only based on three reference points spanning an interval of g hence one should not expect the calibration to be valid far from this interval 16By using the formula at http www npl co uk mass faqs gravity html with the latitude 52 211048 and height 24 m above sea level These values were taken for the William Gates building with the help of Google Earth http earth google com 17 Syst me International d Unit s International System of Units http www bipm fr en si page 15 Calibration for Freescale mote 6 900 T T T T T f x calib txt 3 A
8. 0 8r 206 0 4 0 2 0 96 0 97 0 98 L 0 99 1 1 01 y axis L fi 1 02 1 03 1 04 0 98 1 02 Figure 4 Ratio CDF plot of 5 hours measurements Freescale accelerometer Empirical CDF mote 5 1 T T I r 0 8 g 08 0 4 0 2 i i i i Fr i j i i i 0 96 0 97 0 98 0 99 1 1 01 1 02 1 03 1 04 x axis ir T T T 1 cal 0 8 09r J E gal 4 026 El i i p i i i i i 0 95 0 96 0 97 0 98 0 99 1 1 01 1 02 1 03 1 04 1 05 y axis 1 T T T 0 81 08 J aL J 0 2 4 098 0 985 0 99 0 995 1 1005 101 1015 z axis Figure 5 Ratio CDF plot of mote 5 100 minutes Freescale accelerometer page 17 Empirical CDF mote 6 1 T T T I T T E 0 8r 4 0 6r 4 04 J 0 2 4 0 97 0 98 0 99 1 01 1 02 1 03 1 04 x axis 1 T T n 0 8r 0 6r j 04 4 0 27 A i i nd i i i i 0 96 0 97 0 98 0 99 1 01 1 02 1 03 1 04 y axis 1 T T p T T 0 8r 4 0 6r 4 04H J 02r J i i i i i i i 0 985 0 99 0 995 1 1 005 1 01 1 015 z axis Figure 6 Ratio CDF plot of mote 6 100 minutes Freescale accelerometer Empirical CDF mote 7 1 T T T 0 8 0 6 0 4 0 2
9. 34 37 mA when transmitting receiving and 500 uA in idle mode Two on board antennas are used for sending and receiving respectively Connected to three of the ADC ports are the Freescale MMA6261Q dual axis accelerometer 5 and the Freescale MMA1260D single axis accelerometer 6 Both accelerometers have a 1 5 g range of operation but the MMA6261Q has a noise and sensitivity rating of 3 5 mV and 800 mV g respectively where the MMA1260D is rated as 5 0 mV and 1200 mV g respectively This means that the single axis accelerometer has a higher sensi tivity but with a lower precision Besides these three key parts the MC13192SARD also has 4 LEDs 4 push buttons and 1 reset button The device can be powered either by a 9 V battery or a 3 9 V DC power supply Programming can be done either via the BDM interface Background Debug Module or via the 9 pin RS 232 port which is connected to the SCI module SCI1 on the MCU This port can also be used as a regular communications port 2 1 TinyOS Since the key constraints when developing a wireless sensor networks WSN application are the limited resources on each device such as processing power storage capability and electrical power it is important that the operating system in use honours these constraints TinyOS is an open source operating system built specifically for WSN 8 It uses an event driven execution model and utilises a component based architecture which makes it easy to move the
10. 5r 7 D a 10L 4 15 L L L 1 L 1 0 5 10 15 20 25 30 Time seconds 35 Figure 18 Velocity and position of elevator car with g 9 81258 m s Velocity 0 5 Velocity m s 5 a T L 15 20 Time seconds Position 35 4L Distance meters I 15 20 Time seconds 25 30 35 Figure 19 Velocity and position of elevator car with g 9 7942 m s page 28 which must be removed in order to calculate the travelled distance It remains to be seen if the axes orthogonal to the Earth s gravity field might be less sensitive to this calibration error In the light of the last two experiments we were discouraged to perform the same set of experiments with the MicaZ sensorboards since rotation would have the same negative effect on these devices and the five time lower resolution would have a worse impact on the calibration and subsequently on the numerical integration A reference to the raw data files can be found in Appendix B 3 6 Future Work As seen by the huge differences in the results obtained from walking and driving the elevator the need to com pensate for rotation of the accelerometers is of paramount importance Without the assistance of gyroscopes to adjust for these rotations of the local reference frame it is futile to use accelerometers to achieve a position from the acceleration For future work this integration of gyros
11. D a t alo g is k lI n s t i t u t Evaluation of Accelerometers Mounted on Wireless Sensor Motes Marcus Chang Joint technical report with Cambridge University Technical Report no 06 02 ISSN 0107 8283 Dept of Computer Science University of Copenhagen e Universitetsparken 1 DK 2100 Copenhagen e Denmark Evaluation of Accelerometers Mounted on Wireless Sensor Motes Marcus Chang Department of Computer Science University of Copenhagen Denmark marcus diku dk Abstract In this project we investigate the properties of accelerometers mounted on wireless sensor devices with the purpose of mounting wireless sensors on athletes to facilitate performance measurement To this end a hetero genious data collection architecture with a system wide sampling rate of 200 Hz was developed to facilitate the evaluation A variant of the Reference Broadcast Synchronization protocol was implemented and without MAC layer timestamping we achieved an accuracy of 4 ms for a WSN with a single gateway and 27 ms for a WSN with two gateways connected to the same time source The subsequent evaluation showed that in order to use acceleration measurements to calculate the velocity and position a gyroscope is crucial in order to compensate for rotation and a calibration within 0 1 accuracy is necessary as well With a non rotating accelerometer we were able to establish the position of an elevator with an accuracy within 10 keywords wireless senso
12. Gd eve dem te 14 3 54 Evaluation t ate autant fe iE ala REOR dd ve ts dede is E eR 14 35 1 Calibration sop o A X RUERUE u REGRESAR IS wel 14 32 Zero moton 5 ue IR A AO ood DNO RU A WO dE 16 315 9 POWE o WE Sev Re Vox qe ee wy Cera RN ebd wx ewe 20 39 4 Walking e ERR hon eae SIR eh ke CR tC ae oi ae ng ond 22 3 525 Elevator ei oux osi pec Redde eorom ves en ee Be tree Ee eode iei eub Bd 26 3 6 R t re Works eta god e els amus Sa SAN ae AS 29 3 7 Summary 26 RA UNIUERSIS thie De T A Pa Wee A KG eee 29 4 Time Synchronization 30 41 Related Work metra 2 24 8 448 AE Ee Bee OER AS bee EEE ER Se Ee dd 30 2 2 IRIOLOCOIS s acd ener Tee AN eae Rewed Pe ur Tb DUC EC ex Bad ded 30 42 1 Time Beacon he GRAECE UR RYDER V eq 31 4 2 2 Reference Broadcast pins a s on ZA Seti a Wo a 31 4 3 ImplementationIssues SD RO SO AD bok ree 32 4 Evaluation amp 5 obe tik rd sss A dtes s ROME MS s be A Bd ese AE 33 44 1 Round Trip Time 3 8 spay XOU quen me PRU Pee M Ree e eu 33 442 Clock Drift uh hi e uS reU Giga cies UR NIC Rae Rae n RU D paged me 33 4 4 8 Agreement i Lis hoe lu dors von RUD RSS Eom e RR GR Sar hte RO RR UE Eos 36 7 5 Future Work v a Ape be e ERO NACER m xeu wx REN ves 38 4 6 SUMA i A eC test Qe b er Ce RN We Ru be 40 5 Conclusion 41 References 42 A Application Source Code 44 B Acceleration Data Files 44 C Synchronization Data Files 44 page 4 1 Introduction This report is t
13. T T T T 5 3 5000F 7 O 0 I il j l 1 0 1 2 3 4 5 6 7 8 9 10 Offset milliseconds Slave mote 6 10000 T T T T T T t a 5000F 7 O 0 1 l l 1 0 1 2 3 4 5 6 7 8 9 10 Offset milliseconds Slave mote 7 10000 T T T T T T E 3 5000F 7 5 0 1 l l 1 I 0 1 2 3 4 5 6 7 8 9 10 Offset milliseconds Slave mote 8 10000 T T T T T T E 3 5000F 7 5 0 I I 1 I I 1 0 1 2 3 4 5 6 7 8 9 10 Offset milliseconds Figure 21 Histogram of 10 000 Round Trip Time calculations The histogram for the master mote is the full RTT while the ones for the slave motes are the partial RTT measured transmission time which can thus be disregarded The cause of this systematic drift must be the lack of precision of the PC clock caused by the differences in process scheduling by the operating system on the PC By looking at the relative differences between the master sensor mote s clock and the clocks on the other motes it is possible to remove the PC s clock as a source of error and hence measure the relative clock drift Since the timestamps taken from each mote represents a discrete function evaluated at different points in time they are not directly compatible The timestamps used for the relative offset calculations were found by interpolating the timestamps from the three slave motes These three continously functions were subsequently evaluated at the same points in time as the master mote in effect creating a new
14. boundary between dedicated hardware and software emulation and save power by shutting down unused components Also this component based architecture minimizes the overall program size since only essential components are included The TinyOS core itself only requires 400 bytes of code and data memory combined All these features make TinyOS a much preferred operating system for wireless embedded sensor networks and are therefore available on a wide range of platforms in particular the MicaZ motes from Crossbow 7 However the support for TinyOS on the Freescale EVB13192 is a work in progress currently being carried out at the Department of Computer Science at the University of Copenhagen DIKU The current build is available from CVS at SourceForge under cvs tinyos tinyos 1 x contrib diku Since the MCI3192S ARD is a platform in itself it is only appropriate to have the same support in TinyOS as for the EVB13192 Therefore the work in the following sections has been integrated with the EVB13192 code tree at SourceForge as of October 12th 2005 so the MC13192SARD will benefit from any improvements made to the EVB13192 To compile to the MC13192S ARD platform the compile command make dig536 can be used once inside the contrib diku evb13192 apps application directory The current build re quires direct access to the internet and cur 17 must be installed since the compiling and linking are done remotely on a server located at DIKU This is b
15. must be used This could be a GPS clock connected to either the PC or gateway since many of these devices can operate with nanosecond accuracy Another improvement would be a rotation scheme among the motes to remove the single point of failure from the master mote This scheme could also be responsible to elect a new master mote in the case of a failure Algorithms for the rotation and election scheme have been studied extensively and protocols such as the Bully algorithm 23 could be used to elect at new master 4 6 Summary A 2 tier external time synchronization protocol was implemented on the Freescale SARD First the gateway were synchronized with the connected PC by Christians method Second the sensor motes were synchronized with the gateway with a modified version of the Reference Broadcast Synchronization The protocol were adapted to the highly asymmetrical and continuous data flow in the previous developed acceleration collection architecture resulting in a minimal amount of packet overhead caused by the time synchronization With application level timestamping and a clock resolution of 1 ms subsequent experiments showed the synchro nization between the sensor motes and a single gateway to be accurate within 4 ms Further experiments also found the accuracy between different WSN connected to the same PC to be 27 ms This increase is primarily caused by the inaccurate reading of the PC clock since experiments showed the drift of this clock
16. set of compatible discrete functions The relative offsets can be seen in Figure 23 Compared to the former figure the offset is now of the same order of magnitude as the variance of the transmission time which seems more reasonable since this is of the same order as the expected achievable precision It is also clearly seen that mote 7 has a linear drift compared to the other motes which can thus be compensated for In order to maintain a precision similar to the variance of the transmission time 1 2 ms it is thus necessary to resynchronize the sensor motes at least every 2 min 120000 ms Similarly an accuracy of 4 ms can be maintained by resynchronizing at least every 4 min A reference to both data collections can be found in Appendix C 26This figure is taken as an estimate based on Figure 23 and has the noise taken into account as well page 34 400 T 200 200 Offset milliseconds 400 600 mote 8 800 I i l 0 0 5 1 15 2 2 5 3 Time milliseconds x 10 Figure 22 Evolution of time differences between the PC clock and the four sensor motes clocks after one synchronization m Wr id AMAN MA UNT VASA E Eb J Time milliseconds x 10 Figure 23 Evolution of time difference between the master mote and the three slave motes after one synchronization Each graph represents the time difference between a
17. slave mote and the master All three graphs have been smoothed with a moving average with a window size of 100 page 35 4 4 3 Agreement The most important reason to perform time synchronization in a WSN is the need to be able to combine measure ments taken from several different sensor motes without losing the temporal dimension This can be evaluated by comparing measurements taken from different sensor motes monitoring the same event In this case the four SARD motes were strapped to a metal beam and with the motes synchronizing every 10 s the beam were shaken by turn in all three directions parallel to the accelerometers axes In order for the time synchronization to be successful the measurements must be in agreement with each other However since the magnitude of the acceleration is depen dent on the calibration of each sensor one should not expect the motes to agree on the size of the acceleration but only on when it occurred e g peaks in the acceleration corresponding to the extreme points of the beam s position can be compared directly Figure 24 shows the acceleration measurements sampled from the four sensor motes The three rectangular boxes mark the movement of the beam parallel to the axis in question and a close up of each box can be seen in Figure 25 By looking at the time difference between the first and last mote crossing the zero acceleration line one can estimate the accuracy of the time synchronization to be of the orde
18. the mote on an electrical model train moving backworth and forworth on a fixed track and subsequently measure the differences in the acceleration pattern Finally as seen in Section 3 5 5 the measured accelerations were not precise enough to be used for position estimation by numerical integration Future experiments should be done with accelerometers with both a higher sensitivity and a lower noise density than the ones used in this experiment Preferably the accelerometer should be capable of measuring the gravitational acceleration with a precision of at least two orders of magnitude higher than the desired precision on the position 3 7 Summary A set of TinyOS applications was developed to facilitate sampling of measurements and even with room for several optimizations we achieved a system wide sampling rate of 200 Hz Calibration measurements were collected for the different accelerometers and an upper bound on the precision was found to be 0 0393 m s and 0 196 m s for the SARD and MicaZ mote respectively The zero motion drift was measured on both the MicaZ mote as well as the SARD mote and we found that only the MicaZ mote had detectable drift We also found the power supply to be a source of error if the voltage drops below a certain point Several experiments were performed by walking on a straight line carrying accelerometers and although it was possible to deduce the step duration and the number of steps taken the experiment especial
19. to be of this order of magnitude The relative time difference between the sensor motes were seen to drift and the difference became comparable with the synchronization accuracy of 4 msec after a period of 4 min The network thus has to be resynchronized at least this often in order to maintain the accuracy page 40 5 Conclusion The main goal of this project was to investigate the precision of the physical properties obtained from acceler ation measurements specifically those obtained from accelerometers mounted on wireless sensor devices Two different types of devices were used in order to accomplish this task the Crossbow MicaZ and the Freescale MC13192SARD In order to develop TinyOS applications to both platforms several modifications were made to the motes and the TinyOS contribution from DIKU Specifically the embedded bootloader was changed to one compatible with executables generated with TinyOS the component responsible for UART communication was changed to use the RS 232 port instead of the USB port and finally three components were developed to sample the ADC ports where the accelerometers were connected These components were subsequently used to develop a data collection architecture capable of collecting measurements from both hardware platforms and we achieved a system wide sampling rate of 200 Hz Subsequently the data collecting architecture was augmented with a 2 tier external time synchronization protocol on the Fr
20. to the environment in order to help improve performances One of the physical properties we intend to measure is acceleration With this information it is possible to calculate physical parameters such as velocity position step length and frequency Also with multiple sensors it will even be possible to measure these properties for the body as a whole as well as for each individual body part 1 2 Purpose The key issue we want to understand is the precision of the physical properties obtained from acceleration mea surements This includes the drift of the sensors over time and of course the growth of error over time The main purpose of this project is thus to study and evaluate accelerometers mounted on a wireless sensor device In order to perform this evaluation we had to develop a data collection infrastructure This report consists of three parts which can be read individually Section 2 describes the process of making the Freescale MC13192SARD motes ready for TinyOS and how to use the accelerometers with TinyOS Section 3 consist of the actual development of the data collection infrastructure and the subsequent evaluation of the accelerometers Finally Section 4 consist of the implementation of a time synchronization protocol which is essential when several motes are measuring the same events Cooperating Embedded Systems for Exploration and Control using Wireless Sensor Actuator Networks page 5 2 Freescale MC13192SARD
21. were chosen as to minimize the end velocity as much as possible Experiment g m s max velocity lt m s gt 1st lt m gt 2nd lt m gt 1 9 8019 1 0 6 9 2 9 8026 1 0 7 1 3 9 7967 1 0 3 9 7 7 4 9 7942 1 0 3 6 7 2 Table 7 Elevator experiments With a new g 9 800 0 005 m s the travelled distance of the elevator car was estimated to be 7 2 0 5 m According to these experiments performing an accurate calibration is of high importance since even when the calibration is off by as much as 0 1 the results can change from being useless to fairly accurate However since the resolution of the Freescale accelerometer when used in conjunction with the on board ADC is 0 0393 with the general noise level being of the order of 1 2 ADC units it is not possible to achieve the high s2 lt ADCunits gt f accuracy calibration that is required The problem described above is mainly caused by the movement being parallel to the gravitational acceleration 21 This noise filter is better to preserve the area under the graph than a regular moving average A mathematical definition can be found at http www mathworks com access helpdesk help toolbox curvefit ch data8 html page 27 Velocity 0 5 0 Velocity m s 5 a T 15 1 1 1 0 5 10 15 20 25 30 35 Time seconds Position 5 T T 0 El 3 o E a PEN EALA 8
22. z axis 1000 800 F 600 F 200 l 5000 I 10000 15000 Time Seconds Figure 12 Freescale mote running out of power page 21 3 5 4 Walking After these preliminary experiments we measure the actual acceleration of a person walking A Freescale mote was mounted inside a plastic case and the case was attached to the chest of the test subject The most simple experiment would be to walk a straight line Two tracks were set up by sticking adhesive tape on the floor perpendicular to the walking direction as seen in Figure 13 and Figure 14 The first track was 2 5 m long and the adhesive tape were placed with 50 cm intervals The second track were 15 m long and had adhesive tape placed with 60 cm intervals The purpose of the adhesive tape was to ensure a fixed step length and hence the two tracks were 5 and 25 steps long respectively Figure 13 Short track 2 5 m long and with stickers every 50 cm Several tests were conducted on the two tracks and Figure 15 and Figure 16 show some typical measurements Before and after each walk several seconds of stationary measurements were sampled in order to establish a clear point of reference for the orientation of the mote This data was subsequently used to rotate the local reference frame as mentioned in Section 3 2 The measurements were also filtered for any corrupted data by enforcing that time is moving forward and by removing unrealistic acceleration e g several g s
23. 4000 5000 6000 7000 8000 9000 10000 Time milliseconds Figure 28 Close up of each of the axes as marked with rectangular boxes in Figure 27 y axis 0 02 T T 0 01 4 a o E S 36495 3 ms 36521 9 ms OF lt gt zi 5 o 8 lt 0 01 H 4 0 02 I l 3 649 3 65 3 651 3 652 3 653 Time milliseconds x 10 Figure 29 Close up of the y axis as marked by the rectangular box in Figure 28 The arrow represents the magnitude of the disagreement between the four motes when measuring the same event As indi cated by the readings the accuracy of the time synchronization is of the order of 27 ms when using two gateways page 39 Freescale MC13192 radio MAC layer timestamping combined with increased clock resolution should greatly improve accuracy The current implementation assumes that all the motes master slave s and gateway are within communication range This is however not always the case and the need for a multi hop adaption is necessary Similarly it would be advantageous to have multiple gateways collecting data in unison It would then be necessary to be able to distinguish between synchronization replies transmitted from different gateways in order to calculate the exact transmission time As seen in Figure 22 the PC clock is the largest source of error when an external synchronization is performed If accuracy in the microsecond range is the goal a more precise external clock
24. DC units 100 1 1 1 1 1 15 10 E 0 5 10 T5 m s 2 Figure 3 Calibration curve A more precise calibration could be achieved by inducing a constant acceleration to the sensors of different magni tude so more reference points can be collected This could be done by placing the sensors on the rim of a rotating disk and by varying the rotational speed different magnitudes of acceleration can be measured 3 5 2 Zero motion Drift The zero motion drift is the change of acceleration measured from a stationary mote This kind of drift usually manifests itself over a long period of time and can primarily be attributed to temperature changes Since the accelerometer s temperature rises after prolonged use it is expected that the measured acceleration would change for a stationary mote and if one knew the exact temperature it would be possible to compensate for this drift A measurement of this zero motion drift was made for three of the Freescale motes and one of the Crossbow sensor boards First one Freescale mote was set to run for five hours uninterrupted For each axis the median of the first 1000 measurements were used as reference value in calculating the ratio for all the measurements Rati 0 Ameasured Gre ference The CDF plot of the ratio for each axis can be seen in Figure 4 The experiment was then repeated with three Freescale motes this time only for 100 minutes though The plots can be seen in Figure 5 Figure 6 and
25. Figure 7 The figures show that over 90 of all the measurements lie between 1 of the reference value For the five hour long measurement data a straight line was fitted to each axis and the RMS error was calculated to be 1 29 1 89 and 1 11 for the x y and z axis respectively all in ADC units As it can be seen in Figure 8 the range of acceleration values spanned by the three lines are less than that of the RMS error for each axis respectively Since the magnitude of this apparent drift is smaller than the precision of the measurements it is not possible to measure any drift in the Freescale accelerometers as expected since the Freescale accelerometers have temperature correction circuitry Similar experiments were conducted with one Crossbow sensor boards and the corresponding ratio CDF plots for the five hour long experiment can be seen in Figure 9 As seen with the Freescale accelerometers over 90 of the measurements are within 1 of the reference value In fact the noise is so low on the MicaZ accelerometer that one can say over 90 of the measurements are within 0 4 of the reference value 8This only applies for the accelerometers on the MicaZ motes however since the accelerometers on the SARD motes have built in circuitry to compensate for temperature drift page 16 Empirical CDF mote 5 hour 0 8 gt 0 6 0 4 0 2 P 0 97 0 98 0 99 1 1 01 x axis 1 02
26. PC a gateway program connect ing the sensor motes with the PC and the actual application on the sensor motes The data communication in this setup is highly asymmetrical with the sensor motes transmitting data to the gateway which in turn transmits data to the PC The gateway is in one hop communication range with all the motes This also means that the gateway is a common reference point suitable for the other motes to synchronize with Since the gateway also is permanently connected to the PC this connection can be used to synchronize the gateway with the local time on the PC as well page 30 4 2 1 Time Beacon If only time synchronization between the individual sensor motes is required e g there is no need to associate each measurement with real time a simple time beacon protocol can be used Since all the motes are in direct communication range with the gateway a packet containing a timestamp broadcasted from the gateway will reach all the sensor motes simultaneously if one disregards the propagation time of electromagnetic waves Each sensor mote can thus calculate an offset from the motes local time to the timestamp and by adding this offset to the local time all the sensor motes will in effect be synchronized with each other This protocol does not account for drift and clock skew which means that at some point the local time on each sensor mote will be unaligned again This can be compensated for by transmitting another time beacon bef
27. TOSBase Because of some communications incompatibilities page 13 Byte offset Data 0 mote number 1 4 32 bit packet timestamp 5 6 1st x axis measurement 7 8 1st y axis measurement 9 10 1st 16 bit measurement timestamp 23 24 4th x axis measurement 25 26 4th y axis measurement 27 28 4th 16 bit measurement timestamp Table 2 MicaZ payload structure between the MicaZ and SARD and since it is desirable to have one gateway capable of receiving packets from both the Gateway program was written almost from scratch and only runs on the Freescale platform Since Gateway eventually needs to do time synchronization a high level of control is desired which makes TOSBase unsuitable The program consists of one write buffer and one queue buffer When a packet is received and the UART is not engaged in communication the received packet buffer becomes the new write buffer and vice versa by swapping pointers and the program begins to transmit the write buffer over the UART connection If however the UART is in use the packet buffer is swapped with the queue buffer and once the write buffer has been completely transmitted the write buffer is swapped with the queue buffer and subsequently transmitted over the UART connection If both the write buffer and the queue buffer are in use the packet is dropped One could add more queue buffers but for this project a high sampling rate is desired which means that p
28. VB13192 only has one antenna it uses a special pin to set if either the input or output port of the transceiver is connected to the antenna at a given time Because this pin is left unused on the MC13192S ARD programs written for the EVB13192 are thus also compatible with the MC13192SARD eventhough the radio is connected differently Had the process been reversed and had the original development first been performed on the MC13192SARD instead of the EVB13192 this would not had been the case 2 1 2 Loading Code The pre shipped bootloader was replaced by a new one written by Jan Flora and Esben Zeuthen at DIKU which are the people responsible for a major part of the DIKU contribution to the TinyOS repository The new bootloader is a low speed version of the one used at DIKU and uses the RS 232 port instead of the USB port With a Background Debug Module this new bootloader was installed on all of the MC13192SARD devices Programming through the RS 232 port is done with the Developer s Serial Bootloader program 10 Since the EVB13192 has not been fully integrated with TinyOS yet the usual make platform install id programmer location command is not supported In order to change the value of the system variable TOS LOCAL ADDRESS after compi lation the script set mote id id located at contrib diku evb13192 tools scripts must be used The new executable is called app s19 id and resides in the application
29. ackets will be arriving at a constant rate Since the UART connection is the bottleneck of the system once the sampling rate exceeds the UART transmission rate no amount of queues will prevent packet drops In a burst environment however a higher number of queue buffers would be advantageous 3 4 4 Xlisten The Xlisten program is one of the standard TinyOS programs to log mote data sent over the UART or ethernet The program was modified to communicate with the Gateway program and print the accelerometer data packets in the correct format The program can be started with the following command gt xlisten exe j s com lt number gt 3 5 Evaluation The evaluation of the accelerometers has been divided into five subsections Subsection 3 5 1 contains the cali bration data for all the accelerometers Subsection 3 5 2 studies the drift measured by a stationary accelerometer Subsection 3 5 3 examines the impact of the power supply s voltage on the acceleration measurements In subsec tion 3 5 4 the acceleration from a walking person is measured and finally in subsection 3 5 5 the acceleration from a purely translatorical movement is measured 3 5 1 Calibration The first and foremost thing one must do in order to use the accelerometers to measure an absolute acceleration is to convert and scale the readings from the ADC into proper measuring units Since the voltage response from the accelerometers is linear with respect to the measu
30. adcast messages were indeed received almost simultaneously at the three slave motes This shows that the assumption in which the Reference Broadcast protocol was based upon is indeed sound By looking at the variances between half the master mote s RTT and the slave mote s RTT one should not expect the precision attained from this synchronization to be higher than 1 2 ms 4 4 3 Clock Drift The clock drift is the deviation observed with respect to some reference frame after a certain amount of time By synchronizing all the motes once and subsequently calculate the offset from some reference clock the drift can be measured Two reference clocks were used the PC clock and the master mote clock Since all packets are timestamped upon receival at the PC and the PC clock is used by the sensor motes to perform the external synchronization the offset between the PC and the sensor motes can be used as a measure of the absolute drift of the sensor motes clocks This difference between the PC and the sensor motes clocks can be seen in Figure 22 It can clearly be seen from the figure that the change of offset are almost identical for all four motes since they all possess the same saw tooth characteristic Also the order of magnitude is several times larger than that of the 25 All broadcasted messages are received almost simultaneosly by neighboring motes page 33 Master mote 5 10000 T T
31. copes must be performed Also the data collection programs are far from complete The measurement packets could be optimized by ex ploiting the fact that the ADC only samples measurements with 10 bit accuracy although this is represented with 16 bit in TinyOS By reclaiming this unused space it would be possible to include an entire measurement more in each packet and hence increase the overall sampling rate The gateway could also be made more robust for transmission bursts by increasing the number of queueable packets before one is dropped The power level remains an issue as well since a depleted battery renders the accelerometers useless A work around for this problem could be to constantly monitor the noise level the deviation from the current mean value locally at each sensor mote and sound an alarm if the noise rose above a certain threshold This would however effect the overall performance negatively Another option would be to attach a voltmeter to each sensor mote and regularly read the voltage on the battery This could be done with minimal additional hardware by using the on board ADC as a simple voltmeter although this would increase the power consumption The drift from a stationary accelerometer was measured however it remains to be seen if a moving accelerometer has the same drift This could be measured by moving the accelerometer in a known oscillating pattern and observe deviations over time from this pattern e g by mounting
32. done on the MicaZ mote since this has to be done at compile time 2 2 Future Work In order to perform time synchronization in a wireless sensor network it is essential to have two way communi cation since sensor motes transmitting data must be able to receive synchronization messages as well Since the SimpleMacM module is capable of receiving messages from the MicaZ motes it is possible to develop a heteroge nous data collection architecture by using a SARD mote as gateway However it will not be possible to time synchronize the MicaZ motes with the SARD gateway It is thus important to establish full duplex communication between the SARD motes and other ZigBee complaint motes such as the MicaZ Either by making the appropriate modifications to the SimpleMacM module or by completing the new native TinyOS radiostack Since all analog accelerometers have a characteristic signal response a universal calibration would be useless And as later experiments will show the signal response is highly dependent of the supplied voltage page 9 2 3 Summary With the TinyOS contribution library from DIKU and the new bootloader it is possible to compile and upload TinyOS programs to the MC13192SARD With the alterations made to the UART component communication over the RS 232 port is possible and all the test programs that uses the UART for communication now works on the MC13192SARD With the new accelerometer modules it is possible to sample meass
33. e 25 Close up of each of the axes as marked with rectangular boxes in Figure 24 y axis 0 01 H o E 5 22473 6 ms 22477 4 ms Kk 0 gt 5 D 8 0 01 4 1 1 1 1 2 2473 2 2474 2 2475 2 2476 2 2477 2 2478 Time milliseconds x 10 Figure 26 Close up of the y axis as marked by the rectangular box in Figure 25 The arrow repre sents the magnitude of the disagreement between the four motes when measuring the same event As indicated by the readings the accuracy of the time synchronization is of the order of 4 ms page 37 the timestamp This could also be seen as a large WSN where measurements were collected by the use of two distinct gateways instead of using one gateway and a multi hop protocol As seen in Figure 22 the sampling of the PC clock is not very accurate Hence the two gateways were connected to the same PC to minimize this source of error x axis Acceleration m s Time milliseconds A Figure 27 Acceleration measurements taken from four motes synchronized with Reference Broadcast similar to Figure 24 but with the use of 2 gateways All motes are measuring the same event and each mote are represented by a distinct color with the motes represented by blue and red belonging to one gateway and the motes represented by cyan and green belonging to the other Figure 27 shows the acceleration measureme
34. e accurate within 0 1 in order to measure position parallel to the Earth s gravity As a result the accelerometers used in this project were not sensitive and precise enough to deliver the needed accuracy It is unclear though if measurements obtained from axes perpendicular to Earth s gravity have the same instability page 41 References 1 Freescale Semiconductor Inc Sensor Applications Reference Design SARD User s Guide MC13192SARDUOG Rev 1 5 07 2005 http www freescale com files rf_if doc user_guide MC13192SARDUG pdf 2 Freescale Semiconductor Inc MC13192 Evaluation Board Reference Manual MC13192EVBRM D Rev 0 0 08 2004 http www freescale com 3 Freescale Semiconductor Inc MC9S08GB60 GB32 GT60 GT32GT16 Data Sheet HCS08 Microcontrollers http www freescale com 4 Freescale Semiconductor Inc MC13192 2 4 GHz Low Power Transceiver for 802 15 4 MC13192 D Rev 2 4 07 2004 http www freescale com 5 Freescale Semiconductor Inc 1 5g Dual Axis Micromachined Accelerometer MMA6160Q Rev 2 10 2004 http www freescale com files sensors doc data_sheet MMA6260Q pdf 6 Freescale Semiconductor Inc Low G Micromachined Accelerometer MMA1260D Rev 1 10 2004 http www freescale com files sensors doc data sheet MMA1260D pdf 7 Crossbow Technology Inc MOTE KIT2400 Datasheet 6020 0062 01 Rev C http www xbow com Products Product pdf files Wireless
35. ecause currently no open source compiler compatible with the HCSOS8 in struction set exists The TinyOS code is thus precompiled to C code locally and the result is transmitted to the compile host at DIKU This host has a propriertary compiler installed which finally compiles the executable Work in progress though on porting the current propriertary compiler to an open source compiler This would make it possible to perform the entire build process locally without the need for internet access 2 1 1 EVB13192 vs MC13192SARD Because of several differences between the EVB13192 and the MC13192 several software modifications and additions had to be made in order to run TinyOS on the MC13192S ARD 5 Where noise is measured as the RMS error when operating at 0 1 Hz 1 0 kHz RMS T EN y x Shttp www sourceforge net TCURL http curl haxx se page 7 The two most obvious differences between the EVB13192 and the MC13192SARD are the extra serial over USB port connected to the second SCI module SCI2 on the EVB13192 and the two accelerometers on the MC13192SARD Not so obvious is the bootloader program pre installed on the MC13192SARD which is not compatible with code generated through DIKU In order to develop TinyOS applications to the MC13192SARD these three problems must be solved Additionally as mentioned above the MC13192SARD is equipped with two on board antennas each dedicated to transmit and receive respectively Since the E
36. ect to the local reference frame One transformation can however be done without the use of gyroscopes although it will be a purely cosmetic one By using the Earth s gravitational acceleration which can be measured when the accelerometers are stationary it is possible to use this vector to rotate the local reference frame so the z axes from both reference frames are aligned 3 3 Hardware The three accelerometers used were the Analog Devices ADXL202E Freescale MMA1260D and Freescale MMA6 261Q with the ADXL202E being mounted on the Crossbow MTS310CA sensor board Analog Devices ADXL202E This dual axis accelerometer has a range of 2 g on both axes It has a noise density and sensitivity of 200 ug y Hz and 167 mV g respectively Also the temperature drift is listed to be 20 5 15 Freescale MMA1260D This is a single axis accelerometer with a range of 1 5 g The noise density and sensitivity are listed as 500 ugv Hz and 1200 mV g respectively which makes it a more sensitive accelerometer than the previous one but with lesser accuracy This accelerometer has however built in circuitry to compensate for temperature drift 6 Freescale MMA6261Q This is a dual axis accelerometer with a range of 1 5 g for each axis The noise density and sensitivity are listed as 300 ugy Hz and 800 mV g respectively which puts it between the two previous accelerometers with regard to noise and sensitivity This accelerometer also has built i
37. ection since it is easy to see that the majority of the acceleration is positive and hence leading to a positive velocity at the end The experiment leading to the measurements in Figure 16 was conducted on the large track and was carried out in the same manner as the one on the short The figure shows great resemblance with the former especially the z axis which again enables one to count the actual number of steps taken and calculating the duration of each step which has been summarized in table 6 Although the acceleration for all three axes return to their origin after the experiment is completed the acceleration in the y direction has the same problem as in the former experiment since the overall acceleration is clearly positive which makes it futile to deduce a position from the acceleration No Middle s Duration lt s gt 1 2 572 2 3 236 0 664 3 3 916 0 680 4 4 485 0 569 5 5 109 0 624 6 5 673 0 564 7 6 272 0 599 8 6 841 0 569 9 7 445 0 604 10 7 898 0 453 11 8 587 0 689 12 9 161 0 574 13 9 710 0 549 14 10 26 0 55 15 10 84 0 58 16 11 37 0 53 17 11 96 0 59 18 12 49 0 53 19 13 08 0 59 20 13 62 0 54 21 14 17 0 55 22 14 73 0 56 23 15 31 0 58 24 15 88 0 57 25 16 49 0 61 Table 6 Step duration This trend of not being able to recover the actual events from the acceleration measurements besides the step duration and count is p
38. eescale SARD Further evaluations found the accuracy of the synchronization to be within 4 ms for a single WSN with one gateway and 27 ms for two WSN connected to the same PC With these applications acceleration measurements were collected from different experiments First the accelerometers were calibrated in order to convert the readings to SI units and an upper bound of the accuracy was found to be 0 0393 m s and 0 196 m s for the SARD and MicaZ mote respectively This is caused by the combination of voltage range on the accelerometers and the resolution of the ADC Second the zero motion drift was measured and we found that only the Crossbow motes had detectable drift This was caused by the lack of temperature correcting circuitry which was present on the Freescale motes Third it was discovered that the power voltage had a significant negative influence on both the noise and value of the acceleration measurements resulting in increased noise and drifting values Fourth the necessity of knowing the orientation of the accelerometers became clear when measurements obtained during walking yielded no useful information except the step duration This fact was emphasized when a similar experiment with fixed axes showed the velocity and position of an elevator car with 10 accuracy Fifth the sensitivity of the calibration and the acceleration as a means for calculating the position were investigated and it was found that the calibration had to b
39. elf configuring 4 4 Evaluation The evaluation of the time synchronization protocol has been divided in three sections First the Round Trip Time is measured in Section 4 4 1 second the drift is measured in Section 4 4 2 and finally the precision is measured in Section 4 4 3 However since the MicaZ motes were not able to receive messages transmitted by the SARD motes and thus unable to participate in the time synchronization only the latter were used in the experiments 4 4 1 Round Trip Time In order to measure the calculated RTT at each mote over a period of time a network consisting of four SARD motes and one gateway were resynchronized every 600 ms The time difference between the RTT request and the RTT reply can be seen in Figure 21 and a reference to the data can be found in Appendix C As mentioned above since the slave motes only calculate a partial RTT compared to the master mote s RTT the slave motes should have a difference half of that of the master mote This explains the lower times for the slave motes compared to the master mote as seen in Figure 21 to a certain degree However since these times are generally lower than half of that of the master mote this deviation is probably caused by the RTT calculation being done at the application level and with only two timestamps instead of four It should be noted however that the three slave motes have an almost identical distribution of time differences which means that the bro
40. from the other which could explain the cumulative velocity buildup A reference to the raw data files can be found in Appendix B 3 5 5 Elevator In order to measure whether the effects of the mote s rotational oscillations could account for the observed prob lems the walking motion from the last experiment were transformed into a purely translatorical movement by replacing the human test subject with an elevator car Since the motion of the elevator car is restricted by vertical tracks and controlled by an electrical engine this would in effect give a repeatable environment with a close to per fect translatorical movement The obvious downsides of this are the inability to automate the experiment process and that the primary measuring axis would be skewed by gravity which could lead to boundary effects since the measurements are being carried out at the boundary of the calibrated measuring interval The experiment was carried out at the public elevator at the William Gates Building which is two stories high The distance between each floor was measured to be 3 50 0 01 m Four different kind of trips were carried out with the elevator 1 Moving from the ground floor directly to the second floor 2 Moving from the second floor directly to the ground floor 3 Moving from the ground floor to the first floor and from the first floor to the second floor with the only brake being the doors opening and closing on the first floor 4 Mov
41. h Duty Cycle Output Rev A 10 2000 http www analog com UploadedFiles Data_Sheets 53728567227477ADXL202E_a pdf a Kone http www kone com 17 Kay R mer Philipp Blum and Lennart Meier Time Synchronization and Calibration in Wireless Sensor Networks 18 Bharath Sundararaman Ugo Buy and Ajay D Kshemkalyani Clock Synchronization for Wireless Sensor Networks A Survey March 22 2005 19 Hui Dai and Richard Han Tsync A Lightweight Bidirectional Time Synchronization Service for Wireless Sensor Networks ACM SIGMOBILE Mobile Computing and Communications Review 8 1 125 139 January 2004 page 42 20 Jeremy Elson Lewis Girod and Deborah Estrin Fine grained Network Time Synchronization using Reference Broadcasts In Fifth Symposium on Operating Systems Design and Implementation OSDI 2002 December 2002 21 Joseph Y Halpern and Ichiro Suzuki Clock Synchronization and the power of Broadcasting Distributed Computing 5 2 73 82 1991 22 F Christian Probabilistic clock synchronization Distributed Computing Vol 3 pp 146 58 1989 23 H Garcia Molina Elections in Distributed Computer Systems IEEE Transactions on Computers Vol C 31 No 1 pp 48 59 1982 page 43 A Application Source Code http www distlab dk sensornet sentientsports datacollection zip B Acceleration Data Files Calibration Data http www distlab dk sensornet sentientspo
42. he result of a six week visiting student project carried out at the Computer Laboratory at the Univer sity of Cambridge England The project contributes to the Sentient Computing and Sports research at the Digital Technology Group and is part of the Embedded WiSeNts project in which the Department of Computer Science at the University of Copenhagen Denmark is a participant I would like to thank the people at the University of Cambridge especially Professor George Coulouris and Dr Marcelo Pias for hosting and supervising the project Dr Alastair Beresford for organizing residence and Dr Robert Harle Brian Jones and Alastair Tse for helping with the local computer system I would also like to thank the people at the Univerity of Copenhagen especially Associate Professor Philippe Bonnet for supervising and Jan Flora and Esben Zeuthen for technical support regarding the Freescale devices The project was funded by The EU Framework 6 Embedded WiSeNts Project Cambridge University Computer Laboratory and DIKU University of Copenhagen 1 11 Motivation In the field of sports technological advances have been used for a long time to improve the performance of athletes Until recently this has been done either inside a lab with wired sensors or outdoor with external measuring equip ment e g a stop watch The goal of the Sentient Sports project is to combine these two approaches and enable athletes to wear wireless sensors without restrictions
43. he transmission time can be calculated by assuming that t t The partial RTT is thus t3 f and the transmission time from the gateway to the slave mote can be calculated by Trransmission ts t to ti Since only the partial RTT is measured it is not necessary to divide by 2 to obtain the transmission time This protocol scales well with larger networks and is extendable to multi hop networks as well It does not account for clock drift and skew but this can be compensated for as mentioned in the previous section One flaw with this protocol however is the presence of a master mote which is a single point of failure In order to overcome this problem a rotation and or an election scheme must be added to ensure fault tolerance 4 3 Implementation Issues Since the current MAC layer for the Freescale MC13192 has not been fully implemented in TinyOS yet it was not possible to implement MAC layer timestamping of packages and thus the value t2 t could not be measured since this this difference was neglible at the application layer given the current resolution of the clock Since t1 t only one timestamp was added to the synchronization reply instead of two In this simple implementation the master mote mentioned above was one of the sensor motes since these transmit a constant flow of measurement packets Since each measurement packet contains the transmission time each measurement packet can be used as a RTT reque
44. ing from the second floor to the first floor and from the first floor to the ground floor with the only brake being the doors opening and closing on the first floor As with the other experiments the measurements were filtered for corrupted transmissions and the reference coor dinate system was rotated once to align the z axis towards gravity As an example the measurements of each axis sampled in experiment 4 are plotted in Figure 17 the other experiments showed similar results First and foremost it can clearly be seen from the measurements sampled from the x and y axis that the motion indeed was translatorical since no acceleration can be measured in these two axes Second these two axes also show that if there is any crosstalk between the ADC channels it must be of the same order of magnitude or less than the background noise Also disregarding gravity one can clearly see symmetry between the different peaks on the z axis which one would expect since the elevator car was stationary at the beginning and at the end of each experiment Since this is directly opposite of what was seen in Section 3 5 4 it seems reasonable to try to calculate the velocity and position of the elevator car After subtracting g 9 81258 m s from the z axis and filtering the measurements with a moving average a lowpass filter and a Savitzky Golay filter in that order the acceleration data was integrated twice with respect to time to yield the velocity and po
45. ly showed the effects that rotations of the local reference frame can have on the acceleration measurements and the subsequent need for gyroscopes to compensate for these rotations This was even more clear in the last experiment where the rotation factor had been removed by using an elevator car to ensure a translatorical movement This experiment yielded however another source for error namely the sensitivity of the calibration and that the given accelerometers lacked the desired precision With this taken into account it was possible to deduce the velocity and position of the elevator car with 1096 accuracy page 29 4 Time Synchronization One of the key aspects of a WSN is the ability to observe the environment simultaneously from multiple sites In order to compare measurements collected from different motes a temporal ordering is often necessary Especially 1f several motes are monitoring the same event the collected data must be in temporal agreement Since the overall project goal is to monitor the performance of athletes with several wireless sensor motes placed on different parts of the athlete s body it is essential to have a fine grained time ordering of the collected measure ments in order for the data to be compatible This involves first and foremost a timestamping of all the collected measurements Nevertheless since this has to be done at each and every mote in the WSN the clocks from which the timestamping originates must be
46. ment on the short track x axis 4r T T T So E 2r 4 S o LE M MI NN AN ML ll A U IN MAA a UT Cu II NI lt L J U I I 1 I I I I 0 4 6 8 10 12 14 16 18 20 Time Seconds y axis ai 8r T T T T T J a E 6r N E 2 l S 2 j 2 0 I Hl L IN JM 1 WW AME TYP La aka Ahern o a IHE WTN TY hi IW o eS E Ii 1 Ii I Ii I I Ii 1 0 4 6 8 10 12 14 16 18 20 Time Seconds z axis T T T T T No E 15r 7 E mu 10 A MAINAN ARO TR TIT E A A 2 5L J I I 1 I I I i I I 4 6 8 10 12 14 16 18 20 Time Seconds Figure 16 Walking experiment on the long track page 24 Unfortunately besides the step duration it is not possible to deduce any quantitative information from the three plots Ideally the x and y axis should be integrated twice with regard to time in order to get the velocity and position as a function of time which would be a plot of the walked distance and combined with the step duration each step length could be deduced However disregarding drift the measurements for each axis should return to their original values at the end of the experiment since the test subject was standing still before and after the experiment Since neither of the three axes have this behavior a simple integration would lead to a non zero velocity at the end of the experiment and thus resulting in an unbounded position This fact is most obvious by looking at the acceleration in the y dir
47. n circuitry to compensate for temperature drift 5 3 4 Application In order to sample measurements from the accelerometers and have them stored on a PC three applications were used Transmit Gateway and Xlisten Transmit and Gateway were developed in TinyOS while Xlisten is a modified version of Crossbows logger program A reference to the source code for all three programs can be found in Appendix A 3 4 1 Sampling rate Since this particular type of WSN contains several layers of hardware abstractions e g sensor motes gateway and PC it is important to look at the limiting factors in each layer in order to achieve optimal performance This will in effect determine a maximum sampling rate for each layer Designating the sampling rate for the actual measurements to be at the sensor level the measurement transmission rate over the radio to be at the communications level and the measurement transmission rate between the gateway and the PC to be at the network level an estimate of the overall limiting factor would be the network level Because the transmission speed of the UART is 38 4 kbps and the radio is 250 kbps a standard TinyOS message can Tn its current configuration the SARD UART has a maximum transmission speed of 38 4 kbps although it might be possible to increase this to 115 2 kbps 13The standard size of a TinyOS message is 29 bytes page 12 be transmitted in 6 ms and 0 9 ms respectively disregarding overhead
48. nary all measurements are said to be taken relative to the local reference frame In order to convert these measurements to the laboratory reference frame it is necessary to perform a reference frame transformation l Where common refers to the non relativistic and non quantum mechanical regime page 11 According to the equivalence principle of mechanics 14 measuring an acceleration of a reference frame is equiva lent to measuring an opposite gravitational acceleration inside the said frame Since the acceleration measurements in these experiments are taken with respect to an accelerated reference frame the accelerometers are actually mea suring the gravitational acceleration mentioned above Because this corresponds to an opposite acceleration of the reference frame the actual acceleration as seen from the laboratory reference frame can be deduced by simply changing the sign of the acceleration This is however only true for accelerated reference frames without rotation If the accelerated reference frame is rotating as well the transformation is not quite as simple One must then perform a rotational transformation of the acceleration as well in order to measure the acceleration relative to the laboratory reference frame This however requires that the rotation vector of the accelerating reference frame is known Since this information was not available to us due to lack of gyroscopes all measurements hence forward are all with resp
49. nts collected while the beam was shaken The three rectangular boxes mark the movement of the beam parallel to the axis in question and a close up of each box can be seen in Figure 28 An estimate of the accuracy measured from figures similar to Figure 29 shows this to be of the order of 27 ms This is not surprising since the fluctuations from the PC clock observed over a short time period is of this order of magnitude as seen in Figure 22 A reference to the data files can be found in Appendix C 4 5 Future Work As seen in the last section the accuracy of the synchronization was of the order of milliseconds which is several orders of magnitude worse than the reported microsecond accuracy reported in 20 and 19 Two sources of error can easily be identified as being the lack of MAC layer timestamping and the millisecond resolution of the clock Since without the actual transmission and receival timestamps at the lower level the calculated transmission time also includes the time spent processing at the upper levels which is highly dependent on the current workload and thus susceptible to great variation When a fully native TinyOS radio stack has been implemented on the 27This would be advantageous in a high performance environment where the overhead of packet routing would have a negative impact on performance page 38 a o a o0 o Acceleration m s 0 L L L L L L L 1000 2000 3000
50. ocess can vary from time to time and usual several retries are necessary page 8 2 1 4 Reading from Accelerometers According to the MC13192SARD schematics 1 the dual axis accelerometer s x and y axis are connected to the two ADC ports ADO and AD1 respectively and the single axis accelerometer which becomes the z axis is connected to port AD7 on the ADC Since all values are with respect to the common ground all measurements become absolute values and sampling from the two accelerometers hence can be done simply by reading the values of the three ADC ports The three modules XaxisM YaxisM and ZaxisM do exactly this The readings are however neither in volts nor in m s but some arbitary 10 bit value taken from a linear voltage to value conversion table inside the ADC These values must subsequently be transformed into measuring units in order to get an absolute acceleration Only the raw ADC values are supplied by the modules however and the user is required to perform a calibration of these readings before an actual acceleration can be obtained To use the accelerometers one only needs to insert the line SENSORBOARD accelerometer in the makefile for the application Each of the above modules provides the interface ADC and can be included as any regular TinyOS component in any combination Sampling is done asynchronously with a call to the command call ADC getData and the measurement is returned by the event async event ADC da
51. ody parts during a course of actions would be useful for athletes in order to improve their performances A runner would be able to know the exact length and duration of each step instead of just an averaged value Gymnasts would have direct access to the three dimensional movement of their body without the need to resort to guessing from watching two dimensional video recordings In order to put accelerometers on athletes it is necessary to understand the behavior of these devices and especially their drift The purpose of this part of the project is to evaluate the accelerometers on the Freescale MC13192SARD and the Crossbow MicaZ specifically with the goal of obtaining a position from acceleration measurements Section 3 1 describes work related to accelerometers and Personal Navigation Systems and Section 3 2 discusses the transformations between different reference frames Section 3 3 presents the accelerometers used in this ex periment and Section 3 4 gives an overview of the TinyOS applications developed to evaluate the two different platforms Last Section 3 5 contains an analysis of the experimental results obtained and Section 3 7 summarizes this part of the project 3 Related Work The use of dead reckoning for Personal Navigation Systems has been done by 11 and 12 In 11 the use of a sophisticated measurement device including GPS gyroscopes and compass were used in conjunction with map matching in order to provide an indoor na
52. of the order of nanoseconds Since this is several orders of magnitude smaller than the desired accuracy it is reasonable to disregard this time delay 231t should be noted that all timestamps are taken with respect to the local clock and are thus not directly compatible however since the time difference is only calculated with timestamps taken from the same mote e g both t and t are from mote b the calculated time differences are compatible with one another page 31 Reference Broadcast Synchronization request Master mote Gateway t Slave mote m to reply Gateway t Master mote t 0 2 MCG fen Slave mote EE i m ty zd 23 m to t tp st Figure 20 Overview of the Reference Broadcast Synchronization protocol The protocol is illustrated in Figure 20 The master mote initiates synchronization with the gateway by broadcasting a synchronization request containing the local transmission time tp This is symbolized by the two arrows on the upper part At time t the gateway broadcasts the reply containing fo t and f where t is the receival time of to The master mote then has the complete information to perform a regular RTT calculation by using the algorithm mentioned above The slave mote s however cannot do the same calculation since the timestamps fo and t4 are not compatible since t4 is measured with a different clock than fg Instead an estimate of t
53. ore the drift and skew becomes significant Since this protocol does not account for the transmission delay it is not possible to extend it to a multi hop environment Also since the sensor motes are not synchronized with an external source it would not be possible to extend this protocol to encompass several gateways that are not in communication range 4 2 2 Reference Broadcast To overcome the short comings of the time beacon protocol mentioned above it is necessary to perform an external synchronization which in turn requires the knowledge of the transmission delay In order for mote A to calculate the most accurate transmission delay to mote B at least two packets and four timestamps are required e Mote A transmits a packet with the timestamp fo of the transmission time e Mote B timestamps the packet upon receival with the arrival time f e Mote B returns the packet to mote A but timestamps the packet with the transmission time t2 as well e Mote A timestamps the packet with the receival time t3 and is now in possession of four timestamps e By calculating the difference t3 tg mote A has the Round Trip Time RTT e By calculating the difference RTT t2 t mote A has the total transmission time and by dividing with 2 mote A also has the average transmission time e The average transmission time combined with the timestamp f enables mote A to synchronize itself with mote B This is known as Christian s method 22
54. pdf MICAz Kit Datasheet pdf 8 Jason Hill Robert Szewczyk Alec Woo Seth Hollar David Culler and Kristofer Pister System Architecture Directions for Networked Sensors ASPLOS 2000 Cambridge November 2000 http www tinyos net papers tos pdf c Crossbow Technology Inc MTS MDA Sensor and Data Acquisition Board User s Manual 7430 0020 03 Rev B April 2005 http www xbow com Support Support pdf files MTS MDA Series Users Manual pdf 10 Freescale Semiconductor Inc Developer s Serial Bootloader for M68HC08 and HCS08 MCUs AN2295D Rev 6 11 2004 http www freescale com files microcontrollers doc app_note AN2295 pdf 11 Pierre Yves Gilli ron Daniela B chel Ivan Spassov and Bertrand Merminod Indoor Navigation Performance Analysis Proceedings of the 8th European Navigation Conference GNSS 2004 17 19 May Rotterdam The Netherlands 12 Cliff Randell Chris Djiallis and Henk Muller Personal Position Measurement Using Dead Reckoning Department of Computer Science University of Bristol 13 Y K Thong M S Woolfson J A Crowe B R Hayes Gill and R E Challis Dependence of Inertial Measurements of Distance on Accelerometer Noise Measurement Science and Technology 13 2002 1163 1172 http stacks iop org MST 13 1163 14 J M Knudsen and P G Hjorth Elements of Newtonian Mechanics Springer Second edition 1996 15 Analog Devices Low Cost 2 g Dual Axis Accelerometer wit
55. ponds roughly to one count each millisecond When the buffer is full the data is copied to the radio packet which also contains the address of the mote and a timestamp of when the packet was sent This timestamp is the full 32 bit counter though and if the sampling rate is high enough to avoid overrun of the measurement timestamp the upper 16 bit can be used to restore a full 32 bit timestamp for each measurement The sampling rate is set by the compiler define define SAMPLING RATE number of milliseconds between each sampling and is the actual time between each sampling on each mote With a single SARD mote this can be as low as 5 ms which is quite far from the theoretical 2 ms The packet structures for the above mentioned program can be seen in Table 1 and Table 2 for the SARD and MicaZ mote respectively and both have been designed specifically for this application Byte offset Data 0 mote number 1 4 32 bit packet timestamp 5 6 1st x axis measurement 7 8 Ist y axis measurement 9 10 Ist z axis measurement 11 12 Ist 16 bit measurement timestamp 21 22 3rd x axis measurement 23 24 3rd y axis measurement 25 26 3rd z axis measurement 27 28 3rd 16 bit measurement timestamp Table 1 SARD payload structure 3 4 5 Gateway This program resides on the mote connected to the PC and basically dumps all received radio packets to the UART similar to the standard TinyOS gateway program
56. r networks WSN accelerometers dead reckoning Analog Devices ADXL202E Freescale MMA1260D Freescale MMA6261Q TinyOS time synchronization Reference Broadcast Synchro nization page 2 Figure 1 Four Crossbow MicaZ motes back and two Freescale MC13192SARD front page 3 Contents Contents 4 1 Introduction 5 A SAA A Gs we wo eb eee a wb ba bo we eh oe ee eS 5 1 2 Purpose da is BOS ek ee Sete eMe BG Ot be erue EAS doe Bea Sages 3 2 Freescale MC13192SARD 6 DAL OO ag e In eek pee ese es eMe doe reae de BS nia te Bee deb Os 7 2 1 1 EVBTI3192 vs MC13192SAR DA se A ta ee XE E RE ues 7 2 1 2 Loading Code 4 loss ek eR RE Rhe Eee Rer eg 8 2 1 3 Communicating over RS 232 s es e e eoe o e 8 2 1 4 Reading from Accelerometers een 9 2 1 5 Communicating with MicaZ 2 en 9 2 2 Fure WOTKG Iu hu Rede erbe redo ids eR Ed In e Re Deed 9 2 3 Summary aged aru e et Be A ee am Sev pe eis 10 3 Acceleration 11 3 1 Related Work uo ze npo APR AA Ree XD ed Ha a ao NP RIS 11 3 23 Reference Frames 2 zu 5t 66 te uet ea d VE ex dede ee ON ee 11 3 3 Hardwares ud geo Rieder LD Rim IAS Ge Nen Cie De Rr P RI ERIT Ser rd e 12 3 4 Apphi ation koe Sen RR UG UE Mee Ra rS Ue EH LR UE Reg ne 12 34 1 Sampling rate 2 2 cuni m m RR 9 ay Eo ym XR RS eos 12 3427 Mrans mi oo uy e bak ecl O eed te rk 13 34 3 Galeway nue We e a m PETRUS A M be 13 O manye gs X I DW Dee ett dees See tok RE
57. r of 4 ms An example of such a reading can be seen in Figure 26 This is worse than the former estimate which was based upon the variance of the transmission time A probable cause is the timestamping taken place at the application layer which is prone to errors caused by the current workload This seems reasonable since the gateway in the former experiment only processed packets every 150 ms on average while in the latter the average time between packets has decreased one tenth to 15 ms x axis a 1 a Acceleration m s o U o 0 0 5 1 15 2 2 5 3 3 5 4 4 5 5 Time milliseconds x10 Figure 24 Acceleration measurements taken from four motes synchronized with Reference Broadcast All motes are measuring the same event and each mote are represented by a distinct color Subsequently the experiment was repeated with two gateways instead of one The WSN were partitioned in two identical parts consisting of one gateway and two sensor motes by assigning a different radio frequency to each partition The four sensor motes were still attached to the same beam however The purpose of this experiment was to simulate an environment were measurements collected from different WSN had to be compared solely on page 36 Acceleration m s 0 L 6000 7000 8000 9000 10000 11000 12000 13000 Time milliseconds Figur
58. red acceleration a straight line can be fitted to known reference points and the absolute acceleration can be deduced by the linear relation Since we want to calculate a physical length a relative acceleration is unsuitable 15 According to the specifications this holds for most of the devices measurement range page 14 ADC value a x acceleration measured in lt m s gt b As mentioned in the Crossbow documentation 9 the simplest way to determine a and bis by using the gravita tional acceleration as a point of reference This is accomplished by first measuring the two extreme values each axis senses when it is pointed either vertical up or down Since the accelerometers have a linear response these two extremes would be equal to g and g respectively and the spanned interval would be of the magnitude 2g Second by measuring the acceleration the axis senses when it is oriented horizontal one also learns the value corresponding to Og The calibrations made were done with g 9 81258 0 00005 m s which were obtained from the National Physics Laboratory Hence all values forward will be in SI units 7 The raw accelerometer measurements were obtained by attaching each mote one at a time to a spirit level which was then used to align each axis parallel with the gravitational acceleration Each extreme was measured between one and two thousand times and the median were used for the calibration since the median is more rob
59. resent in all the gathered measurements collected from walking with the accelerometers Several different positions and motes were tried but all yielded the same result A probable cause for this could be the rotation of the mote itself hence leading to the equations of motion for the mote not only including ac celeration but also rotation since the coordinate system which the accelerations are measured with respect to itself is rotating This sounds plausible since other Personal Navigation Systems such as the one mentioned in article 12 are equipped with gyroscopes to compensate for rotation of the reference coordinate system during movement This would explain why the acceleration measured for each axis did not return to the original value 20Since the duration of the experiment is of the order of seconds it is reasonable to disregard drift page 25 once the experiment was over since the reference point for each axis had been changed This mixing of axes could also explain the cumulative velocity measured along the y direction since the oscillations on the mote could cause the overall buildup of acceleration along this axis e g positive acceleration was mostly measured by the y axis while negative acceleration was mostly measured by the x axis Another possible cause could be interference on the ADC on the mote If there was any cross talk between the three channels on the ADC any measurement in one axis would be tainted by values
60. rts calibrationdata zip Drift Data http www distlab dk sensornet sentientsports driftdata zip Low Power Data http www distlab dk sensornet sentientsports lowpowerdata zip Walk Data http www distlab dk sensornet sentientsports walkdata zip Elevator Data http www distlab dk sensornet sentientsports elevator zip C Synchronization Data Files RTT Data http www distlab dk sensornet sentientsports rttdata zip Offset Data http www distlab dk sensornet sentientsports offsetdata zip Agreement Data http www distlab dk sensornet sentientsports agreementdata zip page 44
61. sition as a function of time These two functions can be seen in Figure 18 It can clearly be seen that neither the velocity nor the position are bounded which seems odd as the symmetry in the acceleration would suggest that Since the shape of the velocity and position curves although skewed seem to correspond to the behavior of the elevator we suspect that the calibration of the accelerometers might be off and that the measured acceleration are offset from the real acceleration page 26 x axis 1 T 057 A 0 5r A L 1 l l l 0 5 10 15 20 25 30 35 y axis 1 T N k E 05r 4 E p od 7 o 8 05 4 lt 24 L 1 L L L 0 5 10 15 20 25 30 35 z axis 11 T 9 1 0 5 10 15 20 25 30 35 Time Seconds Figure 17 Measurements from an elevator going from second to first to the ground floor This seems to be the case since when we shift the value of the gravitational acceleration that is subtracted from the z axis to g 9 7942 m s the plots of the velocity and position correspond to the theory as seen in Figure 19 The two plateaus for the position correspond to 3 6 m and 7 2 m respectively and the maximum velocity is 1 0 m s This measured elevator car speed seems reasonable since typical speeds for elevator cars from this particular firm are 0 5 1 0 1 6 and 2 5 m s 16 Similar results were obtained from the three other elevator experiments and the values can be seen in Table 7 Values for g
62. st packet However since a reply is required from the gateway it would not be feasible to turn all the measurement packets into RTT requests It is however part of the TinyOS 4The current TinyOS radio stack is actually a wrapper around a Freescale proprietary MAC layer and thus not easy to modify page 32 packet header to include the packet type and by defining a new packet type that symbolizes a measurement packet with a piggybacked RTT request time synchronization can be initiated without the need for any extra packets The RTT reply cannot be piggybacked though since no transmissions from the gateway to the sensor motes are necessary during data collection This packet is unavoidable but fortunately it only needs to contain two timestamps The gateway was synchronized with the PC in a similar manor except that the RTT request was a completely separate packet Although it would be possible to use the piggy bagged RTT request from the master mote to synchronize the gateway with the PC as well it would not be optimal since only after the second synchronization would the motes be synchronized with the PC With time synchronization embedded in the acceleration collection application the application was modified to enforce transmission windows to minimize packet collision By dividing the time interval between each packet transmission with the sensor motes and the gateway and using the mote address to allocate time slots the process was s
63. synchronized in order for the timestamps to be compatible The purpose of this part of the project is to investigate and implement a clock synchronization protocol in the data collecting architecture developed in the previous chapter Since there are several fundamental constraint differences between regular computer networks and WSN it should be emphasized that the protocol in question must honor these constraints and be scalable in order to be suited for a WSN Section 4 1 describes work related to time synchronization in WSN Section 4 2 analysis two different protocols based on broadcasting and Section 4 3 discusses several implementation issues Section 4 4 evaluates the accuracy of the implemented time synchronization Finally Section 4 5 describes different aspects future work might focus on and Section 4 6 summarizes this chapter 4 1 Related Work The field of time synchronization for WSN have been thoroughly studied and much of the work has been summa rized by 17 and 18 Particularly one of the paradigms studied is the usage of message broadcasts as a common point of reference This technique was first theorized by 21 which also established an upper bound of 1 Le on the precision attainable after K broadcasts where e is the time it takes to receive a broadcast message Since the wireless medium facilitates broadcasts this approach is particularly well suited for WSN Actual implementations based on this work have been carried ou
64. t by 20 and 19 In 20 an internal synchronization of the network is performed by having a reference mote broadcast message beacons The receival times of these beacons are captured by each of the other motes and put in a local timetable By exchanging timetables with all the motes each mote is capable of calculating an offset that minimizes the overall RMS error of the local timetables The drawback of this approach is the increased overhead caused by the exchange of timetables among the motes which has a quadratic increase in the number of participating motes This lack of scalability was improved by 19 which assumes that the transmission delay time between the reference mote and each of the other motes is the same By having the reference mote calculate the transmission time to one particular mote and afterwards broadcast this delay to all the other motes all motes within communication range can be synchronized with a total of three messages transmitted As will be shown later this can be improved to a total of two transmitted messages 4 2 Protocols The main target platform for this time synchronization is the collection of acceleration measurements from sensor motes placed on athletes Since a high sampling rate of the accelerometers is desired the time synchronization protocol must be as light as possible in order not to decrease the overall performance The system we want to time synchronize consists of a logger program located on a
65. taReady uint16_t data Notice that even though the ADC only has 10 bits of accuracy the measurement is stored as a 16 bit value If performance is an issue data transmission can be reduced by only regarding the lower 10 bits of each measurement as actual data 2 1 5 Communicating with MicaZ Since a TOSMsg compatible radiostack for the EVB13192 is still in beta test the current stable radio stack is the SimpleMacM module This component is capable of receiving messages that have been transmitted with the Crossbow MicaZ radio 7 but not the other way around Since the messages used by the SimpleMacM module are not following the IEEE 802 15 4 standard and only contain pure data i e no headers of any kind these messages are automatically discarded by the MicaZ radio This also means that all transmitted messages have no destination address and all received messages must be checked since the module in effect works in promiscuous mode Whether or not it is possible for the SimpleMacM module to achieve full compatibility with the MicaZ mote by inserting the appropriate TinyOS headers remains to be seen Since a native TinyOS radiostack is well under way we did not pursue this issue any further On the flip side it is possible to set the radio s communication channel with the call SimpleMacM setChannel channel command where the channel ranges from 0 15 and corresponds to the ZigBee channels 11 26 respectively which is not as easily
66. urements from the accelerometers and use the measurements directly in TinyOS In effect this makes the MC13192S ARD device as TinyOS compatible as the EVB13192 page 10 3 Acceleration The ability to use simple observations of velocity to deduce a position is known as dead reckoning and has been used by sailors since the early Middle Age By throwing wooden objects overboard and counting how many seconds it took them to fall behind navigators had a simple but efficient tool to estimate the traveled distance The connection between acceleration and velocity and position was however first quantified with the publication of Philosophiae Naturalis Principia Mathematica by Sir Isaac Newton in 1687 and Newton s equations of motions have since then become the basic foundation for all common motion calculations ranging from designing marry go arounds to launching rockets to Mars Using Newton s equations of motion on acceleration data gathered from people in motion with the purpose of obtaining a position however has not been possible until recently when such measurement devices became small enough for a person to carry With the advent of Micro Electro Mechanical Systems MEMS accelerometers personal navigation systems based on dead reckoning made it possible to measure the position of the body as a whole The next step would then be to measure the position of individual body parts The knowledge of the positions of individual b
67. ust to sudden spikes and variations than a simple mean Table 3 and Table 4 shows the calibration data and the RMS error for the Freescale motes and Crossbow sensor boards respectively Figure 3 show an example of an actual calibration curve for Freescale mote 6 A reference to the raw data files can be found in Appendix B mote 5 mote 6 mote 7 mote 8 median RMS median RMS median RMS median RMS x up 255 14 xup 239 1 9 xup 220 14 xup 233 1 7 x hor 501 3 2 x hor 492 2 7 x hor 460 7 1 X hor 490 4 6 x down 747 1 3 x down 746 1 7 x down 713 1 4 x down 745 1 4 y Up 248 19 yup 271 2 1 yup 255 2 0 yup 263 1 7 y hor 494 2 9 y hor 513 2 5 y hor 499 2 9 y hor 499 4 1 y down 741 18 y down 755 2 1 y down 747 2 0 y down 739 1 7 z up 239 1 0 zup 226 1 2 zup 240 1 2 zup 241 1 2 z hor 483 4 7 z hor 474 1 5 z hor 487 5 7 Z hor 485 3 0 z down 728 1 1 z down 723 1 1 z down 732 1 2 z down 732 1 2 Table 3 Calibration data for Freescale motes board 1 board 2 board 3 board 4 median RMS median RMS median RMS median RMS x up 413 0 4 xup 405 0 4 xup 464 0 5 xup 447 0 7 x hor 467 0 5 x hor 455 1 0 xhor 513 4 1 X hor 502 1 4 x down 524 0 1 x down 502 0 5 x down 564 0 5 x down 558 0 6 y Up 436 0 6 yup 481 0 6 yup 466 0 9 yup 476 0 7 y hor 492 1 4 y hor 532 1 5 y hor 519 1 7 y hor 532
68. vigational system By matching the measured position with known physical viable positions the accuracy was improved The need for a 3D map model makes the usability limited however 12 uses an estimate of the step length together with directional measurements compass and gyroscopes to deduce a position with an accuracy of 2 20 m Their method is prone to errors on the step length estimation which is deduced by the magnitude of the vertical acceleration Since the tilt of the measuring equipment was not taken into account the vertical acceleration was ill defined The measurement of accelerometers drift and noise has been done extensively for stationary accelerometers by 13 The error of the position achieved by integrating the acceleration twice was calculated and measured for several factors such as noise sampling rate and time and it was found to grow as the square of time This work did not encompass moving accelerometers however 3 2 Reference Frames In order to understand the samples measured by the accelerometers it is important to understand in respect to what this acceleration is measured When one performs a measurement e g measuring the speed of a car driving by with a laser beam this speed is measured relative to the measuring device Since such devices usually are stationary this reference frame is called the laboratory reference frame or sometimes the geocentric reference frame If however the measurement device is not statio
69. x and y axis respectively Since the error of the fit for both axes is lower than that of the observed drift measured acceleration interval the lines do indeed represent measured drift from the accelerometers This is consistent with the expected drift from temperature and in the absence of correction circuitry this must be taken into account when actual measurements are performed 0 5 T x axis y axis 0 4 2 S 0o5r 4 o a 2 s 5 o 8 ip lt 15r 4 L L L 1 0 0 5 1 15 2 25 Time milliseconds x107 Figure 10 Two straight lines fitted to the 5 hour experiment data All lines have been moved to the common origin to emphasize the drift The RMS errors are 0 40 and 0 26 for the x and y axis respectively MicaZ sensor board A reference to the raw data can be found in Appendix B 3 5 3 Power Another interesting source of error was found in the power supply Since the accelerometers on the Freescale motes require a larger amount of voltage than the rest of the mote in order to function properly it is possible to continue sampling even though the output from the accelerometers cannot be trusted This can be seen in Figure 11 where the left part of the plot is taken from a mote with a battery close to depletion and the right part is taken from a mote with a new battery When calculating the RMS deviation from the mean value one gets 12 6 m s and 2 0 m s for the left and right part respectivel
70. y Although it appears that only the noise increases and the mean remains constant when the battery is nearing depletion Figure 12 shows this not to be the case and that it is vital to use batteries with sufficient voltage It should be noted that neither motes where moving during the collection of data A reference to the data files can be found in Appendix B According to the Freescale accelerometer documentation 5 and 6 the devices stop functioning properly when the supplied voltage drops below 2 7V and 4 75V for the dual axis and single axis accelerometer respectively Hence if one knew the power supply voltage and thus the voltage supplied to the accelerometers it would be possible to determine when the measurements were invalid However it is undetermined if the noise observed in Figure 11 increases as the voltage approaches the cut off voltage or if it only occurs after this point It remains to be seen if the accelerometers on the MicaZ motes have the same kind of behavior page 20 Acceleration ADC units Noise from low power voltage 25 30 35 40 45 50 55 60 65 Time Seconds Figure 11 Comparison between a Freescale mote with low power left and full power right Acceleration ADC units x axis 1000 800 600r 400r 200 0 I 5000 10000 15000 y axis 1000 co o o T o eo o T ES o o T m o o I 5000 10000 15000

Evaluation of Accelerometers Mounted on Wireless Sensor Motes

Contents

Download Pdf Manuals

Related Search

Related Contents