communications infrastructure test group …wireless.feld.cvut.cz/mesaqin2002/echo_detection.pdf ·...

E M P I R I Xwww.empirix.com

Hammer VoIP Test System Echo Detection and Analysis

1111Copyright 2001 Empirix, Inc.

Communications Infrastructure Test GroupEngineering

TechnicalWhite Paper

Hammer VoIPTest System Echo

Detection andAnalysis

ABSTRACT:

This Technical White Paper provides an overview of echo impairment of digital communicationsnetworks and how to use the Hammer VoIP Test System to detect and analyze echo. Theinformation from the VoIP Test System can be used with Echo Canceller devices by wireless anddigital network providers to take corrective action ensuring the highest voice quality across theirnetwork. .

OverviewDigital telephony subscribers are becoming more andmore critical of their network. They expect near-wireline voice quality. A common impairment ofdigital telephone connections is Echo. Echo is when adelayed and distorted copy of a voice signal is reflectedback to its sender interrupting communication.

Echo is a quality problem.

An important technology in network voice quality isEcho Cancellation. The Hammer VoIP Test Systemcan be used to detect and analyze echo in audiosamples captured from telephone connections madewith any one, or combination of telephony protocolsoff a network. This information can be used bynetwork operators to deploy and configure their EchoCancellation resources to provide the highest callquality for their customers or be vendors of networkelements (Gateways) implementing echo cancellationto test their devices.

This White paper discusses the following:

• Echo and Its Causes

• Echo Cancellation

• Hammer VoIP Test System EchoDetection and Analysis

• VoIP Test System Echo Test Example

The example in this document was created on aHammer LoadBlaster 500™ equipped with theHammer VoIP Test System 2.7 in conjunction with.Hammer TestBuilder 2.7. A passing familiarity withthe Hammer, Hammer TestBuilder, a previousversion of the VoIP Test System, the PSTN, digitalnetworks, are assumed with the example.

Figure 1. shows a block diagram of telephone networkarchitectures and echo discussed in this paper.




Figure 1. Network Architectures and Echo

SEQARABICEchoEchoEchoEchoIn digital networks an Echo occurs when hardware intended to amplify or relay the speech of one party picks up thespeech from the other party and reflects it back to them.

Echo in Telephone CallsEcho is perceptual. There is a component of echo in every telephone call. An echo is a signal with a delay andstrength. The length of time it takes for a voice signal to be reflected back determines how much an echo impairs thequality of a telephone call. If an echoed signal’s strength does not rise above the voice call’s background noise, it is notnoticed.

A beneficial echo is called Sidetone. It is an effect and applied technique allowing the speaker to hear their own voice inthe handset’s speaker, convincing themselves their voice is being heard. The benefits of Sidetone are felt when aspeaker hears their own voice at a low volume within between 5 and 25 milliseconds (ms). Call quality begins to be isimpaired when a speaker hears their speech repeated after approximately 30 ms. Too little sidetone at all makes a callunerring, as the speaker cannot hear themselves speak.

Note the brief intervals of time involved in determining sidetone. In digital networks (especially IP) with their inherentlatencies it is very easy to have delays in excess of 30 ms.

Types of EchoThere are two primary sources of echo in telephone communications networks:

• Electrical Echo

LAN\MAN\WAN\Internet PSTN

EchoCanceller

EchoCanceller

AccessGateway

AccessGateway

IP TelephoneHybrid

Hybrid

Electrical Echo

Electrical Echo

Acoustic Echo

Acoustic Echo

Acoustic Echo Intra-NetworkCall

Inter-NetworkCall

LAN\MAN\WAN\Internet PSTN

EchoCanceller

EchoCanceller

AccessGateway

AccessGateway

IP TelephoneHybrid

Hybrid

Electrical Echo

Electrical Echo

Acoustic Echo

Acoustic Echo

Acoustic Echo Intra-NetworkCall

Inter-NetworkCall




• Acoustic Echo

Electrical EchoElectrical echo, sometimes called “Line-Echo” or“Hybrid Echo” occurs in Public Switched TelephoneNetwork (PSTN) connections. It results from animperfect electrical circuit in the network

In a digital Central Office (CO) Hybrids are devices thatconvert between the 4-Wire Switching fabric and the 2-Wire cable of the local loop. The local loop is thetwisted pair cable that connects a subscriber’s telephoneto the CO. The Switching fabric is where the localloop’s lines are connected to long-distance trunk lines.

Note that local calls do not have a problem withelectrical echo. This type of connection does not have ahybrid in the circuit.

A Hybrid is a sensitive device that requires tuning toensure call quality. If it’s not properly tuned to matchthe Two-wire cable, voice signal energy passing fromthe 4-wire to the 2-wire part of the network is reflectedback on itself creating an Echo.

The Hybrid on the opposite side of the networkproduces the echo heard by the speaker. That is, thenear-end Hybrid creates an echo of the far-end caller’svoice signal reflecting it back to the far-end. The near-end caller’s voice signal is reflected back by the far-endhybrid.

In all digital networks, electrical echo is not a problem.However, calls originated from the PSTN mayintroduce this type of echo into digital networks.

Mis-tuned hybrids are a frequent problem found inCOs’. Given information on the performance of aconnection, the Hybrid can be tuned to meet a carrier’scall quality performance needs.

Acoustic EchoAcoustic echo occurs in both PSTN and digitalnetworks. It is a more complex signal than electricalecho. It is also much more difficult to eliminate thanelectrical echo.

Acoustic echo results from the following sources:

• Handset use and design

• Voice encoding and decoding devices (codecs)

• Network delay in digital networks

H A N D S ET EC HO

Handset echo is the result of poor handset design orcomponent quality, or operating the handset in anacoustically unfriendly environment.

A handset’s physical and electrical design contributes alot to reducing echo. A properly designed handsetminimizes the acoustic coupling between the speakerand microphone of the handset. This acoustic couplingis an open-air path that can exist between themicrophone and the speaker. Frequently eliminatingthis pathway requires a compromise between echosuppression and a handset’s aesthetics.

An extreme case of handset echo is called Howling. Thisis when a feedback situation develops between thebetween microphone and speaker. This frequentlyoccurs in the use of “hands-free” telephones.

Some standards for handset design to exist. Although,these standards are primarily for digital wireless handsetsand not wirelines. These standards includespecifications for sidetone tolerance and echosuppression performance.

Another type of handset echo is called multipath echo.This echo can occur in acoustically unfriendly places likeautomobiles

echo occurs when speech indirectly enters the handset’smicrophone late after having been reflected off of walls,windows, floors, ceiling, and furniture. This reflectedsound is called multipath audio or reverberation.Mutipath echo is heard by both parties of the call as anecho.

C O D EC EC HO

Echo occurs on digital networks is the result of anunsuppressed acoustic or electrical echo made worse bydigital encoding.

Digital handsets and digital network elements include acodec for the compressing, encoding and decoding ofspeech to reduce the network bandwidth used in a call.This encoding is typically imperfect reducing voicequality. There is also requisite decode of the speech atthe other end of a digital call. This process of encodingand compressing, then uncompressing and decodingtypically introduces milliseconds of delay to transmittedspeech which can abet an echo.




N E TWO R K EC HO

Additional speech delays occur when a digital call isprocessed through multiple network elements indifferent technology networks. An encoded echowould be delayed at the intersection of two networksresulting in a signal reflection

An example of this might be a long-distance telephonecall from an analog landline on one carrier’s network tothe other party’s digital wireless telephone on anothercarrier’s network. A call like this would initially beanalog on twisted pair copper wire to the CO. At theCO it could be digitized and placed on the carrier’s fiberoptic backbone. The call may leave the original carrieras ISDN going to the wireless carrier’s mobile switchingcenter. From the mobile switching center to the CellSite it might be Ethernet. Finally, from the Cell Site tothe receiving party’s wireless telephone it wouldessentially digital radio. A multipath echo introduced bythe wireless telephone might arrive on the analog lineside 200 ms after the original speech causing asignificant impairment of call quality.

Further, in IP networks, a packetszed voice signal couldhave packets taking different routes to their destination.The delayed arrival of groups of packets in a signalhaving taken separate route results in momentary echo.This echo can quickly disappear as the traffic conditionsand routing on the network change.

Measuring EchoEcho is a reflected audio signal. It is measured in termsof signal strength in decibels (dB) and delay in ms. Themore significant component is the delay.

If the echoed signal is reflected in 25 ms or less it iseither humanly imperceptible or considered sidetone. Ifthe speaker hears their reflected signal afterapproximately 40 ms. or more, the call quality isimpaired by echo.

The internationally recognized metric for measuring thelevel of signal reflected back is called the Echo ReturnLoss (ERL). The units for ERL are dB. The smallerthe echo in a signal, the higher its ERL. The lower theERL the larger the echo in a signal. The VoIP TestSystem measures ERL directly in voice signals.

Eliminating EchoEchoes are eliminated two different ways fromtelecommunications channels, through echo suppressorsand echo cancellers. Carriers position these devices within

the network to ensure call quality. However, owing totheir cost they are deployed sparingly.

Echo SuppressorsAn echo suppressor detects human speech comingfrom one end of a connection, and suppresses all signalsgoing the other way.

An echo suppressor is toggled by a voice recognitioncircuit. They can typically “trip” within 5 ms. to block areflected signal. Unfortunately, this technique results ina half-duplex channel. This half duplex operation is notnoticeable in voice communication, but can adverselyeffect data communications.

To handle data communications, telephone circuits withecho suppressors have an in-band signaling method todisable themselves. When the suppressor detects a puretone at a specified frequency, they shutdown while thewhile the carrier is present.

Echo suppressors are found on the PSTN installed byInter-Exchange Carriers (IXCs). IXCs typically havethem installed at each end of their network to filter callsentering their network. See Figure 1.

Echo CancellersAn echo canceller is a computer-based device thatsamples a call and “profiles” it. An echo will violate theprofile. When an echo is detected it simulates the echo,estimates its magnitude, and then subtracts it from theaudio signal it is sampling.

An echo canceller initially samples the signal on thechannel to characterize the voice and the echo signalspassing through it. The time the canceller takes toprofile or model the signal is called the convergence time.This model is continuously updated during the life ofthe connection. As the signal passes through thecanceller, it is compared to the signal model it hascreated. Any part of the signal that differs from themodel is echo and is digitally subtracted from the signalbefore the signal is relayed on to its destination. Anyecho that may pass through this initial subtraction isdeleted by decreasing the signal’s power, causing theecho to disappear into the background line noise.

Unfortunately, this technique cannot perfectly cancelacoustic echo. The information in the voice signalmodel has a finite lifespan called the tail circuit delay. Anecho with a long delay (approximately 130 ms.) will havebeen aged out of the model and be interpreted by theecho canceller as part of the voice signal.




A common problem with echo cancellers is Doubletalk.Echo cancellers perform poorly in this situation. Atypical telephone conversation is half-duplex; the partiesalternate speaking and listening. Occasionally in aconversation both sides speak simultaneously. Withtwo legitimate voice signals simultaneously active, anecho canceller may interpret either of the two parties’speech to be echo and subtract it from the connection.This result is an call audio impairment called Speech-clipping.

Echo cancellers are found in modern digital networks attheir junction with other networks. See Figure 1.

Echo Control in Digital NetworksEcho cancellers are the superior solution for controllingecho in networks. They are less expensive than theanalog-to-digital technology of echo suppressors, andthey eliminate in-band signaling requirements. They arealso digital. Echo cancellers are typically based on off-the-self Digital Signal Processors (DSPs) that areinexpensive and easy to program.

The following characteristics qualify an echo canceller:

• Circuit Tail Delay

• Convergence Time

• Double Talk Range

The ITU G.167 recommendation for acoustic echocontrollers gives criteria for a number of performancecharacteristics. However, the state-of-the-art echocanceller should cancel across a tail delay of 128 ms;converge in 50 ms; and render speech clippingunnoticeable in double talk situations.

To further evaluate the performance of an echocanceller additional effects to test for are:

1. Residual Echo: Embedded echo. Echoresulting from too short a tail delay appearinglater than expected.

2. Convergence Loss: Embedded echoreappearing in a signal in double talk situationssounding louder than a “typical” echo.

3. Changing Acoustics: Embedded echoresulting from changing the microphonevolume over the duration of the signalsimulating acoustic echo.

4. Howling Rejection: appearance and durationof a squeal when microphone /loudspeakerfeedback is introduced.

5. Doubletalk Attenuation: Detectable loweringof speech signal volume during double talksituations.

6. Half-Duplex Communication: Elimination ofone parties signal during doubletalk situations.

7. Audible Transitions: signal volume changes,noise, and attenuation of background noisebetween words in a voice signal.

Echo and the VoIP TSThe VoIP Test System provides a number of tools tofor making voice call quality measurements includingecho detection. One of these tools: VQScope allowsyou to detect and measure echo in voice signalsrecorded off a network.

Echo DetectionVQScope used in conjunction with the VoIP TestSystem’s PSQM scoring capability can be used to detectimpairment in a transmitted and recorded audio samplein a TestBuilder created test.

Echoes are initially detected through their high PSQMscores. PSQM uses an algorithm (defined in ITU-TP.861) that determines how much a received audiosample sounds like the original sample, taking intoaccount the characteristics of human hearing.

Under PSQM, lower scores are better. Scores can rangefrom 0.0 (Perfect) to 6.5 (Poor) and higher. Audio withscores above 6.5 may still be humanly intelligible, butwould fail most carrier voice quality standards and causeautomated voice recognition applications to fail.




Hammer TestBuilder can create testsperforming PSQM scoring of calls.SEQSEQSEQSEQARABICARABICARABICARABICIn a PSQM scored call, a connection ismade between two scripts executing on separateHammer channels. PSQM consists of a quick in-bandtonal synchronization of the playing and recordingscripts, and then transmission of a single audio sample.The received audio sample is then scored to see to whatdegree it differs from the originally transmitted audiosample.

TestBuilder also has the ability to create PSQM testssimulating the important doubletalk condition, to stressecho cancellers.

Figure 2. shows the sequence of call events to score anaudio sample. Caller A is the prompt (audio sample)playing script. Caller B is the prompt recording script.The received audio sample‘s scoring is handled by theB-Side.

Caller A Caller B

Place Call

Disconnect

Start Protocol

Wait for Call

Play Prompt

Play Synch ToneVerify Tone

Network

Start Protocol

Answer Call

Record Prompt

Caller A Caller B

Place Call

Disconnect

Start Protocol

Wait for Call

Play Prompt


Start Protocol

Answer Call

Record Prompt

Disconnect

Stop ProtocolStop Protocol

Caller A Caller B

Place Call

Disconnect

Start Protocol

Wait for Call

Play Prompt


Network

Start Protocol

Answer Call

Record Prompt

Caller A Caller B

Place Call

Disconnect

Start Protocol

Wait for Call

Play Prompt


Start Protocol

Answer Call

Record Prompt

Disconnect

Stop ProtocolStop Protocol

Figure 3. PSQM Scoring Call Sequence

A higher than expected PSQM score indicates callimpairment.

Note that benchmark PSQM scores for the testequipment and System-Under-Test (or network) areneeded before actual testing begins. These benchmarksmust be established under ideal, controlled conditions.They will later be taken into account with the test resultsto determine the actual degree of call impairment

Echo AnalysisAnalyzing acoustic samples is like detective work. Echois just one of the sources of call impairment in digitalnetworks The analyst needs to be open-minded aboutwhat they might find. Also, to narrow the analysis, theanalyst must understand the signal path the samples

followed through the System-Under-Test, beforebeginning the analysis.

Crosstalk (intrusion of a foreign signal), packet loss, andjitter (variable end-to-end packet transit time) are someof the other reasons for a high PSQM score.Fortunately, call impairments have recognizablecharacteristics that can easily be exposed with analysisusing VQScope VQScope includes three tools used ina three-step process for analyzing the record of a callimpaired by echo: Acoustic Spectrum, The Echo ReturnLoss (ELR) engine and its own built-in Echo Canceller(EC) engine. Both the ELR and EC tools weredeveloped in accordance the ITU-T G.165recommendation for Line Echo Cancellers.

Subjective listening tests can also be performed withinVQScope.

VQScope allows you to do a quick visual inspection of arecorded audio sample’s acoustic signature with itsAcoustic Spectrum tool. This inspection identifiesmany types of impairments based on the observedwaveform.

For example, a high PSQM score resulting from packetloss would show a “cut-out” of several millisecondslooking like silence in an unexpected position within thesignal’s waveform. This cut-out would be clearly shownby the Acoustic Spectrum tool.

Echo also has a characteristic signature, although echo isnot as obvious to detect as packet loss. When echoimpairment is suspected the VQScope ERL tool is usedto confirm its presence..

Echo is typically responsible for a large amount of signalloss. The ERL tool graphically shows the signal lossbetween the recorded audio sample and the originalsignal. With the ERL the temporal location andmagnitude of signal loss in the recorded audio sampleare determined. Repetitive instances of signal losswithin a signal are characteristic of echo.

The EC tool provides an analysis of any echo located inthe recorded audio sample. It performs exactly like anecho canceller. Using an Empirx developed ITU-G 165compliant algorithm. It graphically displays the locationand amount of echo it detects in the recorded audiosignal. If the signal has previously been through anetwork’s echo canceller(s) it can be used to detect anyresidual echo. It also provides metrics on the amount ofecho present including the maximum strength of theecho it detected in the signal (Echo Depth).




By understanding the signal path of the System-Under-Test and using VQScope, an analyst can quicklydiagnose echo and other call impairments in a three-stepprocess.

Echo Detection ExampleThe following section is an example of how to setup,execute and evaluate an echo test using a HammerLoadBlaster 500 with TestBuilder and the VoIP TestSystem.

The example shall include the following sub-sections:

1. Setup and configure the hardware.

2. Setup and configure the test.

3. Execute the test.

4. Analyze test results for echo.

For sub-sections two through four a tutorial format isused to instruct engineers unfamiliar with VQScopehow to use the tool on a mouse-click by click basis.

Hardware Setup and ConfigurationThe System-Under-Test for this example is two AccessGateways which convert between digital telephony(ISDN) and IP telephony then back again. Figure 3below summarizes the System-Under-Test.

Note that two sets of digital telephony connections areneeded. One set connects the Hammer to itself in a“loop-back” connection. This connection is used tobenchmark the test equipment. The second setconnects the Hammer to the Gateways.

Access Gateway Access Gateway

HammerLoadBlaster 500

LAN/MAN/WAN/Internet

T1 (ISDN) T1 (ISDN)

100 MbitEthernet

100 MbitEthernet

Gatekeeper

100 MbitEthernet

T1 (ISDN Loop-back)

Access Gateway Access Gateway

HammerLoadBlaster 500

LAN/MAN/WAN/Internet

T1 (ISDN) T1 (ISDN)

100 MbitEthernet

100 MbitEthernet

Gatekeeper

100 MbitEthernet

T1 (ISDN Loop-back)

Figure 4. Echo Example Block Diagram

The details of the setup are omitted for the sake ofbrevity. However, they include:

1. Cabling the Hammer and the Gatewaystogether.

2. Setting-up the Gatekeeper to recognize androute telephone calls between the Gateways.

3. Configuring the Gateways. This includesenabling their Echo Cancellation capability.

4. Configuring the Hammer’s telephonyhardware. This primarily involves using theConfigurator to setup ISDN for use with theGateways.

5. Configuring the Hammer’s software tointerface with the Gateways. This primarilyinvolves programming the PhoneBook toplace calls through the Gateways.

6. Verify the Voice Quality Server and HammerTelephony Server are running on theHammer in the Configurator application.

Test Setup and ConfigurationTestBuilder with the VoIP Test System comes deliveredwith several tests for performing PSQM scoring. Thisexample uses the test: Voice Quality Testlocated in the TestBuilder CallProfileTestsfolder.

Open the test using TestBuilder displaying the test’sladder diagram. In this example doubletalk will besimulated. Select each of the Voice Quality Play icons inthe test’s ladder diagram. Select Properties from thepop-up menu. In the Voice Quality Properties windowcheck the Enable Doubletalk checkbox. Enable thePSQM Method’s radiobutton.




Figure 5. Test Example Ladder Diagram

Figure 4 shows the TestBuilder ladder diagram for thetest with the Voice Quality Play Properties windowopen.

Test ExecutionThe scheduling and execution of the Echo Test calls isthrough Test Profiler. The execution of the calls can beobserved on the System Monitor and the Call SummaryMonitor.

The tests need to be run in two series. The first series isto benchmark the Hammer hardware. This willdetermine the base PSQM score for the System-Under-Test. The second series is the actual “live” test of thecalls through the Gateways.

T ES T PR O F I LER

In the Test Profiler, select the PSTN SwitchedCircuit Telephony category of tests and the testnamed: Voice Quality Test.

The difference between the two test series is in theCalling Direction Box. For the first “benchmark” testassign the “loop-back” spans. These are the spans thatconnect the Hammer to itself. In the second “live” testassign the spans that connect the Hammer to the twoGateways.

In the Test Profiler’s Calling Direction box, configurethe test for at most a single ISDN span on each of theA and B sides. PSQM scoring is a CPU-intensiveprocess. Scoring with more than a span of channels on

a LoadBlaster 500 would require a remote VQS beassigned.

Finally, select Blast for the Calling Profile. TheNumber of Callers ideally should be set to themaximum number of calls supported by the receivingGateway, within the PSQM single span restriction.

B EN C H MA R K TES T

Execute the first test, using the loop-back spans usingTest Profiler.

The execution of the test can be observed from theHammer System Monitor on a per channel basis. Fromthe Call Summary Monitor PSQM scores for the callscan be observed.

Note the average PSQM score in the Call SummaryMonitor for approximately 10 minutes of calls or untilthe average PSQM score stabilizes. This average scoreis the benchmark PSQM score for the Hammer. TheSystem-Under-Test cannot produce a lower (better)score than this.

A typical PSQM score for a Hammer LoadBlaster 500with an ISDN loop-back is: 0.30.




L I V E T ES T

Halt the benchmark test and reopen the TestBuilderladder diagram for the test. In the Voice QualityPlay Properties window for each of the VoiceQuality Play icons, set the Method Threshold numberto 1.68 times the benchmark PSQM score. Adjustingthe threshold in this way retains the received audiosamples for calls likely to have impairment.

For example, if the benchmark test score is 0.30, thePSQM threshold should be set to 0.50 for sample datacollection.

Execute the second “live” test, the Gateway test usingTest Profiler with the Gateway connected spans.

Note the average PSQM score in the Call SummaryMonitor until the average PSQM score stabilizes.Observe the difference between the “live” average scoreand the benchmark average score. More importantly,note the relationship between the Threshold numberpreviously set and the Average and Maximum PSQMscores.

A live subjective listening test can also be performedduring the test using the channel monitor to hear bothsides of the call, including the doubletalk signal beingtransmitted.

Analysis is performed after the test is complete on thesaved files produced as work products by the VoIP TestSystem. This analysis identifies the source of highPSQM scores. Data will only be saved on calls whosePSQM score exceeded the Threshold.

If no PSQM scores are exceeding the threshold. Thethreshold may need to be set downward and the test re-run to collect data. The tester may have to accept thereis no call impairment in the System-Under-Test, if thelive PSQM scores are close to the benchmark average.

Figure 5 shows the Call Summary Monitor for a typical“live” test. Note the PSQM scores. In the figureshown call PSQM scores have exceeded the 0.50threshold set for the test. The example continues usinga saved audio sample with a PSQM score of 0.57.

Test Results AnalysisAnalysis is performed on calls with higher than expectedPSQM scores. The analysis is a three-step process. Atany step a diagnosis can be reached ending the process.

The three steps are:

1. Spectral Analysis

2. Signal to Noise Ratio Analysis

3. ERL

All of the analysis takes place using the single window ofthe VQScope application in the VoIP Test System

S A M PL E D A TA

Open the VQScope application. From its File menuselect Open. The Input Test Reference and Signal Fileswindow appears to load the audio sample files for theanalysis.

The first file to get is the Test Signal. This is thereceived recorded audio sample from the PSQMscoring. The samples with scores exceeding thethreshold are located in a folder named after the test.

For this example the test the data is located in:

C:\Hammer\Vqs\Save\Voice Quality Test

In this folder will be located pairs of files containing thesample data. The file names are made up of fieldsseparated by pound signs (#). The information in thethird and forth field from the left is the most important.These fields are the audio sample name and the PSQMscore.

An example of the first four fields found in the filename for this example might be as follows:

H9999#1234567#voipboy1p1.pcm#0.57#

Field three and four indicate the file contains thetransmitted and received audio sample voipboy1p1which received a PSQM score of 0.57.

From the Input Test Reference and Signal Files windowbrowse to the stored sample directory and select thesample with the highest PSQM score (fourth field) to bethe Test Signal.

Select only files in the folder that does not have apostfix. The file with the postfix pcm.phf is not anaudio sample used in analysis.

In the window’s Reference Signal box browse and clickon the file in the default directory that has the same




audio sample name as the one found in third field of theTest Signal file previously selected.

In this example, it would be voipboy1p1.pcm.

Click OK, and VQScope will load with recorded sampledata and the reference signal ready to begin the analysis.

Figure 6. Call Summary Monitor for Live Test

S PEC TR A L A N A LY S I S

The first step is to perform a spectral analysis by visuallycomparing the sonic signature of the transmitted audiosample with the audio sample received. This is donewith the Layered View View of VQScope. Figure6 shows the sonic signature of the impaired call.

Foreign SignalForeign Signal

Figure 7. Sonic Signature of Impaired Call

The figure shows the test and reference audio samplesoverlaid upon each other. The graph is signal amplitudein Hz over time. The time axis is in terms of sampleswhere the sample rate is 16K Hz.

A circle has been placed around an area of the graphand labeled “Foreign Signal”. This is a signal that wasnot present in the reference sample transmitted, butappears in the recorded sample. This foreign signalappears as an “overlap” or a “shadow” of a differentcolor on the reference signal’s graph. “Overlaps” or“shadows” are characteristic of echo. They can also becrosstalk.

Note that this analysis would also be able to detect callimpairments resulting from packet loss or jitter. Thisgraph can provide an explanation for high PSQM scoreswithout having to perform any further analysis.

In addition to a visual analysis, a subjective listening testcan also be performed on the recorded audio sample todetect impairment by clicking on Play in the TestSignal graph window.

S I G N A L TO N O I S E R A T I O A N A L Y S I S

The second step is to perform a signal to noise ratioanalysis of the impaired call. This is done using theEcho Return Loss View of the VQScope. Figure7 Shows the Signal to Noise distribution of the recordedsignal.




~25 ms.~25 ms.

Figure 8. ERL Graph of Impaired Call

Figure 7 shows the strength of the foreign signal in therecorded signal. The graph is power in dB over time.The time axis is in terms of samples where the samplerate is 16K Hz.

Attention should be paid to the section of the signal thatcontains the foreign signal detected in the previousanalysis step.

A brace has been placed around an area of the graphand labeled “~25 ms”. Duration can be approximatedby dividing the sample numbers shown on the graphticks by the 16,000 Hz sample rate to arrive atmillisecond duration. From the graph a foreign signalof about 40 dB with duration of about 25 ms can beestimated to exist in the received audio sample.

Foreign signals with a low enough amplitude cansometimes be “buried” in the received signal and goundetected. This graph can also be used to uncover anyadditional foreign signal that might have goneundetected in the Spectral Analysis. Note from theFigure 7’s graph, that the remainder of the recordedsignal is relatively “clean”. . Most echoes repeat withina signal. There is no repetition of the discovered foreignsignal. This would argue against echo impairment of thecall

E R L A N A L Y S I S

The third step is to perform an ERL analysis of theimpaired call. This is done using the EchoCanceller View of the VQScope. Figure 8 Showsthe residual (echo?) signal that the EC tool couldremove from the recorded signal.

EchoEcho

Figure 9. Residual Signal Echo Cancellation

The figure shows the strength and duration of theresidual echo in the received recorded signal. The graphis power in dB over time. The time axis is in terms ofsamples where the sample rate is 16K Hz. Not shownon this figure (but shown in the display) is themaximum strength of the extracted signal: -63.2 dB.

A circle has been placed around an area of the graphand labeled “Echo”. The dense graph pattern displayedis characteristic of an echo and meets all the criteria ofecho listed in ITU-G 165.

Figure 9 shows the entire VQScope window used in thisanalysis. Note the PSQM score and additional voicequality metrics at the bottom of the window. Inaddition, the Reference and Test (received andrecorded) audio samples are shown in separatewindows.




Figure 10. VQScope Window of Impaired Call

Final AnalysisFor the purposes of this example, a final more detailedspectral analysis was performed to verify the results.This is not shown.

The Cool Edit off-the-shelf sound editing softwarewas used to inspect the audio frequency spectrum of thefragment of foreign signal in detail.. This analysisverified that that the “echo” found in this example hasthe acoustic spectrum of the doubletalk signaltransmitted by the receiver while recording.

From this, we can determine the Gateway detected thereflected doubletalk signal and successfully cancelled it.However, approximately 25 ms. of echo “leaked”through the canceller. There is also the possibility theecho was present earlier in the signal, but wentundetected. The 25 ms of echo is probably results fromthe time the Gateway’s echo canceller took to convergeon and eliminate the doubletalk signal.. Further testswith other audio samples would be needed to accuratelydetermine the Gateways: tail delay, convergence timeand full double talk range.

Cool Edit is a trademark of the Syntrillium Corp.

HYPERLINKSEQARABICHYPERLINKSEQARABICSEQARABICSEQAR

ABICSummaryAll voice calls on digital networks require some form ofecho cancellation. These cancellers are deployed oneither side of the compressed signal paths. To achievehigh voice quality telephone calls, understanding theecho effect, how the telecom industry deals with it, andhow the Hammer can be used to detect and analyzeecho are very important.

This Technical White Paper described how to use theHammer VoIP Test System perform echo detectionand analysis.

It discussed electrical and acoustic echo and theirsources. In addition, it discusses echo cancellation andthe criteria for evaluating it.

It discussed testing for echo. A good testing strategy, aHammer LoadBlaster, a TestBuilder built test, and theVQScope application are the tools needed to detect andanalyze echo in telephone calls.

Cool Edit is a trademark of the Syntrillium Corp.




Finally, it gave a step-by-step tutorial of how to hunt areal echo in a network System-Under-Test.

The information in this white paper can be used toexpand the Hammer’s capability to perform voicequality testing on new classes of communication andnon-communication devices that previously could noteasily be tested with TestBuilder.

Technical white papers are a service provided by CommunicationsInfrastructure Test Group Engineering. They are intended to informour colleagues and customers on the Hammer’s capabilities. They arefor the exclusive use of Empirix customers and employees only.

communications infrastructure test group …wireless.feld.cvut.cz/mesaqin2002/echo_detection.pdf ·...

Documents