CN1306352A

CN1306352A - Test method of echo cancel function

Info

Publication number: CN1306352A
Application number: CN00114084A
Authority: CN
Inventors: 张震玮
Original assignee: ZTE Corp
Current assignee: CHINA TECHNOLOGY EXCHANGE Co Ltd; State Grid Beijing Electric Power Co Ltd; State Grid Economic and Technological Research Institute
Priority date: 2000-03-01
Filing date: 2000-03-01
Publication date: 2001-08-01
Anticipated expiration: 2020-03-01
Also published as: CN1167213C

Abstract

A test method for an echo canceller, which firstly collects the actual or customized signal; then processes the signal according to the assumed echo channel at each stage to make it a corresponding echo; uses the original signal and the echo as the remote and Near-end input; capture the output signal and filter coefficients and restore them to a curve, and compare them with the expected results. By adopting the above steps, the correct implementation test, functional integrity test, efficiency test, working condition test at the boundary point, robustness test and comprehensive test of the echo cancellation code can be carried out. The invention gets rid of the limitation of the actual environment, saves the cost, and provides comprehensive test basis.

Description

A kind of method of testing of echo cancel function

The present invention relates to a kind of measuring technology of communication system, relate in particular to the measuring technology of echo cancel function in the communication system.

Active user's telephone set all passes through to support the two-wire line (being called local loop) of transmitted in both directions to link telephone office, but in the toll telephone transmission system, for the ease of the transmission of speech, two transmission directions must be connected on respectively on two pairs of lines separately, thereby form four-wire circuit.The device that connects two wires and four lines part is called hybrid circuit.Because the impedance mismatching of hybrid circuit has produced echo, make the voice signal that comes from first speaker can not be transferred to listener's phone fully, the voice signal of reflection has been got back to the first speaker phone again through oppositely circuit transmission, makes first speaker hear own irritating echo.And in wireless system,, under the toll message state, can not hear echo significantly even make because information source encoding and decoding and channel coding/decoding are handled the time-delay that produces yet.

Existing communication network adopts the DSP digit chip more, and the echo canceller of making by the adaptive digital filtering technology (echo canceller of making as the TMS320C54X series of utilizing American TI Company) is arranged on hybrid circuit place, place, base station or fixed station place and eliminates the echo that local terminal produces.

Though the echo canceller that the method that utilization was mentioned is just now made can be eliminated echo theoretically, how effect needs test just can learn on earth.As everyone knows, fully in actual environment the echo of simulation various requirement to test whether it can both offset almost be infeasible, and test also is extremely inconvenient in actual environment.In addition, in the time of need testing in actual environment by the time, echo canceller has arrived the finished product stage, and at this time test just gets involved the efficient that can reduce exploitation greatly.If the hardware emulator that adopts DSP manufacturer to provide except effect is undesirable, has also brought the rising greatly of development cost.More of paramount importance is not have perfect method of testing for echo canceller, and existing method both can't guarantee to offset the quality of echo, and the test result that provides of test phase can not be provided again, for the developer analyzes, locatees and deal with problems foundation is provided.

U.S. Pat 4727566 example that comes to this, its main contents are in data communication network (two or four line conversion portion), the two wires part is disconnected, connect testing equipment, it produces certain encoded data signals, import as far-end, near-end is input signal not, receive data at the echo canceller output, and press the positive and negative threshold comparison of positive-negative polarity and formulation, promptly be designated as 1 when surpassing threshold value, think error code, otherwise be 0, thereby draw the error rate (promptly producing), so that whether pass judgment on this echo canceller qualified by test point place two or four line conversion portions.And this patent only is applicable to the echo that the conversion of two or four lines produces in the data communication network, inapplicable wireless communication system.Because it must be tested in actual environment, can only be used to verify the work correctness of the echo canceller of (echo channel that only has one or more exists) under particular system, the specific actual conditions, and arrive the finished product stage afterwards.And according to this United States Patent (USP), also need to develop actual equipment, brought inconvenience and development cost is risen.In addition, this United States Patent (USP) is merely able to provide under specific circumstances, the conclusion that whether qualified echo canceller is, and do not divide different test phases to carry out deep test, and comprehensively corresponding test result can't be provided, can't provide foundation for analyzing, locating to deal with problems.

The object of the present invention is to provide a kind of method of testing of echo canceller, this method can be tested the effect of echo canceller code more fully and effectively, and can not test echo neutralization effect, thereby provide foundation for analyzing, locate the solution echo problem by constructing actual test environment.

The method of testing of a kind of echo canceller that the present invention proposes may further comprise the steps:

The first step: with produce under the actual conditions or gather according to the signal of rules such as robustness testing customizations;

Second step: the various echo channels according to each test phase hypothesis are handled these signals, make it to become corresponding echo signal;

The 3rd step: original signal and corresponding echo signal are imported echo canceller as far-end and near end signal;

The 4th step: the coefficient of catching corresponding output signal and echo cancellation filter;

The 5th step: output signal and filter coefficient are reverted to curve;

The 6th step: compare with former remote signaling input curve and by the echo channel coefficient curve of test request simulation, whether differentiate correctness that echo canceller echo cancel function and performance under various actual conditions or under situations such as robustness realize.

Described first step has used the acquisition mode of voice data, and the collection of voice data is used to customize the test case of various reality, and is converted to near-end or the far-end input of corresponding data format as described third step echo canceller.

Described second step is handled and is made it to become the method that meets the echo signal that certain customization requires gathering the data of coming according to the different echo channels of hypothesis, and promptly the data that will gather according to the time-delay number of echo channel are delayed time accordingly.

Described third step is with echo signal after the time-delay and original signal (remote signaling) input echo canceller, by software simulation emulation and call the echo cancelltion code, obtains test data.

Described the 4th step is test result to be deposited in the corresponding file automatically by the analog simulation software platform;

Test data or the next data recovery of customization collection that described the 5th step obtains the analog simulation software platform become corresponding curve form so that relatively differentiate.

Adopt technical scheme of the present invention, can carry out the test of six aspects to the echo canceller code, promptly the efficiency test of the correct realization property testing of echo cancelltion code, perfect in shape and function property testing, echo cancelltion, echo cancelltion code are in working condition test, the robustness testing of echo cancelltion code, the integrative test of echo cancelltion code at boundary point place.

The present invention need not be by making up actual test environment but the correctness of testing the echo canceller code by the comprehensive simulated method of emulation, and the restriction of actual environment has been completely free of in test, and has improved development efficiency greatly; Secondly can not buy expensive simulation hardware equipment, only need carry out the Computer Simulation test, save cost, and simple; Once more, to the echo cancelltion code can be in steps, testing stage by stage, make this test not only provide a row and not all right conclusion, but the method for testing and the foundation of an each side be provided comprehensively, be convenient to analyze and deal with problems.And the present invention is not only applicable to long-distance phone network, also is applicable to wireless system.

Also in conjunction with specific embodiments the present invention is described in detail below by accompanying drawing.

Fig. 1 is the theory diagram of echo canceller;

Fig. 2 a is the flow chart of echo cancelltion adaptive filter coefficient renewal process;

Fig. 2 b is the main program flow chart of echo canceller method of testing of the present invention;

Fig. 2 c is the flow chart that input signal is delayed time customization echo;

Fig. 3 a is the coefficient curve figure (iteration 4000 points) that far-end input sinusoidal signal, near-end are input as the echo cancellation filter of 0 (also not having echo-signal);

Fig. 3 b is under Fig. 3 a situation, the output signal curve figure of this echo cancellation filter (iteration 4000 points);

Fig. 3 c is that remote signaling is 0, the coefficient curve figure (iteration 4000 points) of the echo cancellation filter of near-end input sinusoidal signal;

Fig. 3 d is under Fig. 3 c situation, the output signal curve figure of this echo cancellation filter (iteration 4000 points);

Fig. 3 e is that the far-end near-end is imported sinusoidal signal respectively, the coefficient curve figure of this echo cancellation filter (iteration 4000 points);

Fig. 3 f is under Fig. 3 e situation, the output signal curve figure of this echo cancellation filter (iteration 4000 points);

Fig. 3 g is the echo cancellation filter coefficient curve figure (iteration 4000 points) that far-end input sinusoidal signal, near-end are imported its echo signal;

Fig. 3 h is under Fig. 3 g situation, the output signal curve figure of this echo cancellation filter (iteration 4000 points);

Fig. 4 a is the output signal curve figure (iteration 8000 points) that far-end input voice signal, a near-end are imported the echo cancellation filter of its echo signal;

Fig. 4 b is the output signal curve figure (iteration 8000 points) that far-end is input as 0, near-end is imported the echo cancellation filter of a voice signal;

Fig. 4 c is that far-end is imported the coefficient curve figure (iteration 8000 points) that a voice signal, near-end are imported the echo cancellation filter of its double echo signal;

Fig. 4 d is under Fig. 4 c situation, the output signal curve figure of this echo cancellation filter (iteration 8000 points);

Fig. 5 a is the filter coefficient curve chart of far-end when importing a voice signal, near-end and importing 1000 of the echo cancellation filter iteration of its echo signal;

Fig. 5 b is the filter coefficient curve chart of far-end when importing a voice signal, near-end and importing 3000 of the echo cancellation filter iteration of its echo signal;

Fig. 5 c is the filter coefficient curve chart of far-end when importing a voice signal, near-end and importing 5000 of the echo cancellation filter iteration of its echo signal;

Fig. 5 d is the filter coefficient curve chart of far-end when importing a voice signal, near-end and importing 8000 of the echo cancellation filter iteration of its echo signal;

Fig. 6 a is that near-end is input as 0, far-end input voice, and the coefficient curve figure of anechoic echo cancellation filter (iteration 2000 points);

Fig. 6 b is under Fig. 6 a situation, the output signal curve figure of this echo cancellation filter (iteration 2000 points);

Fig. 6 c is that far-end input voice, near-end input time delay are the coefficient curve figure (iteration 8000 points) of 0 echo cancellation filter;

Fig. 6 d is under Fig. 6 c situation, the output signal curve figure of this echo cancellation filter (iteration 8000 points);

Fig. 6 e is that far-end input voice, near-end input time delay number have just equaled the coefficient curve figure (iteration 20000 points) of echo cancellation filter that this echo canceller is offset the echo signal of ability:

Under Fig. 6 f Fig. 6 e situation, the output signal curve figure of this echo cancellation filter (iteration 20000 points);

Fig. 7 a is that voice signal, the near-end that far-end input one is disturbed by direct current is input as this echo cancellation filter coefficient curve figure (iteration 8000 points) of 0;

Fig. 7 b is under Fig. 7 a situation, the output signal curve figure of this echo cancellation filter (iteration 8000 points);

Fig. 7 c is that far-end is imported a voice signal, near-end is imported the coefficient curve figure (iteration 8000 points) of its time-delay number much larger than the echo cancellation filter of the echo signal of this echo canceller echo cancelltion ability;

Fig. 7 d is under Fig. 7 c situation, the output signal curve figure of this echo cancellation filter (iteration 8000 points);

Fig. 7 e is that far-end is imported a voice signal, near-end is imported the coefficient curve figure (iteration 40000 points) of its time-delay number much larger than the echo cancellation filter of the echo signal of this echo canceller echo cancelltion ability;

Fig. 7 f is under Fig. 7 e situation, the output signal curve figure of this echo cancellation filter (iteration 40000 points);

When Fig. 8 a is integration test, the situation map of far-end and near end signal combinatorial input;

Fig. 8 b is far-end and near end signal when importing with Fig. 8 a situation, the coefficient curve figure of this echo cancellation filter (iteration 30000 points);

Fig. 8 c is far-end and near end signal when importing with Fig. 8 a situation, the coefficient curve figure of this echo cancellation filter (iteration 40000 points);

Fig. 8 d is far-end and near end signal when importing with Fig. 8 a situation, the coefficient curve figure of this echo cancellation filter (iteration 60000 points).

In order to fully understand method of testing of the present invention, at first analyze the operation principle and the mode of echo canceller.As the basic principle of Fig. 1 for the echo canceller employing, far-end speech x (n), near-end speech v (n) leaves x (n) reflection the signal imitation of hybrid circuit and passes through a unknown echo channel for x (n), produce echo-signal y (n), this echo-signal and near-end voice signals addition form r (n).Sef-adapting filter be input as x (n), be output as echo replica signal

\hat{y} (n)

。Use echo residual signal

e (n) = r (n) - \hat{y} (n)

Remove to regulate the coefficient of filter, make echo replica signal

\hat{y} (n)

Follow the tracks of echo-signal y (n), offset echo-signal with this.The object that the present invention tested---echo canceller also is to adopt this principle to realize substantially.Following principle is followed in the renewal of its filter coefficient:

w (n + 1) = w (n) + μ \frac{e (n) * x (n)}{E_{xx} (n)}

Wherein: w (n) is the coefficient of filter;

X (n) is reference signal input vector, i.e. far-end speech;

E (n) is an echo residual signal;

μ is a step-length;

E_{xx} (n) = Σ_{i = 0}^{N - 1} [x (n - i)]^{2}

, be the quadratic sum calculated energy estimated value of N up-to-date sampling;

Echo canceller is when realizing echo cancelltion, groundwork be make the coefficient w of echo cancellation filter (n) level off to actual echo channel coefficient (convergence of coefficient is realized by certain algorithm, its major consideration is the difference of the output signal of filter with actual echo signal), thus make far-end speech offset with the actual echo signal that produces by the signal of echo cancelltion sef-adapting filter output.

Echo cancelltion adaptive filter coefficient updates process is: judge that at first far-end has or not voice,, far-end do not carry out the coefficient update of echo cancellation filter if not having voice, if far-end has voice to judge that again near-end has or not voice, if near-end does not have the coefficient update that voice then carry out echo cancellation filter yet, if near-end has voice then not carry out the coefficient update of echo cancellation filter.Fig. 2 a promptly is the flow chart of adaptive filter coefficient renewal process.Step 201 is when far-end has voice signal, at DSP digit chip (the existing communication net adopts the DSP digit chip to make echo canceller more) interrupt call echo cancelltion code; Step 203 is read in the input data from the address of variable of near-end, far-end, each port of echo signal; And judge whether to have imported again 128 sampling points (this echo canceller is for offsetting echoes every 128 sampling points) in step 205; If input sample is 128 of less thaies still, then forward step 211 to, continue input sample and sampling point counter and add one; If input sample is reach 128, then step 207 is calculated the output of echo cancellation filter and is carried out echo cancelltion; And step 209 is with the sampling point counter O reset; Step 213 is judged when far-end has signal again, and whether near-end no signal; If near-end no signal then enter step 215 is upgraded the coefficient of echo cancelltion sef-adapting filter by certain algorithm; If near-end has signal then non-update coefficients, step 217 is upgraded the address of variable of each port so that read in new data when calling next time, and step 219 interrupts returning.

According to the example of above-mentioned echo canceller, the main flow process of echo canceller method of testing of the present invention is shown in Fig. 2 b.At first step 301 is carried out the initial work that some have related parameter, needs the number of samples of computing when then step 303 is provided with this test; Step 305 is read in the remote signaling data from the corresponding port of internal memory, i.e. the signal that customizes according to different test purpose collections; In like manner, step 307 is read in near-end and echo signal again, is similarly the signal according to different test purpose customizations or collection, and echo signal is that remote signaling produces through certain delay process; Step 309 is called the echo canceller code and is carried out the echo cancelltion test; Step 311 is read dateout from corresponding internal memory port, i.e. the coefficient of echo canceller output signal and echo cancellation filter; Step 313 judges that whether full original computing that is provided with count, if expired then end the operation of program and return, continues to read in far-end near-end and echo signal otherwise return step 305, carries out echo cancelltion; Echo cancellation filter coefficient that draws at last and output signal thereof compare with expected results after reverting to curve form.In addition, because actual echo cancelltion DSP assembler carried out by down trigger, therefore can be according to this point, in this testing process, the input/output interface of modified module makes originally the reading of port, and changes reading internal memory into.In this test main flow, the analog simulation software platform SIMULATE that all uses DSP reads in the remote data or the near-end data (containing the echo data according to the customization of hypothesis echo channel) of customization from particular port with document form, and catches the coefficient (being 128 rank in the example) of echo cancelltion sef-adapting filter and the output signal data (form with file deposits in) of echo canceller at specific port.

When the analog simulation software platform SIMULATE of use DSP carries out the analog simulation test, the input customization data, call tested object (echo canceller code), the output echo canceller output signal and the coefficient of its sef-adapting filter accordingly, for the judgement of test result then by being that graphic form is judged with data conversion such as input, output signal and adaptive filter coefficients.Adopt MATLAB software to transform, show in this example.

Wherein, echo cancelltion adaptive filter coefficient as tested object in this test case is 128 rank, therefore it can only finish 16ms echo (under the actual conditions at most, the echo that is about more than 2000 kilometer generation) counteracting, and because actual echo channel is slowly to change, so in this test case, unless have outside the particular requirement for the input echo signal, the echo channel of general body plan is 64 rank (under the actual conditions, being about the echo of more than 1000 kilometer generation)

Customization as for echo signal then is that the input signal that will collect is to obtain after remote signaling (being converted into data format) is delayed time accordingly according to the time-delay millisecond number of the echo channel of hypothesis.To input signal delay time the customization echo main flow shown in Fig. 2 c.At first step 401 is read near end signal, secondly step 403 is read in remote signaling, and differentiate remote signaling in step 405 and whether exist, then thereby step 407 echo channel as requested postpones several to the remote signaling echogenicity signal of delaying time accordingly if present, otherwise do not handle, final step 409 constitutes the echo signal that produces an input signal of echo canceller with the near end signal addition.Therefore flow chart 2c is actually the time-delay number that is used for according to the echo channel that requires in each test order, to remote signaling carry out delay process so that the echogenicity signal process of (and constitute an input signal with the near end signal addition), the data of its generation finally are used for testing the step 307 of main flow chart 2b.

Adopt above technical scheme, can specifically test at different test purposes the echo canceller code segment, present embodiment exemplifies the test of following six purposes, be equivalent to six test phases, details are as follows for the test purpose in each stage and concrete method of testing: one, the correct realization of test echo cancelltion code

The test of echo cancelltion code correctness, the signal that is about to non-reality is imported as remote signaling, and construct corresponding echo signal, main program calls the echo cancelltion code by the analog simulation software platform SIMULATE of DSP after tested, catch the coefficient of output signal and echo cancellation filter, again these signals are reverted to curve and compare and differentiate or compare, whether correctly realized the acoustic echo cancellation adaptive algorithm of desired design to differentiate this echo cancelltion code with the analog echo channel of customization.

The echo canceller code should be independent of its input signal.If during the input distinctive signal, this code still can normally move, and illustrates that then its mapping for acoustic echo cancellation adaptive algorithm is correct.

Whether the present invention mainly tests this echo cancelltion code as the method for input by sinusoidal signal and moves correctly behind over-sampling and pcm encoder.In addition, according to the working method of the echo canceller of this example: have remote signaling that echo is just arranged, during the echoless channel, echo canceller is not worked, and causes design aspect the echo signal two according to echoless channel and remote signaling when the structure test case.Specifically work by carrying out as following table 1:

Table 1

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. produce the sinusoidal signal of two different frequencies, carry out the 13BitPCM coding.	Do not have	Do not have
2. simulate the echo channel on one 64 rank.(the echo channel time-delay is 8ms)	Do not have	Do not have		Do not have	Do not have
	Do not have	Do not have	3. a sinusoidal signal is imported as far-end, near-end is input as 0.(far-end exists, and near-end does not have, echoless)	The coefficient of filter is 0, and output signal is 0.	As Fig. 3 a, Fig. 3 b
4. far-end is input as 0, sinusoidal signal of near-end input.(far-end does not have, and near-end exists, echoless)	The coefficient of filter is 0, and output signal is a near end signal.	As Fig. 3 c, Fig. 3 d		The coefficient of filter is 0, and output signal is 0.	As Fig. 3 a, Fig. 3 b
		As Fig. 3 c, Fig. 3 d	5. two sinusoidal signals are respectively as near-end, far-end input.(far-end, near-end all exist, echoless)	The coefficient of filter is 0, and output signal is a near end signal.	As Fig. 3 e, Fig. 3 f
6. a sinusoidal signal is imported the echo signal of near-end input simulation as far-end.(far-end exists, and near-end does not have, and echo is arranged)	The coefficient of filter is the echo channel of simulation, and output signal moves closer in 0.	As Fig. 3 g, Fig. 3 h			As Fig. 3 e, Fig. 3 f

According to table 1 listed concrete steps and data, the result of this test phase is as follows:

When far-end exists, near-end does not have and the echoless situation under, the coefficient of filter is 0 (shown in Fig. 3 a), output signal is 0 (shown in Fig. 3 b).

When far-end does not have, near-end exists, under the echoless situation, filter coefficient is 0 (shown in Fig. 3 c), output signal is near end signal (shown in Fig. 3 d).

When far-end, near-end all exist, under the echoless situation, filter coefficient is 0 (shown in Fig. 3 e), and output signal is near end signal (shown in Fig. 3 f).

When far-end, echo signal existence, this routine filter coefficient does not converge to (shown in Fig. 3 g) on the analog echo channel coefficients, and output signal is unmatched to 0 (shown in Fig. 3 h).

The actual result of this test phase output is different with expected results, and the echo cancelltion code not mapping fully on algorithm is realized as tested object then is described, need make an amendment.Two, the basic function of test echo canceller

The functional test of echo canceller, being about to a voice signal gathers, and it is processed into certain echo signal (according to the difference of the echo channel of test request simulation and difference, and can comprise multiple echo), again this voice signal and echo signal are imported as the far-end and the near-end of echo canceller, the various situations that produce according to actual capabilities are with different compound modes, main program calls the echo cancelltion code by analog simulation software platform SIMULATE simulation after tested, catch the coefficient of output signal and echo cancellation filter, these signals are reverted to curve with near end input signal or analog echo channel ratio, differentiate this echo canceller and whether can reach the functional requirement of design in advance.

Remote signaling and near end signal do not have correlation, and remote signaling is relevant with its echo signal, according to this relation, can construct various actual test cases.

Be the integrity that test echo canceller function realizes, the present invention mainly under the situation of analog simulation by covering the input condition (and considering the situation that secondary echo exists) that institute might occur, observation output situation.

When echo situation of test, specifically work by carrying out as following table 2:

Table 2

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. simulate the echo channel on one 64 rank.	Do not have	Do not have
2. a voice signal is imported as far-end, and near-end is imported its echo signal.	Output signal is 0.	As Fig. 4 a	1. simulate the echo channel on one 64 rank.	Do not have	Do not have
	Output signal is 0.	As Fig. 4 a	3. far-end is input as 0, voice signal of near-end input.	Output signal is a near end signal.	As Fig. 4 b
4. far-end and near-end input speech signal simultaneously.(containing echo)	Output signal is a near end signal.	Figure slightly	3. far-end is input as 0, voice signal of near-end input.	Output signal is a near end signal.	As Fig. 4 b

According to table 2 listed concrete steps and data, the result of this test phase is as follows:

When input far-end and echo signal, output signal levels off to 0 gradually, and shown in Fig. 4 a, wherein signal 1 is represented echo signal, signal 2 expression output signals.

When remote signaling is 0, when only importing near end signal, this routine output signal is similar near end signal, is expected results, shown in Fig. 4 b, and the voice signal of signal 3 expression near-ends inputs wherein, signal 2 expression output signal, two signal approximately equals as can be seen.

When far-end near-end all during input signal, output signal is near end signal (figure slightly).

According to the test result of previous stage, this example is amended echo canceller code.The actual result of this test phase output is identical with expected results, illustrates that then this echo canceller code can be offset echo of 64 rank echo channels generations and the working method of this echo canceller meets desired design.

When twice echo of test, specifically work by carrying out as following table 3:

Table 3

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. simulate two exponent numbers, 16 and 64 echo channel respectively.	Do not have	Do not have
2. a voice signal is as input, and near-end is imported echo signal.	Output signal is 0.	As Fig. 4 c, Fig. 4 d		Do not have	Do not have
	Output signal is 0.	As Fig. 4 c, Fig. 4 d	3. far-end is input as 0, voice signal of near-end input.	Output signal is a near end signal.	Figure slightly
4. far-end and near-end input speech signal simultaneously.(containing echo)	Output signal is a near end signal.	Figure slightly	3. far-end is input as 0, voice signal of near-end input.	Output signal is a near end signal.	Figure slightly

According to table 3 listed concrete steps and data, the result of this test phase is as follows:

When input far-end and echo signal, the filter coefficient curve convergence is (shown in Fig. 4 c) to the analog echo channel; Output signal levels off to 0, and shown in Fig. 4 d, wherein signal 1 is represented echo signal, signal 2 expression output signals, and as can be seen, signal 2 is convergences 0.

When far-end is input as 0, near-end is imported a voice signal, and output signal is near end signal (figure slightly).

When far-end, near-end and echo signal are imported simultaneously, output signal is similar near end signal (figure slightly).

Above result all reaches expected results.

Because the actual result of this test phase output is identical with expected results, illustrate that then this echo canceller code can offset twice echo of rank, 64 and 16 rank echo channel generation, the working method of this echo canceller meets desired design.Three, the efficient of echo is offset in test

The efficiency test of echo canceller, be about to a speech signal collection, and be processed into certain echo signal, again these signals are imported as far-end input and near-end, call the echo cancelltion code by analog simulation software platform SIMULATE simulation, when different iterationses, catch the coefficient of corresponding echo cancelltion sef-adapting filter, draw the data that are associated that one group of iterations and echo suppress effect (also being that adaptive filter coefficient converges on the analog echo channel), thus the effectiveness of performance of this echo canceller of qualitative discrimination.

Under the normal condition, after through 5000～6000 iteration, the output of filter should be able to be offset the echo that produces through echo channel as the echo cancelltion sef-adapting filter of tested object.Be the efficient of test echo canceller, the echo channel on one 64 rank of structure, and only add far-end speech signal, and after the inferior operation of 1000,3000,5000,8000 (choosing of number of times do not have particular requirement), the coefficient of difference output filter.After the observation process how many times iteration, the coefficient of filter converges to the analog echo channel, considers that this test is the evaluation of an efficient aspect, therefore mainly passes judgment on qualitatively in this test.Specifically work by carrying out as following table 4:

Table 4

Step	Expected Results	Experimental result
Step	Expected Results	Experimental result	1. simulate the echo channel on one 64 rank.	Do not have.	Do not have
2. a voice signal is imported as far-end, near-end input echo signal.	Do not have.	Do not have	1. simulate the echo channel on one 64 rank.	Do not have.	Do not have
	Do not have.	Do not have	3. it is 1000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 a
4. it is 3000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 b	3. it is 1000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 a
4. it is 3000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 b	5. it is 5000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 c
6. it is 8000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 d	5. it is 5000 points that iterations is set.	The coefficient of record filter.	As Fig. 5 c

According to table 4 listed concrete steps and data, the result of this test phase is as follows:

When iterations is respectively 1000,3000,5000 and 8000 the time, filter coefficient curve convergence situation is respectively shown in Fig. 5 a, 5b, 5c, 5d.According to the improvement of these curve convergence situations and the relation of iterations, as can be seen, when filter (time spent 1S) after iteration 8000 times relatively levels off to the coefficient curve of analog channel.

From above test result, illustrate that the operational efficiency of this echo canceller still can.Four, the disposal ability of test echo canceller code under border condition

The echo canceller code is in the test of boundary point place working condition (considering the situation of a people and many people speech), being about to a voice signal gathers, and it is processed into does not have echo, echo time-delay be 0 and the echo time-delay signal that reaches the critical condition of this echo canceller import with near-end as the far-end of this echo canceller respectively, call the echo cancelltion code by analog simulation software platform SIMULATE simulation, and the coefficient of seizure corresponding output signal and echo cancellation filter, and revert to corresponding curve, thereby judge the correctness of the operating state of this echo canceller on these boundary points.

Because the coefficient of filter is 128 rank, promptly can offset 16 milliseconds time-delay at most, so the boundary point of this echo canceller is 0～16ms.Specifically work by carrying out as following table 5:

Table 5

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. a voice signal is imported as far-end, and near-end is input as 0.(there is not situation in echo)	Filter coefficient is 0, and output signal is 0.	As Fig. 6 a, Fig. 6 b
2. simulate the echo channel of 0 millisecond of a time-delay	Do not have.	Do not have		Filter coefficient is 0, and output signal is 0.	As Fig. 6 a, Fig. 6 b
	Do not have.	Do not have	3. a voice signal is imported as far-end, and near-end is input as 0, and the echo time-delay is 0ms.	Filter coefficient is the analog echo channel, and output signal is 0.	As Fig. 6 c, Fig. 6 d
4. simulate the echo channel of 16 milliseconds of time-delays	Do not have.	Do not have			As Fig. 6 c, Fig. 6 d
	Do not have.	Do not have	5. a voice signal is imported as far-end, and near-end is input as 0, and the echo time-delay is 16ms	Filter coefficient is for being similar to the analog echo channel, and output signal is for should level off to 0.	As Fig. 6 e, Fig. 6 f

According to table 5 listed concrete steps and data, the result of this test phase is as follows:

When echo did not exist, filter coefficient was 0 (shown in Fig. 6 a), and output signal is 0 (shown in Fig. 6 b).

When echo was time-delay in 0 second, the filter approximate convergence is (shown in Fig. 6 c) to the analog echo channel, and output signal has obvious distortion (shown in Fig. 6 d).

When echo signal was the 16ms time-delay, filter can not be restrained (shown in Fig. 6 e), and output signal is not 0 (shown in Fig. 6 f).

Second, third kind phenomenon does not all reach the expection requirement.

This test phase second the third phenomenon does not reach the expection requirement, illustrates that the echo cancelltion code routine processes on (0 rank and 128 rank) on two boundary points as tested object has problem.Five, the robustness of test echo cancelltion code

The robustness testing project of echo canceller mainly is divided into two kinds of situations, first kind of situation promptly gathered some interference signals, as noise, direct current signal, utter long and high-pitched sounds, it is processed into certain echo signal, again these signals are imported echo canceller as far-end and near-end respectively, SIMULATE calls the echo cancelltion code by the analog simulation software platform, catch corresponding output signal and filter coefficient, these signals are reverted to curve, to judge the operating state of this echo canceller under disturbed condition.Second kind of situation then is that a voice signal is processed into the echo-signal that surpasses this echo canceller designed capacity far away, respectively as far-end and near-end speech input echo canceller, SIMULATE calls the echo cancelltion code by the analog simulation software platform, catch corresponding output signal and filter coefficient, it is reverted to curve, qualitatively judge the performance of this echo canceller when far surpassing its designed capacity and whether can stand.

Above-mentioned two kinds of situations concrete as: the input voice are subjected to noise, howling or direct current when disturbing and echo time-delay when surpassing 16ms.Specifically work by carrying out as following table 6:

Table 6

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. simulate the echo channel of 8 milliseconds of time-delays.(64 rank)	Do not have.	Do not have
2. a voice signal that disturbed by direct current signal is imported as far-end, and near-end is input as 0.	Filter coefficient is the analog echo channel, and output signal is 0.	As Fig. 7 a, Fig. 7 b		Do not have.	Do not have
		As Fig. 7 a, Fig. 7 b	3. simulate the echo channel of 32 milliseconds of time-delays	Do not have.	Do not have
4. a voice signal is imported as far-end, and near-end is input as echo signal.	This filter should be offset this kind echo, but should not have a strong impact on the service behaviour of this filter.	As Fig. 7 c, Fig. 7 d, Fig. 7 e, Fig. 7 f		Do not have.	Do not have

According to table 6 listed concrete steps and data, the result of this test phase is as follows:

When input is subjected to the voice signal of direct current interference, filter coefficient can not be restrained (shown in Fig. 7 a), output signal is not equal to 0, shown in Fig. 7 b, wherein signal 1 is an echo signal, and signal 3 is an input signal, and signal 2 is an output signal, output signal 2 has obvious distortion phenomenon, and this phenomenon is different from expected results.

When importing the echo signal of 32ms time-delay, filter can't be restrained, and output signal has obvious distortion phenomenon, 1 second echo and echo sampling in 5 seconds are respectively shown in Fig. 7 c, 7d, 7e, 7f, wherein signal 1 is an echo signal, signal 3 is an input signal, and signal 2 is an output signal, and output signal 2 all has obvious distortion phenomenon in four figure as can be seen.

The actual result of this test phase output is different with expected results, illustrates that then the robustness ability of this echo canceller code has problem.Six, integration test

This test phase is that echo canceller is carried out integrative test, promptly gather near-end and far-end speech signal, and far-end speech is processed into certain echo signal, near-end and remote signaling are carried out certain combination (also having determined the combination of echo signal according to this combination), import as the near-end and the far-end of this echo canceller respectively, SIMULATE calls the echo cancelltion code by the analog simulation software platform, catch corresponding echo filter coefficient and output signal, and revert to curve, judge whether this curve meets the situation of expection.

When four kinds of different input conditions alternately existed continuously, judgement echo canceller code was specifically worked by carrying out as following table 7:

Table 7

Step	Expected results	Experimental result
Step	Expected results	Experimental result	1. simulate the echo channel of 8 milliseconds of time-delays	Do not have	Do not have
2. the input of far-end and near-end exists situation shown in Fig. 8 a.	The convergence situation of filter coefficient is shown in Fig. 8 a.	As Fig. 8 b, Fig. 8 c, Fig. 8 d		Do not have	Do not have

When far-end and near-end adopt shown in Fig. 8 a mode, the convergence situation of filter coefficient is difference along with the difference of counting of test, Fig. 8 b, 8c, 8d represent the coefficient of iteration 30000 points, 40000 points, 60000 postfilters respectively, can find under the situation that iterations increases, coefficient is not restrained on the contrary, and this is different from expected results.

The actual result of this test phase output is different with expected results, illustrates that then the control section of this echo canceller code has problem.

More than be the test that the echo canceller code of realizing based on the DSP digit chip is carried out, because this kind test is the test that is in development, therefore, wherein the test purpose in each test phase can be different.But the method for testing of being used in each stage realizes according to the present invention fully.For in each test phase, after using method of testing described in the invention, modification and testing authentication once more to the tested object that do not reach expected results just repeat no more herein.

In addition, can in the scope that the present invention discloses, revise about some parameters or the data of the discussion of representative instance above being to be understood that.For example, count and to make amendment according to the requirement (as the time) of test with the iteration that analog simulation software platform SIMULATE carries out the analog simulation test by the present invention, equally also can be according to the difference of the echo channel of supposing, change the time-delay number of the echo data that customizes, can certainly customize different voice signals.The difference that can provide according to dsp chip manufacturer as for the analog simulation software platform SIMULATE of DSP and difference.

In addition, even do not adopt the hardware platform of DSP digit chip, as long as this chip that adopts has corresponding software emulation platform also can be undertaken by above invention fully as echo canceller; If the echo cancelltion code is based on certain chip, then can directly carries out software emulation on computers and test and get final product by method provided by the invention.

Description to preferred embodiments makes those of ordinary skill (especially for the tester) in the art can use the present invention above.It is conspicuous to the technical staff in this technical field that this example is made various modifications, and determined here total principle can be used for other examples and do not need creative ability.Therefore, the instantiation shown in the present invention does not plan to be defined in here, but will give with the principle and the corresponding to maximum magnitude of novel feature that are disclosed here.

Claims

1. A test method for echo cancellation function, comprising the following steps:

Step 1: Collect signals generated in actual situations or customized according to various rules;

Step 2: Process these signals according to various echo channels assumed in each test stage to make them into corresponding echo signals;

Step 3: Input the original signal and the corresponding echo signal into the echo canceller as far-end and near-end signals;

Step 4: Capture the corresponding output signal and the coefficients of the echo cancellation filter;

Step 5: restore the output signal and filter coefficients to a curve;

Step 6: Compare with the original remote signal input curve and the simulated echo channel coefficient curve according to the test requirements, and judge the correctness and performance of the echo canceller in various actual situations and various rules. no.

2. The test method for echo cancellation function according to claim 1, characterized in that:

The first step uses the acquisition method of sound data, which is used to customize various actual test cases, and converts it into a corresponding data format as the near-end or far-end input of the echo canceller in the third step .

The second step is to delay the collected data correspondingly according to the delay number of the echo channel.

The third step is to use the delayed echo signal and the original signal as the near-end and far-end input echo cancellers, and obtain test data through software simulation and call the echo cancellation code.

The fourth step is to automatically store the test data results into corresponding files by the simulation platform;

The fifth step is to restore the test data obtained by the simulation platform or the data collected by customization into a corresponding curve form for comparison and judgment.

3. The test method for echo cancellation function according to claim 1 or 2, characterized in that:

Each test stage in the second step includes the correct implementation test of the echo cancellation code, the basic function test, the efficiency test of the echo cancellation, the working condition test of the echo cancellation code at the boundary point, the robustness test of the echo cancellation code, and the echo cancellation code. Offset code synthesis testing.

4. The test method for echo cancellation function according to claim 3, characterized in that:

The correct implementation test of the echo cancellation code is to test whether the program runs correctly by sampling and PCM encoding the sinusoidal signal and using various combinations as input.

5. The test method for echo cancellation function according to claim 3, characterized in that:

The basic function test of the echo cancellation is to observe whether the output meets the expected requirement by covering all possible input conditions and considering the existence of the secondary echo in the case of simulation.

6. The test method for echo cancellation function according to claim 3, characterized in that:

The efficiency test of the echo cancellation is to construct an echo channel of a certain order, and only add the far-end voice signal, after running with different iterations, output the coefficients of the filters respectively, and obtain the relationship between the coefficient convergence and the iterations , observe how many iterations after which the coefficients of the filter converge to simulate the echo channel.

7. The test method for echo cancellation function according to claim 3, characterized in that:

The processing capability test of the echo canceling code in boundary conditions is to test the working condition of the echo canceling code when there is no echo, the echo delay number is 0, and the echo delay number reaches the echo cancellation capability limit of the design of the echo canceller.

8. The test method for echo cancellation function according to claim 3, characterized in that:

The robustness test of the echo cancellation code is to test the working condition when the input voice is subjected to noise, howling sound or DC interference and when the echo delay exceeds the design capability.

9. The test method for echo cancellation function according to claim 3, characterized in that:

The comprehensive test of the echo canceling code is to construct a variety of different input situations according to the echo canceller and make them exist continuously and alternately, so as to test whether the program runs correctly.