Home

TECO3264 32-Channel Echo Canceller Information Manual

1. ESTIMATOR ECHO REFLECTION REFLECTION TAIL END DELAY gt LONG HAUL TAIL END DELAY DELAY 0 ms TO 64 ms USUALLY POSSIBLY HUNDREDS OF MILLISECONDS 5 7083 F ar 3 Figure 6 Long Distance Telephone Connection Lucent Technologies Inc 17 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications continued Wired Telephone Network Applications Local to Long Haul continued Cancelling on the Edge or Maybe Not When a major long haul toll network provider connects to the PSTN with a major local exchange provider there is usually enough of across section of trunks that DS1 or E1 based network echo cancellers with 24 or 32 channels per card can be placed at the edge of the toll network behind the toll switch This avoids picking up the internal long haul delay from the toll network and lets echo canceller systems with a 64 ms tail end delay capacity like sys tems based on the TECO3264 ASIC handle virtually all of the echo control duties The local exchange carrier con centrates their long haul connecting traffic so that the number of connections to the long haul carriers can be handled in as few locations as practical In several countries like the U S A the governments are allowing competition to the current local exchange pro viders by new telephony service providers Most of these new service provide
2. Echoes in Telephone Calls Electrical and Acoustic As mentioned in other sections the main source of electrical echoes in the telephone network is the two to four wire hybrid circuit These days with the telephone system being essentially almost all digital once past the cus tomer interfaces to the networks there will be only two hybrids one for each customer Once the call is set up by the switching machines there will usually be an echo canceller at each end of the long distance portion of the cir cuit with each canceller taking care of the nearest hybrid echo returns Once some signal energy appears on the channel the cancellers can begin to train to the loss and delay characteristics of its hybrid and attached loop and customer terminals The delay for a hybrid tends to narrow in time with only a few milliseconds where the echo occurs say 2 ms or 3 ms within in the 64 ms capability range of most echo cancellers like the TECO3264 The characteristics of the hybrids tend to change very little during the call Once the cancellers have trained only minor fine tuning may be required for the rest of the call The cancellers are monitoring to detect any changes that do occur Multiparty conference calls are a major and real exception to the two hybrids per call since there will be as many hybrids as conference ports connected One bad hybrid without echo control in a call with a dozen parties is one too many However electrical echoes from
3. This is the telephone service that most people have at home Most business voice telephone services are also a form of POTS The main characteristics are voiceband telephones dial tone dual tone or rotary dialing and 20 Hz ringing other ringing frequencies are also used especially outside of the United States The vast majority of telephone services are POTS carrying speech or more and more voiceband modem data or facsimile signals Although various digital data services are growing rapidly the embedded base of POTS is so huge that it will be the dominant service for many years to come In the developing countries which have little or no telephony service POTS type services will probably be the major service but data communications will also be important POTS Loops Metallic Pairs of Today The two wire metallic transmission facility that connects POTS and other customer services between the cus tomer premises not premise as it often incorrectly written and the local telephone switching machine Loops are usually constructed with twisted pair copper wires One main exception is the flat copper clad steel drop wire used for the last short link to the customer premises from a utility pole The physical structure of North American loops varies considerably from one customer to another varying from a few hundred feet to miles long Fine gauge wire 26 AWG is usually used close to the central office Coarser gauge 24 AWG or 22 AWG is u
4. As in the case of wireless local loops the four wire portion of an HFC telephone circuit will extend the four wire network right up the customer premises interface where the HFC hybrid will do the conversion As discussed for wireless loops the customer input impedance is much less controlled than the impedance seen at the central office hybrids and will result in probably lower and more variable ERLs as seen by the network electrical echo can cellers In contrast to the wireless local loops the full 64 kbits s u law or A law companded signals may be trans ported all the way to the HFC terminal hybrid at the customer premises If this is so HFC telephony won t have the nonlinearity concerns of the wireless systems using low bit rate voice vocoders LONG HAUL NETWORK HYBRID FIBER COAX CUSTOMER PREMISES a A SI Le A y a A e ECHO HFC CANCELLER TERMINAL e Gi j H n RS SE TEN ZCP FOUR WIRE gt TWO WIRE gt 5 7087 F r 1 Figure 10 Hybrid Fiber Coax HFC Telephony Lucent Technologies Inc 23 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications continued Integrated Services Digital Networks ISDN Basic Access The New Digital POTS ISDN basic access also called basic rate access BRA has for many years been touted as the digital replace ment for analog POTS ISDN BRA provides a net digital payload to the customer premi
5. Ascot FRANCE 33 1 40 83 68 00 Paris SWEDEN 46 8 594 607 00 Stockholm FINLAND 358 9 4354 2800 Helsinki ITALY 39 02 6608131 Milan SPAIN 34 1 807 1441 Madrid Lucent Technologies Inc reserves the right to make changes to the product s or information contained herein without notice No liability is assumed as a result of their use or application No rights under any patent accompany the sale of any such product s or information Copyright 1999 Lucent Technologies Inc microelectronics group All Rights Reserved August 1999 L t Technologi MN98 071TIC Meeme sel taks ination
6. The loop connects via distribution frames and inside CO wiring to a POTS line unit in the switching machine The switch here is a digital switch The local digital switch is a very spe cialized very reliable computer controlling four wire connections between other local customers or long haul con nections Since internally the switch is a four wire system each POTS customer port has a four wire to two wire conversion circuit called a hybrid In North America digital loop carrier DLC systems are bringing the line unit and hybrid functions out to within 2 miles or less of the customers who are at long distances from the CO LOCAL EXCHANGE SWITCH POTS LINE CARD TWO WIRE LOOP 5 7079 F r 2 Figure 3 POTS Line Card and Telephone Switch 14 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Wired Telephone Network Applications Local to Long Haul continued Local Loop Detail Figure 4 below shows a little more inside the telephone and the switch line unit The telephone has its own hybrid to convert from the two wire handset speaker and two wire microphone to the two wire network loop This hybrid and its balance network is designed for a deliberate mismatch of about 10 dB to provide some feedback called sidetone from the microphone to the speaker This sidetone helps the customer to unknowingly speak at a level that is abou
7. The packet reassembler must wait at least some minimum time like the maximum expected delay difference to put the signal back together in the original prepacket order Also whole packets may be lost or simply arrive too late to be placed back in the stream For data signals where every bit may count forward error correction techniques may be able to restore the errored data at the receiving end else the data block may have to be retransmitted In voice transmission the resulting gaps from missing packets may be filled in with some least harmful default pattern such as a quiet code This would appear to an echo canceller as yet another source of noise and the packet delay would be just another source of delay Packet transmission is also used for permanent point to point connections in addition to being used for switching through a network For example most of the major cellular telephone system manufacturers use frame relay trans mission with variable length HDLC type frames between cell sites and base station controllers These packet paths carry customer voice and data payloads call progress and control data and cell site control and monitoring data A problem that results is excessive delay or latency in packet transmission terminology for the voice transmission The manufacturers are planning to migrate from frame relay to asynchronous transfer mode ATM with its short fixed 53 bytes frames to reduce the delay starting in late 1998 Voice
8. an echo in addition to the transhybrid loss If the round trip delay is greater than about 10 ms and less than about 30 ms the reflected signal is perceived by the far end listener as a hollow sound Delays greater than about 50 ms result in the reflected signal being heard as a separate echo from the original utterance Hybrids are dominant but not the only source of electrical echo in telephone circuits It is possible to design and build hybrids that increase the incoming port to outgoing port loss or echo return loss to gt 30 dB Such hybrids are complex and expensive and are not justified by the vast majority of calls that are local short haul with low delay and no perceivable echo Such hybrids are used for special customer circuits Not shown on most simplified block diagrams is a second two wire hybrid port to which is attached a balance net work that attempts to match balance the input impedance of the two wire cable on the signal port Impedance is the ratio of the signal voltage to the signal current on the wire pair versus frequency For any one wire pair the imped ance varies drastically from 300 Hz to 3000 Hz see Voiceband on page 41 The impedance also varies with other variables such as the temperature of the cable the number and type of telephone sets and other devices con nected and active at the customer end on a given call The physical structure of loops varies considerably from one customer to another varying from a
9. and placed in the ninth bit or alternately over the 15 bit word and placed in the sixteenth bit Parity is not supported when the linear PCM encoding format is used The parity options are provisioned independently for each port In the 4 096 Mbits s companded mode bit 8 bit 15 will pass through unaffected unless the parity generation option is enabled on the output port In that case parity for bit O bit 7 will overwrite bit 8 or bit 15 depending on the parity option chosen Conversion from either A law to p law or u law to A law meets ITU T conversion requirements 2 An alternate adjusted u law to A law or A law to u law conversion is provided 2 Adjusted conversions pre serve 7 of 8 encoded bits for tandom u law to A law to u law or A law to law to A law conversions This feature is provisionable on an all channel basis Digroup often is used to refer to all 24 channels for a 1 544 Mbits s DS1 application In DS1 applications the term group means 12 channels or DSO time slots This document uses the term all channel to refer to all 24 channels for DS1 or all 32 channels for E1 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 TECO3264 Functions Linear Processing Convolution The convolution processor CUP provides for tail end delay of up to 64 ms 512 taps Tail end delay is the round trip delay experienced by a signal exiting the C port being reflected and then entering the Y por
10. be able to distinguish much finer time differences in sound echoes than 10 ms in order to locate bugs and fish It merely blends in as part of the side tone designed into telephone sets Sufficiently large can be 40 dB down If a signal in a bad network comes back larger than the original there are different problems including the possibility of singing like a bad public address system As the delay increases toward 30 ms the circuit starts to have a hollow sound Nearly everyone will hear a reflection as a a separate sound with 50 ms delay if the reflected signal level is not much lower than the background noise 50 ms round trip delay for a sound echo means the reflecting surface is about 200 feet away If an echo returns with 0 dB attenuation inside a two wire to four wire to two wire circuit the circuit can become an oscillator or start singing This is a familiar effect in audio public address systems that squeal until the gain is turned down Usually there is enough ERL that singing doesn t happen often in telephone networks Telephone connections which have the possibility of having a round trip delay of anything like 30 ms or more must have some form of echo control to deliver a circuit over which customers can carry on a satisfactory conversation Before the echo canceller was invented by Bell Labs and the development of very large scale integration made complex devices such as the Lucent TECO3264 practical several approaches to echo co
11. canceller changes such that when a test signal is reapplied to Rin with the convergence circuitry inhibited the returned echo is at the defined level Leak is neces sary to bound the H register response for very narrowband signals such as single tones Near End Near end is the side of the telephone connection which contains the echo path on which the echo canceller is intended to operate Transmission facilities switches the hybrid the loop and terminating customer telephone set are included in the near end side Lucent Technologies Inc 29 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Echo Canceller Terms continued Near End Speech Threshold NEST Near end speech threshold NEST is the minimum attenuation of the signal between port ROUT and port SIN for the echo canceller to declare that only echo is present Nonlinear Processing Loss ANLP Nonlinear processing loss ANLP is additional attenuation of a residual echo level by a nonlinear processor NLP places in the send path of an echo canceller Because human hearing can detect speech like signals that are very small even in the presence of noise that may be higher in level a nonlinear processor function is used to finish the job by adding more loss or cutting off the send path when only the far end speaker is talking The NLP can also insert comfort noise to maintain a live circuit sound When
12. converted to linear PCM The linear PCM estimate of the echo is subtracted from the linear version of the Y signal The difference or residual echoes signal is then con verted to u law or A law before being sent out the E port Usually the raw residual echo is either attenuated further or replaced with the comfort noise when there is no double talking 46 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Adaptive Delta Modulation Pulse Code Modulation ADPCM Adaptive delta modulation pulse code modulation ADPCM is another common standard technique for encoding voice signals with 32 kbits s ADPCM being most common ADPCM works by encoding not the absolute level for each time sample but the difference between the current sample and the previous sample With this approach 32 kbits s of digital bandwidth can achieve nearly the performance of 64 kbits s A law and u law This factor of two bandwidth savings is important in some applications such as rural digital loop carrier systems and some wireless telephony systems The TECO3264 is not designed for ADPCM since it is expected to be used primarily in the toll network External conversion circuits between 32 kbits s ADPCM and A law and u law can be included if neces sary Digital Speech Quality Categories Digitally encoded audio is grouped in four general quality categories broadcast toll
13. even during the call as devices go off and on hook NETWORKS LOCAL amp LONG HAUL RF LINK CUSTOMER PREMISES I AS K A dit SC A SS i Fam N RF RF gt LBRV gt FOUR WIRE TWO WIRE gt ZCP 5 7086 F r 2 Note RF radio frequency Figure 9 Wireless Local Loop The hybrid balance network will probably have to be the standard 600 Q 2 uF network in the U S A used for customer premises switches or a network developed to look somewhat like the impedance of telephone sets The electrical echo as seen by the electrical echo cancellers back in the network will vary accordingly As discussed in the next paragraph this electrical echo may be highly nonlinear as a result of the nonlinear speech coding in the wireless links Lucent Technologies Inc 21 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications continued Wireless Local Loops continued Some wireless local loop systems plan to use the standard low bit rate voice LBRV 32 kbits s adaptive delta PCM ADPCM voice encoding instead of 64 kbits s u law or A law encoded PCM to save radio bandwidth The 32 kbits s ADPCM is very close to the 64 kbits s companding in terms of near toll quality and linearity Some of the wireless local loop systems plan to use technology and speech coding developed for digital cellular and personal communications systems PCS systems These ma
14. from 0 Hz dc to tens of Hz including the har monics of the dial pulses and 20 Hz ringing pulses and pulse rise and fall times These signals are very much lower in frequency than the typical lower cutoff frequency for the voiceband which is about 300 Hz The on hook off hook and ringing are out of band relative to the voiceband while DTMF signaling is in band For digital loop car rier systems connecting a customer premises to the telephone network this out of band call control signaling must be transported back and forth along with the customer voiceband payloads by some means DS1 systems and E1 systems tend to use very different techniques to transport call control signaling DS1 systems rob bits away from the 64 kbits s voice signal to carry control signaling in a robbed bit signaling scheme E1 sys tems gather up the control signaling from 31 of the DSO channels and multiplex it all together in the 30 second common signaling channel DS1 systems take the least significant bit from the eight voice band bits from every sixth 125 us frame This pro vides 64 8 x 6 kbits s or 4 3 kbits s of bandwidth for call control information in each direction which is plenty for call control purposes Older DS1 systems divided this into two channels A and B Newer DS1 systems using extended superframe ESF divide the robbed bits into four channels A B C and D in each direction to allow for more complex interactions between switching systems and othe
15. half to the balance network and none to the send port Likewise half the signal power coming in from the loop goes to the receive four wire port of the hybrid where it is dissipated in the output of the connecting amplifier Half the power from the loop goes out the send side of the hybrid as desired None of the loop input power ends up in the balance network Skipping the details the four wire input to four wire signal transmission function of the hybrid turns out to be of the form ZBN ZLOOP ZBN ZLOOP This is the same form as the return loss RL or the inverse of the reflection coefficient between a transmission line of impedance ZLOoP and a termination of ZBN The other impedances around the hybrid drop out of the THL if they are as shown Ignoring the phase of the complex function the tran shybrid loss THL or echo return loss has the form K 20 log magnitude ZBN ZLOOP ZBN ZLOOP For older two transformer hybrids K was usually about 8 dB when miscellaneous losses were included THL would then be about 8 RL dB as a rule of thumb For the North American loop plant the 900 2 balance network has an average midband return loss of 11 dB and a minimum of about 6 dB The high and low band return loss values are about 3 dB less or 8 dB average and 3 dB minimum which yields a minimum THL of about 8 3 11 dB Note that this is completely adequate for local telephony but not for echo control purposes when the local
16. hybrids are not the only sources of echoes for telephone calls Soeakerphone telephone sets are becoming more popular The incoming sound from the phone is broadcast out into the listener s room and can bounce right back into the microphone With the traditional handset low level sound was fed into the ear and the head tends to block any getting to the microphone and going back out Handsfree speakerphone microphone setups are also becoming more and more popular for cellular telephones in automobiles for safety reasons Again the sound coming in can bounce around inside the car and get back to the microphone The positions of the speaker and microphone and the levels for soeakerphones and handsfree cellular phones must be carefully done to minimize these acoustic echoes There may be several possible reflectors in a room or car producing multiple acoustic echoes These acoustic echoes can vary a lot in level and delays during the call as people and reflecting objects move around Network echo cancellers like the TECO3264 are optimized to provide the maximum performance for the relatively stable electrical echoes Even so they can provide some assistance in cancelling acoustic echoes perhaps 10 dB and more Acoustic echo cancellers are designed to accommodate the more variable characteristics of sounds bouncing around in enclosures These cancellers are being included in the speakerphones and other handsfree devices They normally have to deal with
17. link is part of a long haul circuit The segregated non loaded and loaded balance networks improve this by about 2 dB to yield a minimum THL or ERL of about 13 dB which is still not nearly good enough for echo control Note that a completely mismatched hybrid with a short or open balance network would have an transhybrid loss of 8 dB A lossless hybrid would have a minimum transhybrid loss of 6 dB since half the power 3 dB is first split between the loop and the balance network The reflected power differences from the balance network and the loop get split in half another 3 dB again when going out both four wire ports Miscellaneous losses tend to add another 1 5 dB to 2 dB for the rule of thumb constant 8 dB mentioned above For old style two wire transformer hybrids there is also a fifth low frequency port for feeding dc current from the office 48 Vdc battery supply to power the telephone down the loop and for 20 Hz ringing current to ring the bell These power and signaling currents are fed through large inductors and across shunt capacitors that connect the center tap transformer windings This hides the impedance loading effects on the hybrid at voiceband Two wire transformer hybrids are out of style now because they tend to be rather large with plenty of iron in the cores and copper wire in the windings to avoid magnetic saturation of the core from the dc and ringing currents fed out through the hybrid Small one transformer electron
18. over ATM VoATM has been the subject of industry efforts in recent years to provide quality voice transport Another approach to reduce delay through packet networks is adding priority flags to the packets Packets for delay sensitive services like two way voice would be switched and transmitted before packets for delay insensitive ser vices like one way broadcast video and audio when there is contention for network resources Planning for the next generation of Internet protocol IP includes priority marking and control ITU T recommendation H 323 covers real time audio video and data transport via packet based networks H 323 requires G 711 u law and A law G 722 G 728 G 729 MPEG 1 audio and G 723 1 audio codecs see reference 8 26 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background Echo Canceller Terms This section provides some background information for those new to voice frequency echo control in general and echo cancellation in particular It discusses some of the terms and topics that occur very frequently in the technical specifications and literature Some terms are merely defined with an explanatory sentence or two Other topics are covered in more detail There are other good overall tutorials and primers on the subject of echo control such as the ITU T Recommenda tions G 168 G 165 G 164 and several others in the G series T
19. software The user interface software for the TECO3264 and T7630 is supplied External T1 or E1 communications test analyzer systems or live network circuits may be connected to the line input and output ports The system can be connected to external telephone line circuits to test with live hybrid echo attenuation and tail end delay Full details on the TECO3264EB system are found in the TECO3264EB TECO3264 Evaluation Board System User Manual MN98 072TIC IBMis a registered trademark of International Business Machines Corporation 12 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Evaluation Board System TECO3264EB continued INPUT SIGNALS i E1 T1 w DELAY 1 DELAY 2 O UP TO UP TO SPEUR DATA z 250 ms 128 ms E1 T1 BUS CLOCK M l gt gt lt x x GI I S 1 C lt x ao Z FPGA DATA O BUS CLOCK ZS oO Ee ATTENUATOR ECHO CANCELLER DATA CONTROL BUSES CPU AND MEMORIES MICROPROCESSOR IBM COMPATIBLE 5 7131 F r 2 IBMis a registered trademark of International Business Machines Corporation Figure 2 TECO3264EB Hardware Block Diagram Lucent Technologies Inc 13 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications Wired Telephone Network Applications Local to Long Haul To most telephone customers the worldwide telephone
20. such as source code numerical information and executable code where every bit is important Lossless compandors can shrink the size of computer files and then restore them with every bit intact zip arj tar and Izh are some of the lossless com pandor techniques used for computer data archival and transport Lossy companding techniques discard as much information as possible and still reconstruct a signal at the end that is good enough for the given purpose for human ear or eye and brain processing Specific lossy techniques include signal frequencies limited at low and high ends linear continuous level ranges reduced to discrete stairsteps nonlinear low level signal components near in frequency to high level components discarded by digital filtering linear Some lossy companding schemes include u law and A law for telephony MUSICAM for music audio storage and transmission MPEG for video JPEG and GIF for still images 48 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual References 1 MARWL o Concentration Highway Interface CHI Interface Specification document number DS90 124SMOS Lucent Technologies Microelectronics Group Pulse Code Modulation PCM of Voice Frequencies ITU T Recommendation G 711 Echo Suppressors ITU T Recommendation G 164 Echo Cancellers ITU T Recommendation G 165 Digital Network Echo Cancellers ITU T Recommendation G 168
21. system Digital wireless telephony systems developers found themselves in a situation like the early digital landline systems where every bit of data every bit per second every Hz of analog bandwidth cost circuit complexity and money The very robust 64 kbits s u law and A law schemes were too much when hopefully only one or two analog digital analog conversions would be needed 2 First a standard 32 kbits s low bit rate voice LBRV scheme of nearly the same quality as 64 kbits s was developed just to double the voice channel capacity Then 16 kbits s schemes were developed then 8 kbits s schemes Schemes with only 2 kbits s or even less for voice communications are under study The GSM wire less system which is becoming dominant outside the U S A often uses 13 kbits s voice coding Reportedly cus tomers can be happy with the results The catch is that these lower and lower bit rate voice coding systems require a lot of digital signal processing calculations to implement the many complex algorithms needed to yield acceptable speech for human ear These calculations take time and time equals delay in the voice path The delay increases rapidly as the bit rate goes down Milliseconds of delay appear in calls where the old 64 kbits s approach the con versions are essentially instantaneous Echo cancellers can be used The vocoder functions may be in either the base station controllers or in the mobile telephone switching office For general
22. to connect to the second customer wire pair In modern networks the path through the network or networks is usually a four wire path all the way between hybrids even half way around the world or farther if a satellite link is involved As mentioned elsewhere the four wire to two wire conversion process in hybrids is not perfect Some of the energy that is supposed to go outward from the incoming four wire side to the two wire side leaks across and goes back out on the outgoing four wire port This trans hybrid leakage called echo return loss in echo control termin ology is usually very small relative to the outgoing two wire port signal Even a very poor nearly completely mis matched hybrid would have a 6 dB echo return loss meaning the reflection is down by a factor of four For reason ably well designed systems the echo return loss would run closer to 12 dB or a factor of 16 reflection However the human ear brain signal processing system is very very sensitive when it comes to detecting signals that sound like they might be speech A reflected signal that is 40 dB or even 50 dB down by the time it returns to the speaker can be heard So far even for a network that has very good hybrid performance may need echo control if the round trip delays are large enough When is a reflection perceived as an echo If a sufficiently large echo comes back to the speaker s ear in less than 10 ms it is not recognized as an echo Bats and dolphins must
23. 0 km s or 186000 miles s in round numbers the maximum possible speed Milliseconds are the common time measure in echo control In 1 ms a radio wave travels 300 km 186 miles and a sound wave travels about one foot An electrical signal in plastic insulated twisted pair copper wire travels about half light speed or about 150 km or 100 miles in 1 ms For echo purposes the round trip delay out and back is what counts so the round trip distances for 1 ms propagation delay would be half the one way distances So an electrical signal traveling 75 km or 50 miles trip through wire pairs would experience a round trip delay of about 1 ms if it were reflected back toward the source from the far end The delay in the average North American two mile loop or even two loops connected together for a call does not come close to causing an echo problem for local calls with only microseconds of delay Loops in Europe tend to be shorter and would be of even less concern for delay As an extreme example links through a geosynchronous sat ellite at 22300 miles 35900 km orbital altitude above the equator results in a 0 25 s round trip delay which greatly affects two way speech communications This large satellite delay does not bother one way communications ser vices such as audio and video broadcast data transmission or paging Data transmission protocols do have to account for the delay In 1998 there were low earth orbit LEO voice communications satellite s
24. 048 Mbits s E1 systems and 1 544 Mbits s DS1 T1 systems The eight channels unused by DS1 systems still allow economical system implementa tion The 64 in the name corresponds to 64 ms of round trip tail end delay between the echo canceller and the network reflection source This is sufficient to permit deployment in the vast majority of networks TECO3264 meets the ITU T G 168 objective tests with considerable margin More importantly it meets the subjective expectations of the human listeners in real networks under real call conditions For non speech communications such as voiceband data modem traffic the modems can turn off the echo canceller functions on a per call per channel basis if desired using ITU T recommended 2100 Hz signal ing The TECO3264 has a large number of parameters and settings 186 parameters in the read write regis ters and 28 parameters in the read only registers However only a handful will be of interest if it becomes necessary to fine tune echo canceller vari ables for unusual or unique network characteristics in obtaining maximum performance Settings such as those for pulse code modulation PCM parameters for u law or A law coding for u law to A law and A law to u law conversions and timing alignment are set and forget The recommended set of echo can celler parameter settings are based on years of experience with real networks and accommodate a wide range of normal network conditions For ex
25. 264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Facility Refers to a telephone transmission link line trunk including the terminal electronic equipment and the intercon necting media wire optical fiber radio Some facilities also have intermediate electronics when the distance is too great to span between the terminal equipment In the past the signal delay through a facility was due mainly to the propagation time through the media with a small contribution due to the attached electronics Switch The machine that connects one telephone line or trunk to another line or trunk to build up an end to end call Mod ern digital switches are basically very specialized ultra reliable computers that connect four wire bidirectional 64 kbits s DSO paths to one another Analog two wire switches are still in service and still work very well for local POTS The local switch is usually called the central office and connects POTS lines to other local POTS lines or to long distance toll trunks The long distance or toll switching machines usually just make trunk to trunk connections among themselves and other toll switches or local switches Some toll switches may make direct connections to large company customer private networks and bypass the local switching systems Network echo cancellers are usually located on the long distance side of long distance toll switches pointing with the near end or C and Y po
26. During periods when near end speech is detected the NLP and comfort noise are removed imme diately Thresholds for determining when only far end speech is present are set to eliminate clipping of near end speech The thresholds depend on the relative level difference between near end and far end signals These thresholds are defined as SBETA and FBETA The NLP threshold in dB is given by NLP Threshold in dB 20 logio B 8192 where B is equal to either SBETA or FBETA There is also a transient echo control option designed to remove echo during the initial start up of echo canceller convergence This is engaged whenever the power measure of Eout greatly exceeds power measure of Yin a situ ation that can be only a transient one 6 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual TECO3264 Functions continued Linear Processing continued Noise Matching Noise matching is provisioned all channel and used to alleviate the effects of the NLP switching in and out as heard by far end talkers when they start and stop talking When selected it automatically measures the near in noise level and instead of inserting infinite loss in the echo return path inserts an equivalent amount of noise This inserted noise is sometimes called comfort noise From experience customers expect to hear some low level noise during the gaps in speech and may think a totally quiet line has been discon
27. Information Manual August 1999 microelectronics group Lucent Technologies Bell Labs Innovations TECO3264 32 Channel Echo Canceller Information Manual Introduction This Information Manual supplements the detailed technical information in the August 1999 TECO3264 32 Channel Echo Canceller Data Sheet DS99 241PDH which contains parameter values micro processor interface electrical characteristics etc with a brief description of the TECO3264 echo can celler integrated circuit discussions of network elec trical echo canceller applications and of background information on network characteristics that affect echo cancellers Both traditional long haul telephone networks and newer voice telephony transport sys tems are covered The third document in this series is the TECO3264 Evaluation Board System User Manual which describes a stand alone test system for the TECO3264 The TECO3264 echo canceller has been developed to provide an efficient economical state of the art solution to electrical echoes in the telephone net work It has the flexibility to deal with the traditional long distance network and also with new and emerg ing networks where the sources of delay may not be distance related and where the network design approaches do not match older standards and con figurations The 32 in the name reflects its capacity for up to thirty two 64 kbits s DSO channels This allows the TECO3264 to be designed into both 2
28. Q 2 16 uF see above Modern local digital switches segregate the loops and hybrids into non loaded and loaded populations and use separate nonloaded and loaded balance networks These two new bal ance networks are still simple with only three or four passive resistors and capacitors C 55 lt ZRCV ZH4 Zu Lee EL BALANCE Pal eene TER W NETWORK BRIDGE z BN re CS ZLOOP ZCUST L_ Zm ZBN ZLOOP Y S lt ZSEND 5 7082 F r 2 Figure 12 Hybrid Two to Four Wire Converter Lucent Technologies Inc 39 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Hybrids Electrical Description continued The two wire ports of the hybrid are usually designed to present the standard 900 Q 2 16 uF impedance ZH2 on both sides This can be shown to be the transformation of the input impedances ZNET typically 600 Q in parallel looking back into the network from the four wire ports of the hybrid plus some additional resistances and capaci tance The input impedances of the four wire ports of the hybrid ZH4 are the parallel transformed versions of ZBN and ZLoop The transformer windings are connected such that the phases of the loop and balance network port reflections tend to cancel at the outgoing four wire port Skipping the details it can be shown that ideally half the power input to the four wire port goes to the loop
29. S weer AND CONTROL LOGIC NEAR END L SIGNAL AND ECHO E Zeie NONLINEAR HNEAR Y FAR OUT PROCESSOR NEAR IN NOISE 6 dB IN nx el SEND OUT LINEAR MATCHING Al Al LINEAR SENDIN l l l SEND PATH 5 6158 F ar 3 Figure 1 Echo Canceller Functional Diagram 4 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual TECO3264 System Interfaces and Formats PCM Interface Lucent Technologies Inc The TECO3264 provides two independent serial pulse code modulation PCM input ports and two indepen dent serial PCM output ports The input ports are labeled X equivalently far in or receive in and Y equivalently near in or send in The output ports are labeled C equivalently near out or receive out and E equivalently far out or send out refer to EC diagram in Figure 1 Echo Canceller Functional Diagram on page 4 Input port X and output port C are the receive path of the echo canceller Unless there are optional encoding conversions between u law A law and linear the signal passes from X to C unchanged Input port Y and output port E are the send path and contain the echo canceller subtractor and nonlinear processor functions Ports X and E face the far end of the circuit Ports Y and C face the near end of the circuit which may have a 4 wire to 2 wire conver sion hybrid circuit Hybrids are a main source of networ
30. STN and long distance net works Some military networks use four wire switches facilities and station sets which enables them to avoid echo control problems PSTN Circuit Switching versus Data Network Packet Switching The type of switching used in the PSTN is called circuit switching which means that when a customer places a call a path is constructed from the customer s switch port connection to the called party switch port and it is dedicated to that call as long as both parties are connected even when no one is talking The path through the network may vary greatly from one call to the next for the same pair of customers but it is fixed for a given call In contrast data networks have been designed to use packet switching where the connection may vary from one data packet to the next during one session between two customer data ports More and more voice traffic is being carried on packet oriented data networks with the Internet being the best known example The delay may vary between data packets carrying voice information 32 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Echoes It can be exciting and fun to hear one s voice come back as an echo from a canyon wall a cliff or big building However to hear an echo during a telephone call can vary from merely distracting to annoying to disrupting the flow of conversation complet
31. Speech Coding A Tutorial Review A S Spanias Proceedings of the IEEE Vol 82 No 10 October 1994 pages 1541 to 1582 Telecommunications Transmission Engineering Volumes 1 3 Bellcore ISBN 1 97808 04 02 1990 Packet Based Multimedia Communications Systems ITU T Recommendation H 323 Geneva 1998 Lucent Technologies Inc 49 For additional information contact your Microelectronics Group Account Manager or the following INTERNET http www lucent com micro E MAIL docmaster micro lucent com N AMERICA Microelectronics Group Lucent Technologies Inc 555 Union Boulevard Room 30L 15P BA Allentown PA 18103 1 800 372 2447 FAX 610 712 4106 In CANADA 1 800 553 2448 FAX 610 712 4106 ASIA PACIFIC Microelectronics Group Lucent Technologies Singapore Pte Ltd 77 Science Park Drive 03 18 Cintech III Singapore 118256 Tel 65 778 8833 FAX 65 777 7495 CHINA Microelectronics Group Lucent Technologies China Co Ltd A F2 23 F Zao Fong Universe Building 1800 Zhong Shan Xi Road Shanghai 200233 P R China Tel 86 21 6440 0468 ext 316 FAX 86 21 6440 0652 JAPAN Microelectronics Group Lucent Technologies Japan Ltd 7 18 Higashi Gotanda 2 chome Shinagawa ku Tokyo 141 Japan Tel 81 3 5421 1600 FAX 81 3 5421 1700 EUROPE Data Requests MICROELECTRONICS GROUP DATALINE Tel 44 7000 582 368 FAX 44 1189 328 148 Technical Inquiries GERMANY 49 89 95086 0 Munich UNITED KINGDOM 44 1344 865 900
32. a k a network communica tions and synthetic Broadcast usually has analog bandwidths from 5 kHz to 15 KHz depending on the service AM radio FM radio television etc and relatively flat frequency response requirements Linear PCM encoded broad cast usually takes more than 64 kbits s before any compression such as MUSICAM for MPEG is applied Toll quality includes law and A law encoding at 64 kbits s and can be achieved with an increase in distortion with 16 kbits s ADPCM Communications quality speech is sufficient for high reliability information transfer but with some loss in naturalness Companding schemes with less than 16 kbits s and more than 4 kbits s usually fall in this category Digital wireless cellular systems such as GSM use 13 kbits s speech companding Customers com plained about the 8 kbits s companding used in the early deployments Synthetic speech at the current state of the art having less than 4 kbits s bandwidth has obvious computer or robot characteristics See reference 6 Quantization Distortion Units qdu The distortion resulting from one ideal analog speech to 8 bit u law or A law to analog encoding and decoding process is defined ITU T Recommendation G 113 as the standard against which other digital speech processing schemes are measured The term is quantization distortion unit qdu for one law or A law encoding and decoding and corresponds to a 35 dB signal to distortion ratio Simply inserting a digit
33. al loss pad is allocated 0 7 qdu Thus when the TECO3264 achieves nearly 34 dB ERLE after internal conversions D A conversion before the network hybrid A D on the return from the hybrid and finally echo cancellation that is all that can be done in an ideal situation For comparison a 32 kbits s ADPCM encoding and decoding is allocated 3 5 qdu Low bit rate voice companding schemes have even higher distortion than 32 kbits s ADPCM In general the lower the bit rate the higher the qdus The planning rule for an international telephone connection is to have no more than 14 qdu for end to end distortion from digital processes alone not counting such analog impairments as loss versus frequency attenuation distortion background noise etc High Speed 56 kbits s Voiceband Modems The new 56 kbits s high speed voiceband modem schemes attempt to match their multiple coding levels to the dis crete levels of u law and A law to achieve the maximum rate possible through voice frequency designed channel toward the customer Since the interoffice network is almost entirely digital now only one analog to digital conver sion and one digital to analog reconversion is experienced on many telephone calls not the 2 3 or 4 tandem encodings of the past Voiceband modem designs are trying to work with the existing digital voice coding schemes instead of trying to fight them by using every data compression scheme ever invented The catch is still that call co
34. am ple there may be cases where hybrid echo return loss ERL is consistently lower than the expected minimum of 6 dB The minimum threshold return loss may be adjusted to the appropriate level by setting two parameters ADATA and AVOICE to optimize performance for the lower ERL range The vast majority of parameter settings will never be touched but will be available if ever needed to adjust for different conditions in new wireless land based or satellite packet or Internet applications As refined recommendations for parameter set informa tion are developed for new environments they will be made available The built in u law to A law and A law to p law conver sions of the TECO3264 can be used at the gateways between u law and A law coded networks as a bonus to the basic echo canceller functions A very minimal knowledge of the network in which the TECO3264 will be deployed is helpful to obtain maximum benefit of the available performance but it is not critical for one time provisioning 1 Minimum echo return loss ERL typically 3 dB to 6 dB with a 6 dB recommended default if the minimum ERL value is not known 2 If the network background noise characteristics are known the noise matching options may be adjusted for an optimum comfort noise match Otherwise the recommended settings should prove satisfactory for normal network conditions Information Manual TECO3264 32 Channel Echo Canceller Information Manual Augus
35. ancel lation functions built in to the computer telephone board is the means for echo control at this end see Packet Transmission Impairments on page 26 Internet to Internet Telephony Connections When an Internet call connects to another Internet caller using microphones and speakers connected to a com puter there may not be any hybrids to produce any electrical echo Acoustic echo cancellation would be the sole means for echo control in this case Public Network to Public Network via Internet Long Haul When the Internet is used as an alternative to the traditional long haul circuit switched networks the Internet tele phony service provider must provide echo control just like the traditional network Lucent Technologies Inc 25 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Packet Transmission Impairments Packet transmission introduces its own new impairments to telephony Like the usual digital transmission in tradi tional circuit switched telephony individual bits or blocks of bits can be hit by interference inverted and be in error In circuit switching the customer to customer path is fixed for the duration of the call and the end to end delay is fixed In packet transmission each packet can in theory travel a different path through a complex busy network between the source and destination with each experiencing different delays Packets sent earlier may arrive after packets sent later
36. arly 1960s bits were precious and higher bit rates translated to higher analog bandwidths on the copper wire transport and thus more loss Econom ical high quality voice speech transport was the driving goal Digital data carried within a voice channel was at such blinding speeds as 110 baud and didn t really come into stretching what was needed for voice considerations As mentioned 3000 Hz is high enough for good quality speech Shannon s sampling theorem says that one must sample at some rate greater than twice the maximum signal content frequency to fully represent a signal A sam pling rate of 8000 samples per second was a logical choice That was easy and quick The next question was how many bits are needed per sample To shorten a longer story 8 bits per sample is plenty for good speech but not with equal sized encoding steps per digital sample Multiplying 8000 samples per second by 8 bits per sample yields the now familiar 64 kilobit per second DSO voice channel rate There was still the problem of a wide range of volume levels in human speech and that strained the eight linearily encoded bits per sample Also the network was entirely analog It would be a long long time before we had the universal digital interswitch transmission and digital switch network of today It is still analog from the switches to the vast majority of customers of course It was real ized that a long distance call that crossed the continent would be made up of
37. ation Manual August 1999 Architecture and Functional Description of TECO3264 Description The TECO3264 32 channel echo canceller device is a 3 3 V CMOS very large scale integration VLSI component offering 32 independent channels of echo cancellation The TECO3264 provides echo cancellation for both DS1 and E1 systems The device operates from a single 3 3 V supply and requires only an external 8 192 MHz clock and 8 kHz frame sync Figure 1 shows the basic functions of the TECO3264 Packaged in a 160 pin plastic metric quad flat pack with heat speader MQFPH and handling 64 ms of tail end delay this device combines high performance high channel density low power consumption high flexibility excel lent maintenance capability and low cost into a single package Built in self test logic affords easy device verification while integrated boundary scan capability reduces board ver ification time and cost A high speed microprocessor interface and full user provisionability on device pins provide maximum flexibility 1 l TAIL END lt RECEIVE PATH gt FAR END i NEAR END i Alu x c FARIN XIN HIGH GN NEAR OUT RECEIVE IN LINEAR GC 0dB 6 dB 9 wje oo AND 12 dB pe O CONVOLUTION UPDATE PROCESSOR CUP ECHO ESTIMATOR microprocessor status X REGISTER INTERFACE ALARM H REGISTER f jr PARAMETER CS CONTROL dE CLOCK REGISTER
38. called a loop referring to the full loop path for network supplied direct current to power the sta tion sets and to provide loop closures to indicate the handset has gone off or on hook Large business or government entities may lease four wire wire wireless or optical facilities from telephone ser vice providers and bypass the local telephone company and its two wire loop plant Four Wire Circuit Four wire in telephony used to always mean two pairs of wires used for a two way communication circuit Signals on each pair are unidirectional that is one pair carries the signal in one direction in to out left to right east to west and the other pair carries the signal in the other direction out to in right to left west to east The telephone system has evolved from an all analog network to nearly all digital except for the customer two wire loop and home telephone sets The term four wire now usually refers to the two unidirectional paths inside a digital transmission system such as DS1 or E1 based multiplexed carrier systems or inside digital switching systems Voiceband speech or data signals are converted to binary pulse coded modulated signals usually but not always at a 64 kbits s DSO rate The signal bits appear between logic gate outputs and ground not on two wire pairs but the term four wire circuit is still used to indicate two unidirectional paths Telephone Station Sets Hybrids Side Tone Telephone sets are both two wir
39. ch 186 miles or 300 km traveled Radio waves in air travel at essentially vacuum light speed Electrical sig nals in wire or coaxial cable or light waves in optical fiber travel at slower speeds Electrical signals in wire travel at about one half the light speed or about 100000 miles per hour depending on the cable insulation and structure This gives about 1 ms of delay for each 100 miles one way or 2 ms for round trip delay for an echo Older elec tronic terminal equipment switches amplifiers and the like contributed much smaller delay than the distance A long distance call of 1500 miles would have a round trip delay of about 30 ms and thus needs echo control Now there are new sources of delay in the terminal and transmission equipment that result in echo control being needed even for short distances Low bit rate voice LBRV compression techniques to use 8 kbits s or even less for voice circuits require time to perform the compression and decompression LBRV techniques tend to be used in wireless telephony where bit rates and analog bandwidth to the mobile transceivers is a precious resource Packet voice transmission through ATM the Internet or other packet networks means that the voice signal is bro ken up into small packets before being transmitted Each packet may travel a different path to the destination and experience different delays through switches and transmission links Packets may arrive at the receiving end out of order Th
40. cho Canceller Information Manual August 1999 Echo Canceller Background continued Hybrids General Description Hybrid or more completely two wire to four wire hybrid or two wire to four wire termination set as used in voice band telephony is a balanced bridge network used to convert between two wire telephony circuits and four wire circuits Hybrids are used at the two wire ports of digital switching machines and digital carrier systems which are designed as four wire systems internally Signals coming in from the customer to the hybrid to its two wire port are sent out on the outgoing unidirectional port on the four wire side Signals coming into the hybrid on the incoming four wire port of the hybrid are directed to the two wire port to go out toward the customer Ideally none of the incoming four wire signal gets across to the outgoing four wire port However hybrids are real physical devices and some of the incoming four wire signal does leak across to the outgoing port If the leak or reflection in echo terms is large enough the other customer at the far end can hear his own voice even 50 dB down The loss from incoming four wire port to the outgoing four wire port is called echo return loss ERL In non echo canceller telephone transmission terminology this leak loss is called transhybrid loss The ERL as measured at an echo canceller includes any signal level adjustments up or down in the network going to the hybrid or returning as
41. cost have as much or more to do with political reasons than technical reasons Internet tele phony is not subject to most of the regulations and tariffs as the normal local and long distance networks and does not have the overhead costs of the embedded regulated networks The current political climate seems to be that the U S government is not willing to increase control over the free Internet The first major voice over Internet protocol VoIP network was put into service in late 1998 Echo control was a major engineering concern Internet to Public Connection In the voice telephony application the Internet is taking the place of the circuit switched networks From an echo control point of view the concerns are basically the same when the Internet originated call connects to a caller on the regular public network Transmission delay results from the distance voice bandwidth companding packet assembly switching and reassembly Whatever the origin of the delay it is still just delay like that from distance in traditional long haul circuits The result is that there needs to be echo control at the edge of the Internet connection facing the public network to deal with electrical echoes from hybrids An echo canceller such as the TECO3264 with normal settings would fit just as it does at the edge of circuit switched networks At the computer end any echo would be acoustic echo from speaker energy being bounced back to the microphone Acoustic echo c
42. ctor indicates that the EC is not converged 5 Each of the aforementioned indications are optionally enabled or disabled on an all channel basis Action takes place on a per channel per call basis 6 When the circuit is recognized as not converged by one of the EC convergence indications the convolution gain control circuit may apply an alternate provisionable convolution gain for a provisionable amount of adap tation time 7 When the circuit is recognized as not converged by one of the EC convergence indications the convolution gain control circuit may apply an alternate provisionable NLP differential sensitivity threshold for a provision able amount of adaptation time 8 When the circuit is recognized as not converged by one of the EC convergence indications the convolution gain control circuit may apply an alternate provisionable peak factor value to the gain normalization circuit for a provisionable amount of adaptation time 10 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual TECO3264 Functions continued Control continued Conirol of Channel Processing Features The TECO3264 system provides a per channel 64 clear channel or bypass option When enabled on a given channel the 8 bit data words for that channel will pass through unaffected This occurs in both directions of transmission and overrides all other channel control options The eight unused channels
43. d con trol signaling transport Companding The term companding is made up from two words compressing and expanding Companding in telephony is a technique used to preserve the dynamic level range of a signal such as speech when it must be transmitted through an electronic channel that has less dynamic range and may distort low levels or high levels The compres sor circuit at the transmitter end reduces the input level range by a stated rule The expandor circuit at the receiver end restores the original dynamic range by applying the inverse of the compressing rule A law and u law compan dors work by allocating more bits for low level signals than for high level signals thus preserving accuracy and a low background noise at the cost of increased nonlinearity for loud signals for which human hearing is more toler ant Other speech companding schemes include those which work on a syllable by syllable basis and those that use input output ratio processing like 1 dB output increase for 2 dB input increase and reverse at the receiver Low bit rate voice algorithms employ various level companding techniques to help reduce the data rate Companding can be divided into lossy or lossless categories in terms of how faithfully they can restore the original information at the receiving end Lossy companding is often used for speech music and video data compression and lossless for digital data archival Lossless companding is used for digital data files
44. d not happen as forecasted 1990 was proclaimed as the Year of ISDN in the U S A In 1998 ISDN was just beginning to be installed in significant numbers in some areas since the price and the available services such as Internet access matched up with customer desires and willingness to pay 24 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Internet Telephony Applications There is a forecast that by year 2002 perhaps 25 of long distance telephony traffic will be packet voice over the Internet rather than traditional circuit switched voice by long haul networks What began a short time ago as a low quality voice service offered for a low price by small start up telephony networks has quickly evolved to good qual ity voice service offered by traditional long distance network operators The large long distance networks are start ing to offer the Internet telephony services in competition with their normal circuit switched services in order to keep as much market share as possible For a price that is half or less than normal long distance services customers were willing to accept less than the usual voice toll quality Now after a few years of work on packet voice transport techniques the voice quality deliv ered over an Internet phone call has greatly improved At the same time the price has not increased The reasons for the much lower
45. e 404 Hz and 2804 Hz points measured end to end At least one telephone network does some boosting of the signal below 400 Hz and above 3000 Hz to restore more natural sounding speech This boosting can aggravate echo control problems because hybrid balance tends to degrade at the low end below 500 Hz and at the high end above 2500 Hz of the voiceband POTS Call Control Signaling Transport and Echo Cancelling In addition to the customer signals voice or data carried within the DSO channel the network has to transport additional information to allow network control of the call setup alerting of the called customer and final discon nect For POTS calls such information includes on hook and off hook signals to indicate whether the customer has picked up the telephone handset to initiate a call or to answer an incoming call and the presence of ringing typically pulses of 20 Hz coming from the originating switch For old style rotary telephone sets using dial pulse addressing for sending the calling number to the network the on hook off hook signal is modulated typically at 10 pulses per second Modern systems tend to use dual tone multifrequency DTMF dialing within the voiceband for sending phone number addressing When the network is connecting to a private customer switch called a pri vate branch exchange PBX additional call control signaling functions are required The on hook off hook and ringing signals are at very low frequencies
46. e and four wire The connection to the telephone network for the vast majority of analog station sets is via a two wire metallic cable pair called a loop see Two Wire Circuit above The connection to the human side is four wire with two wires to the microphone and two wires to the earphone Thus the ordinary telephone must have a two wire to four wire hybrid conversion circuit inside it If the station set hybrid were per fectly matched to the impedance of the connecting two wire loop none of the speaker s voice energy from the microphone would go to the earphone In fact the hybrid has a deliberate carefully controlled mismatch to the loop causing a fraction of the speaker s voice energy to be heard in the earphone This designed in microphone to ear phone reflection is called side tone Since the delay is very very small the speaker listener is usually not aware that it even exists unless the circuit fails and there is no side tone in which case the phone sounds dead Tele phone systems designers discovered that if the customer hears his own voice in the earphone he will tend to con trol his speaking volume much more so than if he doesn t hear his voice This more controlled range of speech volumes helps in designing the telephone equipment and setting the network operating parameters Side tone helps the customers feel better and the network work better and cheaper Lucent Technologies Inc 37 Information Manual TECO3264 32 Channel E
47. e hybrid and back to the echo canceller send in Y port With no speech compression packet switching or forward error correction the switches and other transmission terminal gear contribute little to the delay The 64 ms tail end delay capacity of the TECO3264 will account for the greater majority of such local to toll connections Note that the long haul delay inside the toll network is not addressed by the echo cancellers even though it can be hundreds of milliseconds if geosynchronous satellite links are included The echo cancellers are intended to pre vent the echoes from the hybrids from getting back to the long delay links More and more intercontinental voice and data traffic is being moved to an ever growing network of undersea optical fiber cables where the delays are much shorter but not negligible and the bandwidths are much greater than the radio links through satellites Wavelength division multiplex techniques are allowing many light signals in one fiber with terabits of digital data transport capacity Such optical undersea cables are nearing deployment with all but Antarctica on the schedule While the occurence of long 100 ms delays may decrease the long haul delays just due to terrestial distances will still be enough to require echo control LONG HAUL NETWORK PUBLIC SWITCHED PUBLIC SWITCHED TELEPHONE NETWORK TOLL ECHO FACILITIES ECHO TOLL TELEPHONE NETWORK PSTN SWITCH CANCELLER PSTN AND CANCELLER SWITCH SWITCHES
48. e on an all channel basis a The threshold value is provisionable on an all channel basis Near End Speech Detector NES The TECO3264 system provides two independent indications that near end speech signals coming in from the 2 wire circuit via the hybrid and into the Y port is present These indications are combined with other indications to control convergence a The first indication of near end speech is derived by comparing the X input power measure to Y input power measure When the Y input power measure exceeds the X input power measure by a provisionable threshold value near end speech is declared The H register values are frozen during NES TRUE The H register values may adapt during FES TRUE and NES FALSE The threshold used in the first indication should be selectable from one of two all channel provisionable values The means to select between the two values will be the narrowband energy NBE detector state or alternately an external voiceband signal classifier VSC input indication The voiceband signal classifier with the narrow band energy detection decides whether a wideband signal such as speech or a narrowband signal such as a tone is present An external pin is provided for the VSC serial control input a The second near end speech indication is provisionable on an all channel basis Narrowband Energy Detector NBE a The narrowband energy detector is used to classify the far end X input signal as wide or
49. e packet reassembler must account for the maximum differential delay of the packets before reconstruct ing the signal If a packet voice transmission system also uses LBRV the echo control concerns may be com pounded when the packet network is linked to traditional local telephony networks Large scale integration has made not just digital error detection feasible but also error correction for transmission errors that occur between sender and receiver Forward error correction requires storing long strings of input at the transmitter and inserting redundant data codes and structures At the receiver the input string must all be ana lyzed errors detected and then the corrected bits substituted The bits on the data stream may also be inter leaved or very carefully mixed before transmission H a burst of interference produces errors in a long block of data the deinterleaving process at the receiver distributes the errored bits as single bit errors which are easier to detect and then correct All of this takes time and means delay in the terminal equipment For some types of ser vices such as one way television and data transmission such delays are of much lesser concern than error free reception When two way voice links are run through error corrected links echo control can become a concern 42 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued
50. e with new facility installations being four wire digital paths There are many left in service in the embedded plant because they still work 44 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Linear Nonlinear Nonlinear Processors u law A law The nonlinear processor is an important complement to the echo cancellation function that has been carried over from echo suppressor system design The adjective nonlinear is used for good mathematical terminology reasons Further the u law and A law coding and encoding rules for the payload signal carried in a 64 kbits s DSO signal are also nonlinear processes For a system with an input and an output to be called linear it must obey a rule If the input signal is multiplied by some factor before entering the system the output of the system will be multiplied by that same factor A result of this property is that no frequency component will appear in the output that does not appear in the input signal as viewed in the frequency domain Some components may appear in the output with nearly zero amplitude i e they have been filtered out by the system Even these tiny components increase or decrease with the input level If an electronic system is not linear the frequency components of the input signal are mixed by the system to produce new components that are the sum and difference frequencies of the inp
51. echo cancellers at the analog stages of the cell phone before encoding to digital Residual acoustic echo can be reduced by adjusting TECO3264 parameters 22 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Hybrid Fiber Coax HFC Telephony Figure 10 depicts a simplified view of a hybrid fiber coax HFC telephony connection Here is another use of the word hybrid to mean a mix of transmission media in this case metallic coaxial cable of the type used to carry tele vision signals in electrical form and optical fiber for signals in light form The cable television industry has been working for years to develop technology to transport telephone communica tions on their wideband coax and fiber facilities along with the television channels If they could do that they could support POTS like telephone service and tap into the huge telephone system revenues and bypass the traditional local telephone companies For various reasons including technical political reliability funding concerns HFC telephony has not caught on in any major way As of June 24 1998 that may well change The largest long dis tance telephone network announced it is acquiring the second largest CATV system for 48 billion With this size of financial commitment HFC telephony could well turn out to be a major part of voice telephone communications in the years to come
52. eeeseceaaeeeeeeeaaaeeeeeaaeeeeeseaeeeeeeaaes 26 Echo Ganceller Background isureid ia aeia deed ea ae di EENS eedae deeded 27 Echo Ganceller Term S aeaea A geseet Ree Seege Ree tad suetieiitws cdasateruecodensssteasateneectendarimtaateaes 27 Plain Old Telephone Service POT 31 POTS Loops Metallic Pairs Of Today 31 lg AE cere eter rere teeter EE E AEAEE coer EE teeter T E rere ceri ccecr terre reer eect cee ee eee 31 ul EE 31 ele e EE 32 le EE 32 Central ue 32 Public Switched Telephone Network DGTN nnn 32 PSTN Circuit Switching versus Data Network Packet Switching ccccscceeeeeeeeeeeeeeeeeeeeeeaeeeeeeeeeeaaeeesaeeenaaes 32 ENEE 33 DONA iss dented teed bea elie E avi ahs kite E A sults ta eaube A E E N rakes h TE 35 The Telephone System Not Perfect for Very Good Heasons nnt 36 Two Wire Ci eUi a r e aa ohh ivecducht sst eet ei Seege leede S ilen eb aan aaae ia eee 37 FOUR Wires ee 37 Telephone Station Sets Hybrids Side Tone cccccccesssececeseeceeeecseeeeeeceeeeeeeeeceeeeeeesseeeeeeeseceneeeeesseseeaeesessnnaeess 37 Hybrids General Description c cceccceceeeceeeceeeeeneeeceaeeeeeaeeeeeaeeeeeaaeeeaeeeeeaaeseeaeeesaeesseaaesecaeeeseseeeeesiaeeesaeseeaes 38 Hybrids Electrical Descriptions s E E T E gbeee ghedavssteeedavtviaepevs LES deer 39 YCHAIN WEE 41 POTS Call Control Signaling Transport and Echo Cancellling cccccccceeeeeeeeeeeeceeeeeeeeeeseeeeeeeaaeseeneeessnaeeeeeaes 41 Delay So
53. einserted before continuing in the send direction For older interoffice digital carrier systems connecting local switching machines within a network operator area the situation for local calls is was the same as for the digital loop carrier The call control information has had to be car ried along with the voiceband payload on a call by call basis For modern networks once a call has progressed into a modern digital switch possibly on its way to a toll network the call control information extracted and is routed to an entirely separate signaling network Look up signaling system 7 in a telephone system reference At the receiving end the signaling must be reinserted into the DS1 E1 carrier if a digital loop carrier is involved Network echo cancellers systems are often deployed at the edges of long haul networks and the local telephone networks H as mentioned above all the call control signaling for the traffic that the echo canceller will see is trans ported in the separate signaling network the echo canceller would not have to deal with preserving either DS1 robbed bit signaling or E1 common signaling channels Delay Sources Traditional Distance Speech Compression Packet Switching Error Correc tion Etc The main source of signal delay in traditional telephone circuits came from the simple fact that nothing travels faster than light in vacuum about 186000 miles per second or 300000 km s This gives one millisecond of delay for ea
54. ely There are three conditions necessary to have an echo perceived 1 some sort of reflector which redirects some of the passing energy electrical or sound back toward the source 2 sufficient returned energy relative to speech and background noise levels to be detected and 3 sufficient round trip time delay between the utterance and its return to the listener s ear How bad a telephone call echo is perceived depends on the relative loudness of the echo and also on the time delay For telephone echoes the reflector is most often a device called a hybrid which converts from a four wire electrical transmission path to a two wire transmission path See Four Wire Circuit on page 37 and Two Wire Circuit on page 37 Practically every telephone customer is connected to the local serving telephone company switching machine by a single pair of metallic usually copper wires or a two wire path This pair of wires is a bidirectional path carrying both directions of signal at once Modern telephone switching systems and transmission systems use two unidirectional paths internally called a four wire path These four wire paths may be on copper wires fiber optic cables or radio waves Some military telephone networks use all four wire circuits and four wire telephone sets and thus need no hybrids So when a customer call connection is made there must be a hybrid to get from the first wire pair into the telephone network and another hybrid at the far end
55. ephony network may have audio compact disk quality 16 bit linear analog bandwidth of 20 Hz to 20 kHz and all four wire transmission with no reflecting hybrids but not for a long time In the mean time devices such as echo cancellers make good engineering and economic sense 36 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Two Wire Circuit Two wire in telephony literally refers to two metallic wires in a pair that are used for a two way communication cir cuit The wires are usually copper and usually twisted to reduce outside electrical interference to and from other pairs and other electrical noise sources A two wire telephony circuit is usually bidirectional and full duplex that is signals travel in both directions in out left right east west at the same time Voice circuits are usually symmetric in bandwidth that is they have about 3000 kHz or 64 kbits s available in both directions to carry signals Data channels are not necessarily symmetric A 15 1 asymmetry in two way data bandwidth requirements can happen Speech and voiceband computer data modem signals are two common signals Nearly all of the several hundred million telephone customers in the world are connected by a single two wire pair from their home or business all or at least part of the way to their local telephone switching system This two wire connection is often
56. essor interface to reset the H register on a per channel basis 2 In addition to the microprocessor interface control an input pin is provided to globally reset the H registers for all channels A mode is provided to also allow per channel serial control of the H register reset function Lucent Technologies Inc 9 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 TECO3264 Functions continued Control continued Nonlinear Processor Control The NLP feature in the TECO3264 system is controlled by several sources The following is a list of these sources 1 Per channel NLP disable an external pin is provided to allow per channel serial control of the NLP Residual echo e vs X calculation this is the dominant means of NLP control If the power measure on X after coprocessor exceeds the power measure on e by a certain provisionable threshold value the NLP is operated 3 The differential sensitivity threshold used in the vs X calculation has the capability of taking on two indepen dently provisionable values depending on the convergence mode refer to the Convolution Gain Control Fast and Slow Convergence Modes section on page 10 4 Per channel microprocessor control the NLP can be controlled via the microprocessor interface on a per channel basis This control either allows real time control of the NLP by the echo canceller or turns the NLP on full time 5 A mode is provided to freeze
57. ex echo control systems precision hybrids are used as the echo control devices of choice With modern echo cancellers it is possible to design a long distance network that has a 0 dB loss from end to end and insert echo control at the ends that interface with the local networks having the troublesome hybrids An echo canceller in a traditional long distance network will be placed at each end of the link with each facing the nearest local network The round trip delay between the echo canceller position and the local network hybrids is seldom more than 64 ms Each echo canceller only has to account for this local tail end delay to its hybrid not for the delay for the portion of connection between the echo cancellers The long distance portion can be hundreds of mil liseconds if satellite links or very heavy signal processing or bit rate compression is involved If the total delay gets into large fractions of a second even with echo cancellers the human speaking protocol may have to become more formal and less natural in flow with military type conversations with over at the end of each sentence so the listening party knows when the speaker is done For each call the long distance trunk echo cancellers will be connected to a different local telephone trunk and then to a different local customer with his own hybrid two wire loop and customer premises equipment Thus for each call the echo cancellers may see a different hybrid with different echo retur
58. few hundred feet to miles long construction of fine gauge wire like 26 AWG to mixtures of gauges connected unused parallel shunts called bridged taps to added inductors called loading coils on long loops The input impedance among loops varies considerably The approach to a reasonable hybrid balance network that does a good enough job on local calls is an electrical circuit called a compromise balance network that provides a hybrid mid band 500 Hz 2500 Hz echo return loss of about 15 dB averaged across the population of loops connected to a switch and with a minimum in the range of 6 dB for the worst loops The return loss or impedance balance degrades below 500 Hz and above 2500 Hz The minimum return loss not the maximum or average is the key because that is the cause of the trouble Ninety nine great hybrids and one bad is still bad A hybrid with 50 dB echo return loss across almost all the voice band but only 2 dB at any one frequency is still bad More modern digital switches segregate the loops into two groups nonloaded about 80 in the U S the shorter loops and loaded about 20 the longer loops gt 18000 ft with inductive loading Europe has no loaded cable Using separate nonloaded and loaded compromise hybrid balance networks provided about another two dB of echo return loss on the average and some improvement in the minimum for a given loop plant served by a switch ing machine This is still not good enough to prevent ec
59. for which a one dB increase results in only a 0 5 dB increase in the output level may be a definition of the upper limit for acceptable linearity One way to express the linearity nonlinearity of an amplifier is by its total harmonic distortion which is a measure of the new frequency components that appear in the output and their relative magnitude to the original signal compo nent output portions A high fidelity amplifier may have a harmonic distortion of a very small fraction of a percent at its maximum specified input output level For speech the linearity can be a few percent distortion and still be virtu ally undetectable so telephone amplifiers can economize there also Of course a narrow bandwidth and relatively high distortion will wreak havoc and set limits on fancy schemes to cram ever more digital data into a communica tions channel optimized to the hilt for speech These high rate voiceband data techniques use very small granula tions in amplitude and phase to differentiate distinct valid data points and cry for all the bandwidth possible to maximize data throughput For human speech the ear brain processor seems to ignore phase information Lucent Technologies Inc 45 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Linear Nonlinear Nonlinear Processors u law A law continued When the digital transport for telephony was being developed in the 1950s and e
60. goes off hook the loop impedance seen at the hybrid will change thus changing the ERL In non echo canceller telephone transmission terminology the term echo return loss ERL is also used for a mid band 500 Hz 2500 Hz weighted return loss measurement referenced to the standard network input compro mise balance impedance or other test impedance of choice The North American standard network two wire input impedance is a 900 resistor in series with a 2 16 uF capacitor often referred to as 900 2 This double definition of ERL sometimes results in minor confusion when the two transmission subcultures communicate The corresponding low band 200 Hz 500 Hz and high band 2500 Hz 3000 Hz impedance comparisons are called singing return loss low SRL LOW and singing return loss high GPL HIGH SRL LOW and SRL HIGH values tend to be lower than the impedance ERL measurement and these low and high frequencies tend to be the source of the echo canceller ERL value 28 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Echo Canceller Terms continued Echo Path Delay td also Called Tail End Delay or Tail Length Echo path delay td also called tail end delay or tail length is the delay from the ROUT C port to the SIN Y port due to the delays present in the echo path transmission facilities including dispersion time due to the network ele
61. his section is meant to supplement the tutorial material in these other echo cancellers or general telephony sources see reference 2 not to replace them Some good tutorials on various current telecommunication topics may also be found at http www webproforum com The subtopics are intended to be read independently As a result a few points are repeated occasionally but fuller discussions are under one heading General The basic theory of echo cancellation dates back to 1966 when it was first proposed to build a voice frequency echo canceller for one voice channel which resulted in a rack of circuitry that consumed a high amount of power This was state of the art Thirty years later a single application specific integrated circuit ASIC can handle 32 voice channels with milliwatts of power per channel An echo canceller forms a replica of the echo by passing the far end signal through an adaptive filter that attempts to match the characteristics of the hybrid and the facilities between the canceller and the hybrid see Figure 11 Echo Canceller Standard Four Port Configuration below This replica is then subtracted from the signal that enters the near end port RECEIVED PATH c RIN tr Lain GEN ECHO ESTIMATOR AND OTHER CONTROL CIRCUITRY LRET ey E NONLINEAR Sout SUBTRACTOR PROCESSOR SEND PATH Lres 5 6157 F r 3 Figure 11 Echo Canceller Standard Four Port Configuration Figure 11 shows
62. hoes when the hybrid and the connecting two wire loop are part of a long distance circuit However since the switch has uses two hybrids to connect two customers the total increase of 4 dB is critical to the design of a 0 dB loss from switch two wire output to two wire output Otherwise the switch would have to insert loss or often sound like a barrel 38 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Hybrids General Description continued The trend in hybrid echo return loss seems to be deteriorating rather than improving with time The traditional long distance networks depended on the careful reasonably predictable hybrid echo return loss as described above and provided by local switching machines With the introduction of voice telephony by packet systems over the internet the connection to other private and public networks in which the concept of good overall network and hybrid design for echo control is a lost art there have been instances reported of echo return gain Fortunately a modern echo canceller such as the TECO3264 has parameters that can be adjusted to accommodate less than desired network performance Hybrids Electrical Description As mentioned in the general discussion of two wire to four wire hybrids they are impedance balance networks Figure 12 below depicts a hybrid network as a block with the impedances with i
63. i son Grey and others developed better microphones earphones and other equipment Stowger a funeral home director invented the electromechanical switch controlled by customer dial pulses The quality of voice transmis sion improved and the transmission distance increased and increased until today At each stage design decisions had to be made For the telephone to be successful it couldn t be just a business or rich man s tool It had to be cheap enough for everyone to be able to afford it Universal service was a real goal The vast bulk of the equipment and investment was and is in the local network serving local POTS customers While four wires to the customer and four wire terminal equipment as used in some military systems would be nice two wire would serve and be a lot cheaper Local switches were also two wire While short toll trunks could be two wire long distance transmission required four wire transmission and switching machines so amplifiers could be inserted This meant hybrids had to be introduced into the interface between long distance and local networks Hybrid design impedance matching and return loss control were developed into fine arts with exacting design rules to minimize echoes on long distance calls where the distance meant transmission delays Hybrids in two wire toll switches had to meet terminal balance requirements where the average midband return loss for a group of trunks had to be 18 dB The minimum for any
64. ic hybrids with external feed for dc and ringing currents are now the norm The transmission analysis for transhybrid loss echo return loss is still the same The old reliable rule of thumb of an 8 dB minimum transhybrid loss seems to be gone with the number these days often coming in at 6 dB 3 dB or less 40 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Voiceband Voiceband in telephony usually means the signal spectrum from about 300 Hz to about 3000 Hz sometimes given as 3400 Hz Human speech in the air may go from approximately 20 Hz to 10 kHz Two wire copper pairs can carry electrical signals from 0 Hz DC to 30 MHz and more For human understanding of speech the range from 300 Hz to 3000 Hz is adequate for acceptable quality The connecting circuits in the telephone sets the transmis sion systems and switches filter out frequencies below 300 Hz and above 3000 Hz to reduce the overall cost and complexity of the network The sampling that is done as part of the analog to digital conversion cuts off rapidly above 3000 Hz and the signal is essentially gone by 3400 Hz Legal tariffs that specify voiceband telephony ser vices are often written in terms of the loss at 404 Hz and 2804 Hz relative to the end to end loss at 1004 Hz For example a high quality private line service could allow up to 1 dB above and 4 dB below 1004 Hz levels at th
65. ice transmission is clear and quiet probably because there is no source of electrical interference from any other electric systems of any kind These ancient systems are good candidates for replacement by solar powered earth station sets for low earth orbit satellite systems thus skipping 120 years of intermediate technology See Echo Cancellers in Nontraditional Networks Connections to PSTN on page 19 and Delay on page 35 Line A telephone transmission link that has customer terminal equipment connected to at least one end A POTS loop is a line If both ends of line are connected to customer equipment and has no switch connections it is a private line service or basically an intercom Lines tend to be at least partly two wire from the central office out to the cus tomer premises Any interoffice links of private lines constructed today will use digital four wire facilities thus intro ducing two wire to four wire hybrids Trunk A telephone transmission facility connecting two switches The trunk is made up of the terminal electronic equip ment amplifiers multiplexers transmitters receivers etc the media wire radio optical fiber and possibly intermediate amplifiers The switches may be local long distance or private network switches Modern trunks are almost always four wire Trunks connect the local network to the long haul networks and to pri vate switched networks Lucent Technologies Inc 31 Information Manual TECO3
66. in DS1 applications may be set to 64 clear mode The basic EC convolution processing NLP and high pass filter HPF function can be controlled on a per chan nel basis When disabled on a given channel and the channel has not been placed in the 64 clear mode A law to u law or u law to A law conversion can still take place if provisioned The basic EC function can be enabled or disabled on a per channel basis via the microprocessor interface The basic EC function can be enabled or disabled per channel via an external pin ECDIS pin 26 The basic EC function can be controlled via in band tone disabling refer to Tone Disabling section on page 9 The echo canceller function may be optionally disabled with 2100 Hz tone detection with phase reversal 2100 Hz tone detection with or without phase reversal or not at all The echo cancelling function can be controlled via in band VPA tone disabling refer to Tone Disabling section on page 9 The 64 clear function can be enabled or disabled on a per channel basis via the microprocessor interface The 64 clear function can be controlled via in band tone disabling refer to Tone Disabling section on page 9 The 64 clear function may be optionally enabled with 2100 Hz tone detection with phase reversal 2100 Hz tone detection with or without phase reversal or not at all Note A provisionable minimum signal power must remain on the channel to maintain the 64 clear function The frame delay thro
67. information on speech coding see reference 6 Another aspect of digital cellular systems is that packet transmission is often used to transport voice packet call control signaling and systems control between the wireless base stations in the field and the mobile switch Long or variable frame packet schemes like frame relay can introduce considerable delay see Packet Transmission Impairments section on page 26 CELLULAR S SS e SY PSTN MOBILE TELEPHONE ECHO A SWITCHING OFFICE CANCELLER f D BASE STATION CONTROLLER feds OOF nw RF rg PACKET TRANSMISSION DELAY REFLECTION LOW BIT e RATE VOICE ie Note RF Radio frequency 5 7084 F r 4 Figure 8 Cellular Telephone 20 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Wireless Local Loops The International Telecommunications Union ITU predicts that by 2002 seventy five percent of newly installed POTS customer loops will be wireless not copper twisted pairs see Figure 9 below This will likely be in rural areas in developing countries that do not have the huge embedded base of wire that developed countries have Wireless technologies promise the possibility of reaching customers with voice and data services without installing copper wire to every home There is speculation that there ma
68. ior Electronic amplifiers are never perfectly linear but they can be very good and acceptable for a given purpose over a wide range of input and output amplitudes and frequency ranges A high fidelity stereo amplifier can be good from a few Hertz up to 30 kHz and from millivolts to volts input with careful design Telephone amplifiers do not need such a wide frequency range to transport acceptable human speech A frequency range from 300 Hz to 3000 Hz has provided adequate speech fidelity for decades There are some networks that try to provide a some what wider frequency range higher and lower for a more natural sound A narrower range of amplitudes also eases design complexity and cost For very small amplitude input signals the amplifier device turn on turn off characteristics may tend to be nonlin ear or the signal may be small enough to be lost in the internal noise generated by the amplifier itself until some higher level is reached Above some input level the amplifier may be very linear For some high levels of input or output the amplifier may become nonlinear as the signal level approaches the amplifier stage supply voltage level The range from smallest linear amplitude to the largest linear amplitude is called the amplifiers dynamic range usually expressed in dB The dynamic range is sometimes measured by the level range for which an dB input in the input level results in a dB increase in the output For example for the input level
69. k Applications Local to Long Haul continued Echo Cancellers in Nontraditional Networks Connections to PSTN Many of the new voice transport schemes and networks such as digital cellular phones PCS wireless local loops low earth orbit satellites speech compression forward error correction and packet switching create transmission delay that is not distance related Figure 7 shows only the end points delay and echo control of these networks The delays in a hybrid fiber coax HFC telephony system will still be mainly due to distance unless very low bit rate voice or packet transmission is used When networks interconnect to the traditional PSTN they can experi ence echo that results from the delay inside their networks and from the hybrids in the local POTS networks A solution is to install network electrical echo cancellers in these networks at the gateways to the PSTN The echo cancellers are pointed at the PSTN local hybrids PUBLIC SWITCHED TELEPHONE NETWORK PSTN ECHO CANCELLER DELAYS DISTANCE LBRV PACKET ERROR CONTROL 5 7083 F r 2 Figure 7 New Telephony Transport Network Lucent Technologies Inc 19 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications continued Digital Wireless Cellular Low Bit Rate Voice Figure 8 below depicts a simplified view of a digital cellular wireless
70. k electrical echo reflections Each of these serial ports X C Y and E conform to the concentration highway interface CHI time division mul tiplex TDM standard 1 The PCM encoding format is inverted or noninverted u law A law or 16 bit linear for the X C Y and E ports The choice between linear and companded encoding is provisionable per port independently The choice of companded encoding law is provisionable for all channel for each port independently Most applications use inverted law or A law coding The serial bit rate on all ports is either 4 096 Mbits s with thirty two 16 bit time slots or 2 048 Mbits s with thirty two 8 bit time slots The selection of the I O bit rate applies to all four ports Mixing of the 2 048 Mbits s and 4 096 Mbits s is not supported The linear PCM format is not supported for the 2 048 Mbits s mode The TECO3264 provides parity checking on the X and Y PCM input streams The parity is programmable for odd even or none at all Parity can be calculated over the 8 bit PCM word and placed in the ninth bit or alter nately over the 15 bit word and placed in the sixteenth bit Parity is not supported when the 16 bit linear PCM encoding format is used The parity options are provisioned independently for each port The TECO3264 provides parity generation on the E and C PCM output streams The generated parity is pro grammable for odd even or none at all Parity can be calculated over the 8 bit PCM word
71. ments In case of multiple paths all delays and dispersion of any individual echo path are included Echo Return Loss Enhancement ERLE ACANc Echo return loss enhancement ERLE ACANC is the attenuation of the echo signal as it passes through the send path of an echo canceller This definition specifically excludes any nonlinear processing on the output of the can celler to provide for further attenuation ERLE is a measure of how much the echo canceller reduces the echo coming back into the send in Y port Because of the nonlinear nature of the u law and A law encoding decoding process the maximum ERLE that can be obtained is about 34 dB which the TECO3264 can achieve This 34 dB ERLE is still not good enough and a nonlinear processor is used to finish the echo control job Far End Far end is the side of the telephone connection that does not contain the echo path on which the echo canceller is intended to operate For long distance circuits this is the long distance portion of the end to end circuit The delay in the long distance portion may be hundreds of milliseconds H Register H register is the register within the echo canceller which stores the estimated impulse response model of the echo path from the C ROUT port back to the Y SIN port Leak Time Leak time is the interval between the instant a test signal is removed from the receive in port of a fully converged echo canceller and the instant the echo path model in the echo
72. modern switch can insert loss to maintain a minimum loss If one counts the hybrids in the station sets and line units there are at least four hybrids in the circuit none of which are perfectly balanced But with the short distances and low delays echo is not per ceived and is not a concern Loss 0dB 9dB 0dB 0 dB 9 dB 1 kHz 1 kHz 1 kHz LOOP 2 LOOP 1 LOCAL DIGITAL SWITCH DELAY TENTHS OF A MILLISECOND 5 7081 F Figure 5 Local POTS Call Connection 16 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Wired Telephone Network Applications Local to Long Haul continued A Long Distance Call Connection Figure 6 below shows an end to end long distance call connection starting with a POTS customer and local switch at each end The connection is symmetric about the middle The local switches connect to trunk facilities that con nect to the toll switches On the other side of the toll switches there is an echo canceller oriented for cancelling echoes that come from the local loop hybrid This position allows the echo canceller to be shared among many trunks from the local network on a call by call basis The tail end delay is mainly the signal propagation delay from the echo canceller receive out C port through the toll switch the toll to local trunk facilities and through th
73. n loss level with a different delay from the previous calls The echo cancellers must adapt as quickly as possible in order to cancel echoes at their respective ends hopefully before the human ear hears any echo The hybrid echo characteristics do not tend to change much during a call but when changes do occur the echo canceller must recognize them and adjust as necessary to maintain quality When one person speaks and a second party listens until the first party is finished speaking then the second party speaks while the first party listens etc This is called single talking People do not talk in this manner Very often both parties speak at the same time It may be as simple as a yes or no or just a hmmm to indicate that the lis tener is indeed still listening If the conversation becomes more excited both parties may be speaking at the same time These situations are called double talking If humans only single talked the task of designing an echo control device such as an echo canceller would be much simpler The TECO3264 employs advanced detection tech niques and algorithms to reliably determine when the far end speaker is speaking or not when the near end speaker is speaking or not so that the echo canceller can be trained at the appropriate times the nonlinear pro cessor can be activated deactivated and comfort noise inserted and removed at the right times Low ERL values make it harder to distinguish between an echo and a quiet
74. narrowband a A means is provided to substitute the internal NBE indication with an external VSC indication This feature allows the use of a more accurate voice data classification of the signal by an external processor 8 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual TECO3264 Functions continued Control continued Tone Disabling The echo canceller is equipped with two independent tunable tone disabler functions TD1 and TD2 to disable the echo canceller when voiceband modem facsimile data or other specified tone signals are transmitted through the canceller The tone disabler characteristics conform with ITU T Recommendations G 164 3 G 165 4 and G 168 5 Tone detector TD1 is usually used The tone disabler is used to disable the echo canceller on the detection of specified tones but does not disable on speech The tone disabler provides tone disabling function for 32 channels in send Y E and 32 channels in receive X C directions The tone disabler responds to a disabling signal that may be present in the send or receive direction The tone disabler detection circuit is capable of detecting a 2100 Hz 21 Hz tone and disabling the echo can celler as specified in ITU T Recommendation G 164 3 For correct operation of V series modems the tone disabler detection circuit is capable of detecting a 2100 Hz 21 Hz tone with periodic phase reversal
75. nceller is a device used for reducing near end echo present on the send path by subtracting an estima tion of that echo from the near end echo Echo Return Loss ERL AECHO Echo return loss ERL AECHO is the attenuation of a signal from the receive out port ROUT C to the send in port SIN Y of an echo canceller due to transmission level adjustments and transhybrid loss i e the loss in the near end echo path In voiceband echo canceller terminology ERL is the signal level loss of a signal exiting the receive out C port and returning to the send in Y port Since digital networks are usually set for 0 dB loss end to end at 1 kHz the ERL value is basically the four wire input port to four wire output port loss of the hybrid network on a given call Usually a single number often 6 dB for a good network is given and is understood to be the minimum ERL for any fre quency within the voiceband for a given call Return loss tends to be higher in midband 500 Hz 2500 Hz and lower at the band edges below 500 Hz and above 2500 Hz because of poorer impedance matching between the customer loop and the hybrid compromise balance network For electrical reflections from a hybrid the ERL is usu ally relatively stable in level versus frequency and in delay spread during a call but may vary some One example of the echo varying during a call is when the customer is very close to the central office thus with a low loss loop If a second telephone
76. nected Soft Nonlinear Processor a The soft nonlinear processor SNLP mode allows a smoother and slower transition from NLP disabled switch closed to NLP enabled with noise matching An abrupt change in background noise is noticeable a The SNLP mode is provisionable on an all channel basis Noise Transparency The transparency mode allows the noise matching data insertion to be colored by low level near end input sam ples intended to improve spectral noise matching performance The inserted comfort noise sounds more like the real background noise that it is replacing and the change is less noticeable to the customer The transparency mode is provisionable on an all channel basis Lucent Technologies Inc 7 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 TECO3264 Functions continued Control Far End Speech Detector FES Two measures of far end energy i e signals coming in the X port are calculated The first is signal power com puted from all the samples stored in the X register The second is based on the peak signal magnitude stored in the X register This measure is adjusted by a peak factor that depends on the spectral narrowband vs wideband con tent of the signal and state of the canceller a Far end speech FES is declared FES TRUE if either of these measures exceeds a threshold value Note that there is no hangover count involved a The peak factors are provisionabl
77. ntrol signaling on hook off hook ringing etc in North American digital carrier systems steal the least signifi cant bit of the DSO signal every sixth frame of the 1 544 DS1 signal to use for call control purposes This scheme Lucent Technologies Inc 47 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued High Speed 56 kbits s Voiceband Modems continued is called robbed bit signaling and is an in band control signaling transport scheme since the voice and call control share the same DSO Thus only 7 bits of the 8 bits per channel have a chance to get through intact yielding a net maximum modem rate of 56 kbits s not 64 kbits s Further the U S Federal Communications Commission FCC signal power limits result in an actual maximum rate of only 53 kbits s The 56 kbits s voiceband modems can toler ate only one analog to u law conversion The new V 90 standard for 56 kbits s modems is nearing final approval The 56 kbits s data transport is asymmetric that is the 56 kbits s maximum rate is only available from the service provider to the customer The customer to service provider direction uses normal voiceband modem techniques such as V 34 E1 based transmission systems do not use robbed bit inband signaling but carry all of the call control for 31 voice circuits in one of the 32 DSO slots for a common control channel usually channel 0 This is called out of ban
78. ntrol were tried with vary ing degrees of partial success Getting rid of the reflection with perfect hybrids is impossible Providing very high quality hybrids for every one of the hundreds of millions of customers is impractical The past approaches usually involved adding full time loss to the long distance trunk links that increased with distance The North American echo control plan was called via net loss VNL These long distance trunks are not dedicated to any one customer but are shared among all calling customers on demand under control of the switching systems in the local and long Lucent Technologies Inc 33 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Echoes continued distance networks Echo suppressors were developed that attempted to detect when the far end party was talking but not the near end party and switch in a fixed loss When the near end party started talking the suppressor tried to switch out the loss without the humans detecting it It was better than nothing and the only practical solution at the time Being able to share the complex expensive echo suppressors by placing them back in the edges of the long dis tance portions of the networks made the per customer per call cost feasible For small small in numbers of cir cuits not in physical extension or delay aspects networks where there is not the ability to share relatively compl
79. oice frequencies Echoes from hybrids and impedance mismatches in two wire links are usually much larger Some customer premises four wire circuits do not use twisted pairs Quadded cable is made with four nontwisted conductors in one cable sheath Two of the conductors are used for transmission in one direction and the remain ing two are used for the other direction Without twisting the pair to pair crosstalk is often 10 dB or more worse than for twisted pairs So if a voice circuit has long runs of quad cables at the customer ends the circuit is four wire all the way with no hybrids and acoustic echo cancellation is used at the ends then the crosstalk echo component could become the main source of echo This could be the case for some of the new telephone networks where the two wire loop hybrid interfaces at the central offices are bypassed If this does become a problem an electrical echo canceller such as the TECO3264 would be a solution Unlike echo from a hybrid where the echo is from a point source with a very small spread in delay this crosstalk echo coupling is distributed along the wire pairs as they run near each other Thus the echo is spread out in time as the signals propagate along the wires Voice frequency signals in wire travel at about half vacuum light speed or about 150 meters us If the crosstalk takes place along a 3000 meter run of cable the crosstalk is spread out for 20 us This would still be represented in only one of
80. one trunk had to be at least 13 dB translating to a transhybrid loss of at least 21 dB Since the long distance network investment is considerably smaller than the local plant this was the logical place to make up for necessary compromises in the local systems As local four wire digital switching machines began to replace the two wire analog switches in the 1970s the four wire to two wire hybrid interface moved out the interface to each local customer s loop at the central office end or with digital loop carriers out to a location in the field closer to the customer premises These local customer inter facehybrids do not indeed can not get the intense degree of impedance matching and return loss control that the far fewer toll hybrids back in the network once received One aspect of telephone system design that was not compromised is reliability Telephone equipment is expected to be installed turned on and work night and day for decades The failure of a telephone switching system for any reason other than a natural disaster is a news event and the failure is analyzed in detail to help avoid similar fail ures The most common large failure in telephone systems occurs when a construction company digs up a tele phone cable by accident Compare this to normal computer systems where a day without a crash or data loss of some sort is a success The TECO3264 is designed for mean time between failures measured in decades Some day the voice band tel
81. only a few ms of delay not the 64 ms of the network cancellers The target for attenuation of the acoustic echo is 40 dB to 45 dB between the physical acoustic path between the speaker and the microphone and the elec trical echo control circuit Lucent Technologies Inc 43 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Metallic Pair Crosstalk and Echo Just to be complete there is another source of voice frequency electrical echo that is usually small compared to hybrid echo Outside of customer premises wiring it is also becoming obsolete However since the four wire inter face is migrating out to the customer it may be of more concern than in the past There can be electrical crosstalk between the metallic pairs carrying signals in each direction for a true four wire metallic transmission circuit The output signal to the transmit wire pair couples back into the receive pair and adds to the received signal This is a result of the inductive and capacitive coupling between pairs in the vicinity of one another The level of this crosstalk echo depends on the signal frequency the relative closeness of the pairs and the distance along the path that the pairs are close to each other The twisting of the individual pairs the placement of the pairs and careful wir ing practices can easily keep this echo to an ERL of 40 dB and more down relative to the passing signals at v
82. r control functions Note The 20 Hz ringing pulses are not sent by PCM encoding but just the presence absence of the 20 Hz bursts are detected so that they can be recreated at the far end This DS1 robbed bit signaling approach leaves a net voiceband capacity of 64 4 3 kbits s while introducing a little low level noise With the robustness of the u law encoding the human listener never hears anything wrong What happens if an echo canceller and a carrier transporting call control signals were to encounter For E1 based systems it is simple With the TECO3264 the common signaling control channel can be set to 64 clear bypass mode and the signaling passes through the echo canceller untouched If a DS1 based system must preserve the robbed bit signaling the ABCD signaling bits can be copied out of each channel before the echo canceller and Lucent Technologies Inc 41 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued POTS Call Control Signaling Transport and Echo Cancelling continued then reinserted after the echo canceller The echo canceller treats the robbed bits passing through just as very low level noise since the signaling bits in each transmission direction are not necessarily related The robbed bits are very likely to be mangled by the echo canceller process and wiped out when the nonlinear processor and comfort noise are activated Hence they must be r
83. rameters in the read write Page 0 registers 2 Normally the microprocessor does not perform real time echo canceller calculations except for one possible option The TECO3264 takes care of all real time functions internally with an option for external call boundary indication to restart echo path cancellation convergence Normally the echo canceller detects the start of a new call or a change in the echo path from the signal itself However for some networks such as wireless cel lular networks the external systems have the start and stop time information for each call If desired this external call boundary information can be fed to the TECO3264 via CALLB pin 34 3 Interfacing and monitoring maintenance and error indicators mostly in the read only Page 1 registers Details of the microprocessor TECO3264 interface are presented in the August 1999 TECO3264 32 Channel Echo Canceller Data Sheet DS99 241PDH A fully functional stand alone evaluation board system TECO3264EB is available to allow laboratory and field tests of the TECO3264 for DS1 T1 and E1 applications A brief description is found in section Echo Canceller Terms on page 27 See TECO3264EB TECO3264 Evaluation Board System User Manual MN98 072TIC for detailed information Motorola is a registered trademark of Motorola Inc Intelis a registered trademark of Intel Corporation Lucent Technologies Inc 3 Information Manual TECO3264 32 Channel Echo Canceller Inform
84. rs are small and may have only a few circuits in any one city When these circuits involve long distance connections the interconnections have to be car ried farther back inside their long haul network to pick up echo control A result is that some of the long haul delay may be included in the tail end delay seen by the echo cancellers Instances of tail end delays greater than 64 ms and as great as 150 ms in a very small number of cases have been encountered Such cases are expected to continue to be in the minority and there are means to deal with them If it is known that the delay for a given trunk group is always greater than 64 ms a fixed delay can be added externally to the TECO3264 chip The potential market is being studied for a new version of TECO that can handle longer end delays When cross sections of trunks having less than approximately six circuits need more echo control DSP based solutions using algorithms similar to those built into the TECO ASIC may be considered When there are ten or more trunks using the TECO3264 with external delay padding may be economical even when most of the channel capacity is not used The long end delay and small cross section application situation is a matter of study to develop the best technical and economic approaches 18 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Applications continued Wired Telephone Networ
85. rts facing the local switch to cancel the local network hybrid echoes The far end or X and E ports face back toward the toll network Signal delay through an old analog circuit switch was very small and was mainly due to propagation through the intra building wiring Signal delay through a packet based switch such as ATM can be considerable Central Office Strictly speaking the local telephone switching machine identified by the first three digits called NNX of the local telephone number in the U S A as in the 879 of 879 6143 Central offices are housed in buildings called wire cen ters because the local network of customer loops radiate outward from there A wire center may contain one or several central office codes with up to nearly 10000 customer lines each served by one or more switching machines The term central office is often used to refer to the whole building as well Public Switched Telephone Network PSTN PSTN is the name for the collection of telephone networks local and toll switches and trunks that make up the public switching network originally intended for voice service Many large companies governments the military and other entities have their own private telephone and data networks that carry most of the traffic between their locations For interconnections outside their networks they may have one or more gateway connections to the PSTN Hopefully these private networks use the same care in echo control as the P
86. s inserted in that tone a The TD2 tone disabler is capable of detecting tones in the frequency range of 2000 Hz to 2110 Hz tone for voice path assurance VPA VPA is an end to end path continuity test that may be performed by the network before the circuit is given to the customer for use a VPA detection is provided on the receive path H Register Freeze The H register contains the echo canceller s model of the tail end echo path Each of the 32 E1 or 24 DS1 chan nels has its own echo model calculated for each customer call The model adapts for changes during the call Control of the H register freeze function can come from any of the following sources 1 External global control an external pin is provided to globally freeze the H register for all channels A mode is provided to allow per channel serial control of the freeze function 2 NES FES control the NES and FES speech detectors are combined to determine when to freeze The freeze function is enabled whenever NES is present or FES is not present 3 NBE control NBE detector is used to freeze the H register The freeze function is enabled whenever NBE is present 4 There is a per channel freeze control via microprocessor interface H Register Reset The H register reset setting of all H register taps to zero function is performed on a per channel basis The sources that can affect a reset are as follows 1 External control a means is provided through the microproc
87. s tweaked the u law coding levels just a little to develop their A law coding scheme to provide a more constant signal to quantization distortion ratio for various levels of signals The u law and A law yield a net result with only 8 bit samples that is equivalent to 11 or 12 linearily coded bits in terms of dynamic level range This was well worth the effort to save the bits all through the switching and transmission systems Most of the telephone net works outside the U S A use A law speech companding Digital echo canceller systems like the TECO3264 depend on linear PCM samples to do all of the calculations for the digital signal processing Inside the echo canceller the u law and A law samples must be converted to linear PCM before calculations can be done In order to preserve arithmetic accuracy 16 linear bits are used not just the 11 or 12 that would be enough for human hearing There are some music lovers with highly sensitive hearing who claim that the 16 bit linear coding used for audio compact disc CD recordings is not good enough Indeed some say that only analog with infinite bit resolution is good enough Some recent music CDs mention 20 bit processing For the receive side of the canceller the far end X input encoded bits are passed to the near end C output port The copy of the X signal that is passed to the convolution update processor CUP is converted to linear PCM For the send side of the canceller the near end Y input signal is
88. sed farther out for longer loops Thus the typical loop has a mixture of gauges Telephone cable is generally available in 500 foot reels To reach the average customer on a 12000 foot loop the pair must be spliced two dozen times For flexibility in assigning pairs to customers as they apply for phone service and eventually move on connected unused parallel shunts called bridged taps are used to connect a cable to more than one customer area Added inductors called loading coils are installed on loops longer than 18000 feet to reduce transmission loss Longer loops in the U S A are gradually being replaced with digital loop carrier DLC transmission multiplexer systems that bring the four wire hybrid interface out from the local switch to within 12000 feet about 4 km or less of the customer European loops tend be shorter and to use coarser gauge wire with no bridged taps or loading coils Sometimes old technology is around for a long time As recently as the mid 1990s and possibly still in 1998 remote villages in the South American mountains are linked to civilization by the one iron wire with earth ground return telephone transmission loop used in 1878 These circuits extend from the closest town with some modern telephone service up into the mountains for many miles and may link more than one village in a party line Such systems were still in use in the U S A in the 1930s Reports are that the noise levels on these lines are very low and vo
89. ses of 144 kbits s This 144 kbits s is usually divided into two 64 kbits s bidirectional channels labeled B bearer and a 16 kbits s channel labeled D data The B channels can be used independently for voice or digital data and the D channel is used for call control replacing the POTS on hook off hook dialing 20 Hz ringing etc and low speed customer data pay loads The two B channels can be linked to provide a 128 kbits s channel This combination of customer channels is often referred to as 2B D The most common transport for ISDN is by a digital subscriber line DSL system that operates over a single ordi nary two wire nonloaded loop between the network and the customer premises DSLs have their own type of echo cancellers to enable four wire digital to two wire digital path conversions at the network and customer ends of the loop for the DSL signals The DSL extends four wire transmission all the way to the customer The customer termi nal equipment can be many types of digital voice data or facsimile devices An ISDN voice telephone set could be digital up to nearly the microphone and speaker Side tone coupling from microphone to speaker would be added to make the phone sound live just as in analog sets Any echo that could get back to an electrical network echo canceller would be acoustic echo from handsfree terminal sets Because of the huge embedded base of analog voice telephone terminal equipment the ISDN DSL can also attach
90. several trunk links some analog and some digital A speech signal would have to undergo several analog to digital encodings and then digital to analog decodings to make it The digital encoding decoding scheme would have to be very robust and still provide toll quality speech The answer is the u law coding scheme that compressed large signals by using large sample steps and provided fine grain level resolution sample steps for very small quiet speech When the encoded signal was decoded back to analog speech the linear multiplied input yields multiplied output rule is violated but it sounds good For any pulse code modulation technique for analog to digital conversion there is some distortion since each coded digital word corresponds to a range of analog input level For an original analog sample level near the bot tom or top of a given level range for a given digital code it gets decoded back to the center of the range at the range in the digital to analog conversion The center assumes the reference levels are set exactly the same at each end of course but at any rate a range in becomes a point level value out This introduces some nonlinear distortion The more bits in the digital code words the smaller the end to end A to D to A distortion For u law and A law the out of voiceband distortion products are filtered out by the sampling filters The small inband distortion components are not noticed in the wideband speech signals The European
91. speaker at the near end The whole double talking accommodation process by the echo canceller should be transparent to the listening talking customers Telephone call connections in the modern world are usually made up from several pieces supplied by several tele phone network providers starting with the callings customer s local operating company by a long distance network provider and finally by the called customer s local operating company The long distance portion of the call may be made up from one or more national links international links and links by provided by other circuit providing entities Wireless mobile calls are linked into the landline networks and then to a landline customer or back out to another wireless customer In over the air television transmission echoes from hills buildings airplanes etc are seen as visual ghosts on the television screen Ghosts can occur in poorly engineered cable TV systems as well from reflections off poor cable connections or improperly terminated cables 34 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Delay Formerly the transmission delay between the time a signal left the transmitter and arrived at a receiver was just a result of the fact that physical things can only travel at finite speeds Radio signals and light in air travel at nearly the speed of light in a vacuum 30000
92. system is a big box whose only visible parts are the phones on their desks and sometimes some cables hung on poles throughout the neighborhood One picks up the phone handset dials a number magic happens and one can speak with someone across town or halfway around the world Your voice goes in here and comes out over there Even some technically minded customers have little understanding of the many parts of the network that are involved This section will present some basic sketches and explanations as background to where voice frequency echo cancellers such as the TECO3264 fit in the over all picture Other subsections and paragraphs deal with individual topics in a little more depth and detail A Local POTS Customer Connection Figure 3 below depicts the local connection for a typical plain old telephone service POTS customer On the cus tomer premises the most visible piece is the telephone station set In addition to station sets there may be data modems facsimile FAX machines answering machines and caller identification boxes connected in parallel The inside wiring with two wires starts at the interface to the network two wire loop at the entrance to the building in a basement garage or other convenient point The inside wiring runs about the premises and appears for a connec tion wherever the customer desires The network cable called the loop connects from the customer premises to the local telephone building central office CO
93. t Gain Normalization Gain normalization is based on a measure of Xin power It is provided to make convergence time independent of X input signal level The prenormalized gain is provisionable on an all channel basis and provided by the convo lution gain control circuit The set of peak factors available for use in the gain normalization process are provisionable on an all channel basis Various indications are used to select the proper peak factor under a given set of conditions refer to the Control section on page 8 High Pass Filter a All channel selectable high pass filter gt 10 Hz is provided for X input signal a All channel selectable high pass filter gt 10 Hz is provided for Y input signal a The high pass filters are disabled on a per channel basis whenever a channel is either in 64 clear or EC_DISABLE mode These high pass filters block any direct current present on the echo canceller ports from reaching the echo cancel ler circuitry Nonlinear Processing Even with the 34 dB echo return loss enhancement residual echo e called ehat can be occasionally heard on low noise trunks In order to reduce this effect a nonlinear processor NLP inserts either loss finite or infinite or a controlled level of noise in the echo return path when there is no double talking i e only far end speech is present The 34 dB echo return loss enhancement approaches the maximum possible with law and A law com panded signals
94. t 1999 Table of Contents Contents Page IMTOCUCTION EE 1 Architecture and Functional Description of TECO3264 0 0 ceeecceceeeeeeeeeeeeeeeeecaeeeeeaeeeeaeeeeeaaeeseaeeeesaaeeeecaeeeeeaeeeeaees 4 Uescppion ias ENa Ea AEEA AEEA a E EE AAE EENE gen ted AA e a NES 4 TECO3264 System Interfaces and Fommais 5 PCM interface sinora t iaia ES Ae eed eee a ae ee OS 5 TEC OS3264 FUNCUHONS a iiig taronatia Desa the heey Seege ee div tye vente ke negate dete ben tape aaaea tae laand aieia 6 Linear Processing uaiiie detec eed anne seeded e NEEN ENEE aed ed ea gaa ee de 6 SOMO sc fessor EEE AAA TEO E eg tad shes eege ts at da epech Eege 8 Echo Canceller Evaluation Board System TECOOOGAER nnt 12 Echo Canceller Applications ccccccccecesceeeeeceeneeeeeneeeceaeeeeeaeeeceaeeeeeaaeseaeeeseaeseeaaeeseaaesseaaeeseaeeeeeeaeesseeeeseieeeeeaeeess 14 Wired Telephone Network Applications Local to Long Hal 14 Digital Wireless Cellular Low Bit Rate Voice 20 Wireless Local LOOPS aide hee ete ete aes EEA me ee EE EEN 21 Hybrid Fiber Coax HFC Telephon 23 Integrated Services Digital Networks ISDN Basic Access The New Digital POTg 24 Internet Telephony Applications 0 0 2 2 cccceeceeseeeeeeeeeeceeeeeeeeeeeeaaeeeeeeeaaaeeeeeeesaeeaeseesaaeseseeeaaeeseeneeeaeseeeseeaeeeenenaes 25 Public Network to Public Network via Internet LONG Hal 25 Packet Transmission Impairments eccceeeeeeeeee eee eeee cette eeeeeaee eee eeaae eee eeaaeeeeeceaaa
95. t interfaces and the impedances that it presents to the outside Point C is the receive out C port of an echo canceller system The signal travels through the intermediate network transmission and switching equipment which is not shown to the four wire input port of the hybrid Likewise point Y is the send in Y port of an echo canceller that receives the signal from the four wire output of the hybrid via the intermediate unshown equipment One two wire output port of the hybrid connects to the metallic loop that in turn goes out from the central office and connects to the customer s inside wiring and telephone sets The second two wire port of the hybrid connects to an impedance balance network BN This balance network is usually not shown in telephone connections The task of the balance network with its input impedance ZBN is to match as closely as necessary the input impedance of the loop as seen at the hybrid This loop impedance ZLOoP is a complex function of frequency It varies from one customer loop to the next It can vary some with the outside temperature It can vary with the number of tele phones being used on a given call that change the input impedance to the customer premises ZCUST especially for short low loss loops As discussed above ZBN does not have to be a perfect match to ZLOOP for local tele phone connections For older analog two wire switching systems in North America a single compromise balance impedance 900
96. t right for the network The two wire metallic loop connects to the local switch or to a digital carrier sys tem The switch POTS line unit performs the hybrid two wire to four wire conversion discussed in more detail later analog to digital and digital to analog transmission conversion and also supplies the interface for the CO battery to power the telephone sets down the loop and apply a ringing generator to ring the customer phone for an incoming call The balance networks called compromise balance networks CBNs for the line unit hybrid are designed to match loop input impedance well enough for low loss switch operation POTS LINE CARD TELEPHONE MICROPHONE gt Cis SPEAKER 5 7080 F r 2 Figure 4 POTS Line Card and Telephone Set Lucent Technologies Inc 15 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Applications continued Wired Telephone Network Applications Local to Long Haul continued Local POTS Call Figure 5 below depicts a local POTS call with two customers connected by the switch North American loop design rules limit the maximum loss at 1 kHz to about 9 dB with the average loop having about 4 dB loss at 1 kHz Loss across the switch is normally close to 0 dB So the customer to customer loss at 1 KHz can be as much as 18 dB and averages about 8 dB Too little loss results in too high levels of speech so a
97. the 125 us samples of the 64 kbits s DSO channels that the TECO3264 echo canceller sees The signals experience loss as they propagate and thus the echo level also var ies Hybrid echo also varies with frequency across the voiceband because of the varying level of match balance of the compromise balance network with the connecting loop impedance For simplified test purposes a hybrid and the intermediate circuitry is often modeled with a flat echo return loss and a single delay total tail end delay time such as 6 dB ERL and 22 ms Metallic wire end to end loss increases roughly as the square root of the frequency and crosstalk coupling tends to increase roughly as the 3 2 power of the frequency Crosstalk is an increasing concern as the signal frequency increases At voice frequency crosstalk among twisted pairs is there but the level is usually well below the echo from hybrids and impedance mismatch reflections Thus this crosstalk echo component should rarely be of con cern for past network electrical voiceband echo canceller applications Nontwisted pairs sometimes used in cus tomer premises wiring can result in audible crosstalk between telephone calls in premises with more than one line Pair to pair crosstalk is a major system performance limiting concern for the echo cancellers used in digital subscriber lines DSL where the signal frequencies reach into the MHz range Four wire analog circuits in telephone networks are becoming scarc
98. the near end speaker makes an utterance even very brief the nonlinear processor must quickly restore the send signal path When near end and far end speech are both present double talk the ERLE should be sufficient for the near end speech to hide the residual far end echo Pure Delay tr Pure delay tr is the delay from the ROUT to the SIN port due to the delays inherent in the near end path transmis sion facilities In this case the delay associated with the hybrid is assumed to be zero Residual Echo Level LRES Residual echo level LRES is the level of the echo signal that remains at the send out port of an operating echo canceller after imperfect cancellation of the circuit echo It is related to the receive in signal LRIN by LRES LRIN AECHO ACANG Any nonlinear processing is not included In the TECO3264 the residual echo is labeled Returned Echo Level LRET Returned echo level LRET is the level of the signal at the send out port of an operating echo canceller that will be returned to the talker The attenuation of a nonlinear processor is included if one is normally present LRET is related to LRIN by LRET LRIN AECHO ACANC ANLP If nonlinear processing is not present note that LRES LRET 30 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Echo Canceller Background continued Plain Old Telephone Service POTS
99. the reference model for echo cancellers that is consistent with ITU T Recommendation G 168 In the definitions L refers to the relative power level of a signal expressed in dBm0 and A refers to the attenuation or loss of a signal path expressed in dB The following terms are consistent with both ITU T Recommendations G 165 and G 168 Lucent Technologies Inc 27 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued Echo Canceller Terms continued Convergence Convergence is the process of developing a model of the echo path that will be used in the echo estimator to pro duce the estimate of the circuit echo by iteratively estimating the tail end impulse response model in the H register calculating the estimated echo measuring the residual echo error LRES updating the impulse model checking etc Convergence Time Convergence time for a defined echo path is the interval between the instant a defined test signal is applied to the receive in port of an echo canceller with the estimated echo path impulse response initially set zero and the instant the residual returned echo level at the send out port reaches a defined level An echo canceller should converge in a fraction of a second with 0 5 s being relatively long Dispersion Time Dispersion time is the time required to accommodate the band limiting and hybrid transit effects Echo Canceller An echo ca
100. the state of the NLP whenever the H register is frozen This option is provisioned on an all channel basis via the microprocessor interface Convolution Gain Control Fast and Slow Convergence Modes The TECO3264 system is equipped with a convolution gain control circuit whose purpose is to accelerate the con vergence rate under certain conditions on any given channel This is called the fast convergence mode In conjunc tion with this feature the system provides several means for detecting when the echo canceller EC is not converged The specific measures for EC convergence are done on a per channel basis and are listed below 1 H register reset when the EC transitions out of the H register reset state for any reason the EC may be considered not converged 2 External control the EC provides a means for an external to the TECO3264 system source to indicate that the EC is not converged This indication is communicated via the microprocessor interface In addition an external pin is provided for this function 3 vs Y calculation a parameterized calculation determines if the EC is converged by examining the power measure at vs the power measure at Y If the power measure at is sufficiently larger when compared to the power measure at Y the EC is considered not to be converged 4 VPA detected to not detected transition since VPA provides an indication of call setup a VPA detected to not detected transition on the VPA dete
101. to a terminal adaptor TA that connects to existing customer wiring telephones and analog terminals This terminal adaptor contains the analog hybrid that would be at the central office for analog POTS Electrical echoes can get back to a network echo canceller from the impedance mismatches between the terminal adaptor hybrid with its compromise balance network and the mixes of devices connected to the inside wiring This is similar to wireless local loop and hybrid fiber coax connections at the customer premises There are recommendations for the on hook and off hook impedances of voiceband terminal devices nominally 600 Q off hook at 1 KHz When the number of devices off hook can vary even during a connection the customer premises input impedance can vary considerably A 6 dB impedance return loss is easy to get when connecting between the network termination such as an ISDN terminal adaptor and the inside wiring For instance three telephones off hook yield a 200 Q shunt loop impedance for a 6 dB ERL This customer premises echo source is usually buffered by the two way loss of the loop Loops have an average of 4 dB loss at 1 kHz each way or 8 dB for a reflected signal This 8 dB plus 6 dB mismatch yields an ERL of 14 dB and is usually of less concern than the office hybrid echo The terminal adaptor may insert about 4 dB of analog or digital loss in both transmission direc tions to simulate the average loop loss to keep the end to end loss abo
102. ugh the device in each direction is the same for all modes of operation Lucent Technologies Inc 11 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Evaluation Board System TECO3264EB To allow laboratory and field tests of the capabilities and performance of the TECO3264 integrated circuit Lucent has developed a fully functional stand alone echo canceller evaluation system for T1 DS1 and E1 applications See Figure 2 TECO3264EB Hardware Block Diagram on page 13 A Lucent Technologies Microelectronics Group T7630 Dual T1 E1 Terminator performs the E1 or T1 line interface and framing functions for the TECO3264 PCM data and clock information exchange between the TECO3264 and T7630 is by a programmable 2 048 MHz or 4 096 MHz serial time division multiplex TDM bus called the concen tration highway interface CHI 4 The CHI input output is connected to four Lucent T7270 time slot interchange TSI circuits configured as a four channel time slot channel network switch The external control interface to the T7630 is by an RS232 link to an onboard Intel 87C51FC microprocessor The external control interface to the TECO3264 is by an RS232 link to an onboard Motorola MC68360 microprocessor The external equipment required to operate the TECO3264EB system consists of sources of 3 3 Vdc at 1 0 A and 5 0 Vdc at 3 0 A and an BM compatible computer to run the ANSI terminal interface user interface
103. urces Traditional Distance Speech Compression Packet Switching Error Correction Etc 42 Echoes in Telephone Calls Electrical and Acoustic c ccceccceceeeeeseee cece eeeaaeeeeeeeeeeeeeesaaeeseseeeceaeeeseas caaaeeenees 43 Metallic Pair Crosstalk and Echo 44 Linear Nonlinear Nonlinear Processors U laW A lAW ccccceecceeeeeeeeeeeeeeeeeeeeeeeeeeeceaaaaaaecaeeeeeeeeseseeesceeaeeeeeess 45 Adaptive Delta Modulation Pulse Code Modulation ADPCH 47 Digital Speech Quality Categorie Sisenen vedi anaa riaan aa iaae a aaaea aani Aaaa aeiaai dah 47 Quantization Distortion Units qdu cccecese cece ee eeeeeeeeeeeeeeaeeeeeeeeseaaeeeteaeeseeaeeeeaaaeseeeeeesaaaesseaeeseeeeessaeeseaeeee 47 High Speed 56 kbits s Voiceband Modems 47 Selaa ere tae Te EEN 48 FRREISKE er 49 2 Lucent Technologies Inc Information Manual August 1999 TECO3264 32 Channel Echo Canceller Information Manual Introduction continued 2 If it is known that the signal levels in the given network are consistently higher than normal there are 6 dB and 12 dB loss pads that may be switched into the receive or send signal paths as desired 4 Tone disabling options for voiceband data modems and facsimile can be changed if needed The TECO3264 microprocessor interface can be used with either Motorola or Intel microprocessors The micro processor performs the following external functions for the echo canceller 1 Static provisioning of the pa
104. used Lucent Technologies Inc 35 Information Manual TECO3264 32 Channel Echo Canceller Information Manual August 1999 Echo Canceller Background continued The Telephone System Not Perfect for Very Good Reasons Today s telephone system goes all the way around the world and from the northernmost to the southernmost inhab ited points on the earth It involves an incredible investment in money time material and human ingenuity The system is not perfect but it is ideal in the sense that it provides acceptable voice service for a price the customer is willing to pay Devices like echo cancellers exist because the system is not perfect The design and implementation of the telephone system like any engineered system or tool involves choices and balances and trade offs among many conflicting concerns like cost performance state of the art timely deploy ment reliability among others The design choices have a history starting with the pioneers like Bell Grey and Edi son over 100 years ago In the beginning even a barely perceivable hello heard across town was a miracle The word hello was coined for telephone use Bell wanted to use ahoy As the technology improved that quickly became not good enough To start with the transmission facilities were one iron wire with an earth ground return still used and still works well in some South American mountains in the 1990s Then Bell came up with insulated twisted copper wire pairs Ed
105. ut The multiplied input yields multiplied output rule is no longer valid When the echo cancellation has done its best over 30 dB of echo return loss enhancement for the TECO3264 human hearing can still detect speechlike signals coming back Something else must be done for a final clean up The nonlinear processor comes into play This function is a holdover from older echo suppressor system design and can actually cut off the return path completely so nothing gets through when near end speech is not present When the nonlinear processor opens the switch back toward the far end on the E port it can insert digital code word corresponding to silence Another common term used for the nonlinear processor is center clipper since it clips out the center or the low level residual signal after the echo canceller has done its best Telephone customers expect to hear something even when no one is talking So the nonlinear processor inserts comfort noise in the form of wideband white noise at about the level of the background noise as measured when no one is talking It can be fancier and use generated noise that has some time and frequency correlation to the original background so that the transition from actual noise to artificial noise is less noticeable When the comfort noise is inserted the far end customer can talk loudly or softly with a low deep bass voice or a high tenor and the output of the nonlinear processor does not change That is nonlinear behav
106. ut the same as for analog service delivery The 6 dB mismatch is then buffered when reflected back to the network as before The impedance mismatch can be worse less than 6 dB at frequencies in the voiceband especially at the low and high ends If the voiceband attenuation in the adaptor is less or more than 4 dB then the ERL seen at by a network echo canceller will change accordingly ISDN should normally use four wire full rate 64 kbits s u law or A law encoded voice all the way to the customer premises The speech quality delivered to the customer is toll quality A DSL is allocated 0 5 ms for one way trans mission or 1 ms two way so there is little additional delay over analog loop transmission This very rarely puts an ISDN circuit beyond the 64 ms tail end capacity of the TECO3264 In contrast some other telephony arrange ments that move the four wire circuit path all the way to the customer premises use very low bit rate voice LBRV in the final link to the customer and thereby introduce the delay and nonlinearity that comes with the aggressive voice companding techniques Examples of these are wireless local loop and wireless cellular systems especially some of the new digital systems ISDN DSL technology has been ready since 1987 Some large business locations have been using ISDN for sev eral years particularly in Europe and Japan and in some large companies in the USA Large scale residential and small business deployment just di
107. y not be enough copper available to deploy in China Links from the remote rural areas may well be wireless via earthbound or satellite facilities as well These wireless local loops will bring the four wire two wire conversion all the way out to the hybrids in the cus tomer station sets As shown in the figure the terminal equipment box for the wireless loop will contain a hybrid to convert from the four wire radio link to interface with the two wire circuits in the customer premises The choice of the hybrid compromise balance network see the hybrid discussions below will much more of a compromise than the balance networks used at the central office or in digital loop carriers From the central office the hybrids see the loop input impedance which except for very short loops is determined mainly by the cable characteristics with the customer premises input impedance buffered by the loop loss While individual loop input impedances at the cen tral office vary considerably from one another they are similar enough for the simple compromise balance net works to control stability and singing in the switch For the wireless terminal hybrids the customer input impedance will be determined mainly by the various customer devices that may by connected in parallel and off hook at the same time telephone sets answering machines caller ID fascimile and modems etc with the short cable runs having little effect The customer input impedance can vary greatly
108. y use speech compression techniques such as linear predic tive vocoders at 13 kbits s 8 kbits s or 5 6 kbits s to transport voice Vocoders at 4 kbits s are in the lab In spite of linear in the name these devices use highly nonlinear encoding and decoding approaches to achieve ultra low rates The output of a vocoder is not a compressed version of not quite but very nearly linear law or A law PCM but is a set of instructions on how to build a signal at the decoder that sounds like the original speech to the human ear and brain using white noise and a library of sound samples that can be scaled in frequency and amplitude One problem with the ultra low rate vocoders is developing algorithms that will work for most human languages Vari able bit rate voice transmission is another scheme that may be used where the bit rate varies with time as the speaker is talking or silent Network electrical echo cancellers like the TECO3264 depend on the echo being no more nonlinear relative to the original signal than u law or A law or about 34 dB As more nonlinearity is added to the circuit the maximum avail able ERLE will degrade accordingly With the more variable electrical echo due to the more variable customer premises hybrid reflections that pass through nonlinear vocoders the challenge to control echo in the network will be greater Acoustic echoes with handsfree speakerphone home phones on wireless loops or in cell phones are handled by acoustic
109. ystems nearly ready for service These satellite orbits tend to be about 1400 km 870 mile altitude which results in a round trip radio propagation delay of about 9 ms just from the air space path The slant range from the earth based transceivers to the satellites will usu ally be longer than the satellite altitude since they will not be directly underneath So the delays will usually be longer than the minimum 9 ms Some systems plan intersatellite relays of signals by radio or laser before returning to earth which will result in still longer propagation delays These satellite systems tend to use low bit rate voice speech compression to preserve analog and digital bandwidth Use of digital forward error correction will add more still more delay Echo control is a consideration in the LEO satellite voice systems Delay through the terminal equipment and switches connected to the copper fiber or air space paths used to be negligible being typically a few ms For the high bit rate digital subscriber line HDSL there is an allocation for 0 5 ms of one way delay due to the line encoding and decoding This was considered a large delay for terminal equipment in 1991 This is not the case any more with tens of ms being introduced by digital speech compression packet and ATM transport and digital forward error correction applications The asymmetric digital subscriber line ADSL has an allocation of 20 ms for one way delay when the error correction option is

TECO3264 32-Channel Echo Canceller Information Manual

Contents

Download Pdf Manuals

Related Search

Related Contents