CA2231605C - An enhanced echo canceller for digital cellular application - Google Patents
An enhanced echo canceller for digital cellular application Download PDFInfo
- Publication number
- CA2231605C CA2231605C CA002231605A CA2231605A CA2231605C CA 2231605 C CA2231605 C CA 2231605C CA 002231605 A CA002231605 A CA 002231605A CA 2231605 A CA2231605 A CA 2231605A CA 2231605 C CA2231605 C CA 2231605C
- Authority
- CA
- Canada
- Prior art keywords
- echo
- signal
- coefficients
- estimate
- return channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B3/00—Line transmission systems
- H04B3/02—Details
- H04B3/20—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other
- H04B3/23—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other using a replica of transmitted signal in the time domain, e.g. echo cancellers
Abstract
An echo cancelling system for communication systems. The system includes an adaptive filter (34) that processes the speech signal re-constructed by a speech decoder (30) to generate an echo estimate. The echo estimate is subtracted from the echo containing signal propagating in the communication link to remove or at least reduce the echo corruption. The tap coefficients of the adaptive filter are established on the basis of an error signal that is representative of difference between the echo estimate and the actual echo and on the basis of a signal element that is a component of the re-constructed speech signal. The signal element is highly decorrelated using a decorrelation filter (38) to enable the adaptive filter to converge faster and manifest an improved echo return loss enhancement.
Description
TIT~E: ECHO C7iNCELLING SYSTEM FOR DIGITAL TELEPHONY
APPLIC~TIONS
F~ELD OF TEE INV~1~N
The present invention relates to a method and a system ~or inhibiting echo in a c~ll~ication line The system is particularly useful for communication systems that convey voice coded signals consisting of speech model parameters and excitation information In a most preferred embodiment of the invention the excitation information is extracted from the re-constructed voice-coded signal and used to train an adaptive filter to effect echo inhibition. The excitation signal is ~requency rich and allows to increase the convergence rate and the echo return loss enhancement of the adaptive filter.
Ra. .~r;~ ~ OF TEE 1NV~n 1~
In a typical telephone network a hy~rid converter is provided to connect the unidirectional four wire link from the public switched telephone networX (PSTN) to the local two wire loop. The basic function of the hybrid converter is to separate the transmitted signal originating in the local loop from the received signal in the PSTN section, CA 0223160~ 1998-03-10 W O9~06185 PCT/CA96100771 and vice versa. This process requires the energy o~ the received signal to pass fulIy in the local loop. However, due to an impedance mismatch in the hybrid converter, part of the received energy is ref~ected back to the transmitting port. As a result, a talker hears his own delayed speech which, of course, is undesirable. This kind of echo generated by interactions occurring in the communication network is called an electrical echo.
Another phenomenon that generates echo is from a hands-~ree telephone termi~l. The speech signal generated from the speaker in the t~rmi n~ 1 will propagate in the ~orm of an acoustic wave through an acoustic environment and part o~ it will be captured back by the microphone of the t~rmi n~l . This residue signal will be transmitted back to the talker and creates echo. This kind of echo is referred to as acoustic echo.
To avoid echo problem suppressors are used in the communication networks. A typical echo suppressor is a switch that monitors the voice signals travelling in both directions. The suppressor detects which person is talking and blocks the signal travelling in the opposite direction.
CA 0223160~ 1998-03-10 In essence, the echo suppressor transforms the communication link into an unidirectional path allowing only one person to spea~ at the time. The drawback of such echo suppressors is that they tend to "chop" speech signals when the subscribers talk back and forth quickly. This is because the suppressor is not able to switch direction fast enough. Moreover, during double talk, i.e., when the subscribers talk simultaneously, the suppressor fails to control the echo.
One possibility to avoid the problems of echo suppressors is to provide circuitry that instead of blocking speech signals in one direction in the communication link cancels the echo by using an adaptive filter. In essence, the echo canceller synthesizes the echo which is then subtracted from the composite signal (speech signal+echo signal). If the echo canceller models the true echo path well, the resultant signal is substantially free from any echo corruption.
To make the echo cancellation concept useful, the distinctive characteristics of the echo signal must be measured and stored in the echo canceller for operation.
CA 0223160~ 1998-03-lO
W O 98/06185 PCT/C~96/00771 A direct measure beforehand is not practical because first of all, the echo is dependent on the clrcuit connections in the electrical echo case and on the acoustic environment in the acoustic echo case. This makes the measurement extremely difficult if not impossible. Secondly, the echo may be varying due to dynamic link connections or an unstable acoustic environment. As mentioned be~ore one way to overcome this di~ficulty is to use an adaptive filter to gradually identify the characteristics of the echo in order to perform canc~Llation. Although an adaptive echo canceller can model the echo signal without requiring any prior knowledge of its characteristics, it has a finite learning time to reach a final solution. The convergence speed is a measure of how fast an adaptive filter reaches an acceptable error level. As a conse~uence, the echo will be present at the beginning of a call because at that m~m~pt the adaptive filter is beginning the learning process and the magnitude of the error signal is most likely not to be optimum yet. In addition, the user will hear a short period of electrical echo when the call is switched to a new link. In the case of an acoustic echo the echo characteristics are changing from time to time.
Effective echo cancellation in these situations requires CA 0223160~ 1998-03-10 a fast convergence rate.
The convergence speed of a typical adaptive filter is such that effective echo cancellation will occur a few seconds after the learning process has been initiated.
Therefore, at the beginning of a call there is a period of time during which the subscriber will hear at least some echo. This problem is particularly annoying in cellular telephony applications where the subscriber moves from one reception cell to the other. At the entry in a given reception cell the learning process of the echo canceller must be repeated which involves the short but objectionable period of echo presence.
The generation of a linear echo can be modelled as follows:
Y ( k) =~, h~x ~ k-i ) +~ ( k ) i =o Where Y(k) is the kth output sample of the echo signal are the parameters that characterize the echo path W 098/06185 ~CT/CA9.'~C7/1 ~ (k) is random noise or modelling error x (k) is the kth input sample of the source signal N-l is the order of the model Echo cancellation can be effected by using an adaptive filter which estimates Yfk~ by the filter output Afk) to nullify the speech signal that constitutes the echo. The output of the adaptive filter is expressed by the following equation:
N-l i~0 i The echo tends towards nullity when the error e (k) N-l N-l 3 (k) i~hix(k~ , (hP-hi)x(k-i) +~(k) lS between the two quantities is m;n;m; zed:
The filter coe~ficients are adjusted to mintmi ze the output mean-squared error (MSE) E [e2~k) ]:
E~e2 (k) ] =Et (~ (hl-hi) x(k-i) ) 2] +E[C~)2(k) ~
i =O
-CA 0223160~ 1998-03-10 After some learning period the adaptive filter will converge to hl =h OsisN-1 In order to reach the solution, one has to find a procedure to guide the adaptive filter to reach the minim-l~
error. This is referred as adaptive algorithm. There are two well known adaptation algorithms: the least-mean-square and the recursive least squares. The former is popular for its simplicity of implementation, while the latter has a much faster learning speed. These algorithms are well known in the dlgital telephony art and there is no need to describe them in detail.
The least-mean-square algorithm converges fastest when the input is white. One can therefore enhance the convergence speed by pre-whitening the input to the echo canceller. This is accomplished by placing a whitening filter before the canceller. However, a fixed whitening CA 0223l60~ l998-03-l0 W O 98/06185 PCT/CA3~,'~ //l filter can improve convergence speed only in the statistical sense. ~hat is, it enhances the speed when the input matches the correlation function used in designing the filter. A mismatch, on the other hand, will slowdown the convergence rate. A better approach is to make the whitening filter changeable by using a certain adaptive algorithm. Such an adaptive whitening filter, however, is ineffective, difficult and expensive to implement and for that reason it does not represent an optimal solution.
~ v~S AND sUMMaRY OF T~E lNV~llVN
In view of the foregoing problems encountered in the ~nown prior art, it is an object of the present invention is to provide an improved echo canceller system that is capable of quick adaptation to the particular echo path during a given call.
Another object of the invention is an improved method for effecting echo cancellation that allows to reduce the error between the echo estimate and the true echo more rapidly by comparison to prior art methods.
CA 0223160~ 1998-03-10 Another object of the invention is to provide a communication system using the aforementioned echo canceller.
A further object of the invention is to provide a method for reducing the convergence rate of an echo canceller system.
Yet, another object of the invention is to provide a novel adaptive filtering system having a high convergence rate, particularly well suited for use in echo cancellation systems.
It should be noted that for the purposes of this specification the expression "echo canceller" should be interpreted in a broad sense to designate systems that substantially eliminate the echo or at least partially reduce the echo magnitude. Thus, "echo canceller" is not intended to ex~luslvely designate a system that totally eliminates the echo.
APPLIC~TIONS
F~ELD OF TEE INV~1~N
The present invention relates to a method and a system ~or inhibiting echo in a c~ll~ication line The system is particularly useful for communication systems that convey voice coded signals consisting of speech model parameters and excitation information In a most preferred embodiment of the invention the excitation information is extracted from the re-constructed voice-coded signal and used to train an adaptive filter to effect echo inhibition. The excitation signal is ~requency rich and allows to increase the convergence rate and the echo return loss enhancement of the adaptive filter.
Ra. .~r;~ ~ OF TEE 1NV~n 1~
In a typical telephone network a hy~rid converter is provided to connect the unidirectional four wire link from the public switched telephone networX (PSTN) to the local two wire loop. The basic function of the hybrid converter is to separate the transmitted signal originating in the local loop from the received signal in the PSTN section, CA 0223160~ 1998-03-10 W O9~06185 PCT/CA96100771 and vice versa. This process requires the energy o~ the received signal to pass fulIy in the local loop. However, due to an impedance mismatch in the hybrid converter, part of the received energy is ref~ected back to the transmitting port. As a result, a talker hears his own delayed speech which, of course, is undesirable. This kind of echo generated by interactions occurring in the communication network is called an electrical echo.
Another phenomenon that generates echo is from a hands-~ree telephone termi~l. The speech signal generated from the speaker in the t~rmi n~ 1 will propagate in the ~orm of an acoustic wave through an acoustic environment and part o~ it will be captured back by the microphone of the t~rmi n~l . This residue signal will be transmitted back to the talker and creates echo. This kind of echo is referred to as acoustic echo.
To avoid echo problem suppressors are used in the communication networks. A typical echo suppressor is a switch that monitors the voice signals travelling in both directions. The suppressor detects which person is talking and blocks the signal travelling in the opposite direction.
CA 0223160~ 1998-03-10 In essence, the echo suppressor transforms the communication link into an unidirectional path allowing only one person to spea~ at the time. The drawback of such echo suppressors is that they tend to "chop" speech signals when the subscribers talk back and forth quickly. This is because the suppressor is not able to switch direction fast enough. Moreover, during double talk, i.e., when the subscribers talk simultaneously, the suppressor fails to control the echo.
One possibility to avoid the problems of echo suppressors is to provide circuitry that instead of blocking speech signals in one direction in the communication link cancels the echo by using an adaptive filter. In essence, the echo canceller synthesizes the echo which is then subtracted from the composite signal (speech signal+echo signal). If the echo canceller models the true echo path well, the resultant signal is substantially free from any echo corruption.
To make the echo cancellation concept useful, the distinctive characteristics of the echo signal must be measured and stored in the echo canceller for operation.
CA 0223160~ 1998-03-lO
W O 98/06185 PCT/C~96/00771 A direct measure beforehand is not practical because first of all, the echo is dependent on the clrcuit connections in the electrical echo case and on the acoustic environment in the acoustic echo case. This makes the measurement extremely difficult if not impossible. Secondly, the echo may be varying due to dynamic link connections or an unstable acoustic environment. As mentioned be~ore one way to overcome this di~ficulty is to use an adaptive filter to gradually identify the characteristics of the echo in order to perform canc~Llation. Although an adaptive echo canceller can model the echo signal without requiring any prior knowledge of its characteristics, it has a finite learning time to reach a final solution. The convergence speed is a measure of how fast an adaptive filter reaches an acceptable error level. As a conse~uence, the echo will be present at the beginning of a call because at that m~m~pt the adaptive filter is beginning the learning process and the magnitude of the error signal is most likely not to be optimum yet. In addition, the user will hear a short period of electrical echo when the call is switched to a new link. In the case of an acoustic echo the echo characteristics are changing from time to time.
Effective echo cancellation in these situations requires CA 0223160~ 1998-03-10 a fast convergence rate.
The convergence speed of a typical adaptive filter is such that effective echo cancellation will occur a few seconds after the learning process has been initiated.
Therefore, at the beginning of a call there is a period of time during which the subscriber will hear at least some echo. This problem is particularly annoying in cellular telephony applications where the subscriber moves from one reception cell to the other. At the entry in a given reception cell the learning process of the echo canceller must be repeated which involves the short but objectionable period of echo presence.
The generation of a linear echo can be modelled as follows:
Y ( k) =~, h~x ~ k-i ) +~ ( k ) i =o Where Y(k) is the kth output sample of the echo signal are the parameters that characterize the echo path W 098/06185 ~CT/CA9.'~C7/1 ~ (k) is random noise or modelling error x (k) is the kth input sample of the source signal N-l is the order of the model Echo cancellation can be effected by using an adaptive filter which estimates Yfk~ by the filter output Afk) to nullify the speech signal that constitutes the echo. The output of the adaptive filter is expressed by the following equation:
N-l i~0 i The echo tends towards nullity when the error e (k) N-l N-l 3 (k) i~hix(k~ , (hP-hi)x(k-i) +~(k) lS between the two quantities is m;n;m; zed:
The filter coe~ficients are adjusted to mintmi ze the output mean-squared error (MSE) E [e2~k) ]:
E~e2 (k) ] =Et (~ (hl-hi) x(k-i) ) 2] +E[C~)2(k) ~
i =O
-CA 0223160~ 1998-03-10 After some learning period the adaptive filter will converge to hl =h OsisN-1 In order to reach the solution, one has to find a procedure to guide the adaptive filter to reach the minim-l~
error. This is referred as adaptive algorithm. There are two well known adaptation algorithms: the least-mean-square and the recursive least squares. The former is popular for its simplicity of implementation, while the latter has a much faster learning speed. These algorithms are well known in the dlgital telephony art and there is no need to describe them in detail.
The least-mean-square algorithm converges fastest when the input is white. One can therefore enhance the convergence speed by pre-whitening the input to the echo canceller. This is accomplished by placing a whitening filter before the canceller. However, a fixed whitening CA 0223l60~ l998-03-l0 W O 98/06185 PCT/CA3~,'~ //l filter can improve convergence speed only in the statistical sense. ~hat is, it enhances the speed when the input matches the correlation function used in designing the filter. A mismatch, on the other hand, will slowdown the convergence rate. A better approach is to make the whitening filter changeable by using a certain adaptive algorithm. Such an adaptive whitening filter, however, is ineffective, difficult and expensive to implement and for that reason it does not represent an optimal solution.
~ v~S AND sUMMaRY OF T~E lNV~llVN
In view of the foregoing problems encountered in the ~nown prior art, it is an object of the present invention is to provide an improved echo canceller system that is capable of quick adaptation to the particular echo path during a given call.
Another object of the invention is an improved method for effecting echo cancellation that allows to reduce the error between the echo estimate and the true echo more rapidly by comparison to prior art methods.
CA 0223160~ 1998-03-10 Another object of the invention is to provide a communication system using the aforementioned echo canceller.
A further object of the invention is to provide a method for reducing the convergence rate of an echo canceller system.
Yet, another object of the invention is to provide a novel adaptive filtering system having a high convergence rate, particularly well suited for use in echo cancellation systems.
It should be noted that for the purposes of this specification the expression "echo canceller" should be interpreted in a broad sense to designate systems that substantially eliminate the echo or at least partially reduce the echo magnitude. Thus, "echo canceller" is not intended to ex~luslvely designate a system that totally eliminates the echo.
2~
As embodied and bro-adly described herein the invention provides an echo canceller for reducing a magnitude of an CA 0223160~ 1998-03-10 W O9~06185 PCT/CA96/00771 echo occurring in a return channel of a c~mmnnication system, said echo canceller including:
- first means for generating an echo estimate, said first means including a first input for receiving a first signal that is a precursor of the echo in the return channel of the comm--ni cation system and a second input for receiving a second signal related to said first signal and ~eing more decorrelated than said first signal; and - second means in operative relationship with said first means, said second means being responsive to said echo estimate for conditioning a signal propagating on the return channel of the c~mmlln-cation system to inhibit echo therein.
The present inventor has made the unexpected discovery that the echo canceller can be caused to converge faster by utilizing in the adaptation process a component of the source signal (the signal from which the echo estimate is generated) that is more decorrelated than the source signal. This additional signal element allows to reduce the learning period of the echo canceller very significantly.
CA 0223160~ 1998-03-lO
W 098/0618~ PCT/CA96/00771 In a most preferred embodiment of the invention the echo canceller includes an adaptive filter provided with a decorrelation unit whose function is to whiten the re-constructed speech input generated by the voice decoder.
The decorrelated data along with the error signal e(k) are used to change the adaptive filter characteristics in order to keep the error signal at minimum. The decorrelation ~ilter employs the same parameters used by the low bit-rate speech decoder so no additional processing is required to tune the decorrelation filter. Typically, the information extracted by the decorrelation filter is closely related to excitation signal contained in each frame of the coded speech signal. One characteristic o~ the excitation signal is that it is highly decorrelated. The decorrelated nature of the excitation signal allows the adaptive filter to converge faster toward the minimllm error by comparison to prior art systems where the input of the adaptive filter is only a correlated speech signal.
For the purposes of the present specification the expression "decorre~ated" and "correlated" will be defined with relation to one another. A "decorrelated" signal will designate a signal that is characterized by a more uniform CA 0223160~ 1998-03-lO
energy distribution across the frequency range of interest than a "correlated" signal. Thus, for a given frequency range divided in a predetermined number of bands the energy content of a decorreLated signal will typically be distributed more uniformly in the frequency range and among a larger number of bands than in the case of a correlated signal. The correlated nature of a given signal can also be expressed in terms of spectral density or spectral energy distribution which is defined as the power carried by the signal within some interval o~ frequency. A
correlated signal will exhibit a substantially less uniform spectral density over the same frequency range than a decorrelated signal. As an example, a totally random or unpredictable signal in the time domain produces a perfectly decorrelated signal that has a frequency spectrum which is flat across the frequency range of interest. On the other hand, a speech waveform exhibits a much higher level of correlation.
~ he definition above as to what constitutes a correlated signal and a decorrelated signal applies to long-term measurements or observations, not to short term occurrences. Indeed, a totally random signal may over a CA 0223160~ 1998-03-lO
W O 98/06185 PCTlCA9''~C7/l short period of time appear highly correlated (the energy is concentrated in a narrow frequency band) while over a longer term period the signal is decorrelated because the energy load "walks" across the entire frequency range of interest. In contrast, in a correlated signal most of the energy will systematically be found to occur in a narrow frequency band within the frequency range. For example, in speech coding applications "long term" could refer to a period of time in the order of 20 - ~0 milliseconds.
In a preferred embodiment of the invention summarized above, the decorrelation filter is updated at the same time as the voice decoder which, typically occurs at each frame of the coded speech signal. The updating procedure of the decorrelation filter simply consist of copying the coefficient of the voice decoder.
In a most preferred embodiment the echo canceller uses a novel least-mean-square type algorithm to update the parameters of the adaptive filter in order to maintain the ~ 25 error signal low. This algorithm uses the three following inputs:
CA 0223160~ 1998-03-lO
A) The error signal e(k);
B) A power estimate signal that is the exponential running average of the output of the decorrelation filter;
C) A number of sub parameters established on the basis of the output of the decorrelated signal.
The adaptive algorithm processes those signals to calculate a set of coefficients that are communicated to the adaptive filter of the echo canceller.
As embodied and broadly described herein the invention further provides a commllnication system including:
- voice decoder for receiving coefficients data and excitation information to re-construct a speech signal;
- a return channel in which an echo is susceptible to occur;
- an echo canceller for reducing a m~-agnitude of the echo in said return channel,-said echo canceller comprising;
CA 0223160~ 1998-03-10 a) first means for generating an estimate of the echo in said return channel, including I) a ~irst input in operative relationship with said voice decoder for receiving said re-constructed speech signal;
II) a second input for receiving a component of said re-constructed speech signal, said component being more decorrelated than said re-constructed speech signal;
III) a third input for receiving an error signal which is representative of a divergence between said echo estimate a real echo propagating on said return channel;
IV) means for processing the signals at said first, second and third inputs and for outputting an echo estimate signal;
b) second means in operative relationship with said first means, said second means being responsive to said echo estimate for conditioning a signal propagating on said return channel to inhibit echo therein;
CA 0223160~ 1998-03-10 c) a divergence detector means for calculating said error signal between said echo estimate and an actual echo propagating on said return channel of the c~m~l~nication system, said divergence detector being operatively associated with said third input to supply thereto said error signal.
As embodied and broadly described herein the invention also provides a combination that includes:
- a voice decoder for generating a re-constructed speech signal;
- means ~or generating an estimate of an echo of said re-constructed speech signal, comprising:
- an adaptive filter in operative relationship with said voice decoder to condition said re-constructed speech signal and in turn generate said estimate of an echo, said adaptive filter having a transfer function determined by a plurality of coefficients:
- processing means including:
a) a first input for receiving a signal derivable from said re-CA 0223160~ 1998-03-10 W O98/06185 PCTICAg6/00771 constructed speech signal and being more decorrelated than said re-constructed speech signal;
b) a second input for receiving a error signal indicative of a difference between said estimate of an echo and a desired output, wherein said processing means determines at least partially from the signal at said first input and said error signal said plurality of coefficients.
As embodied and broadly de'scribed hereln the invention yet provides a method of increasing a convergence rate of an echo canceller that reduces the magnitude of an echo occurring in a return channel of a communication system, said echo canceller including:
- first means for generating an echo estimate, said first means including an input for receiving a first signal that is a precursor of the echo in the return channel of the comm-1nication system;
- second means in~operative relationship with said first means, said second means being responsive to CA 0223160~ 1998-03-lO
said echo estimate for conditioning a signal propagating on the return channel of the communication system to inhibit echo therein, said method comprising the steps of:
- providing to said first means a second signal derivable ~rom said first signal, said second signal being more decorrelated than said first signal, said first and second signals influencing said first means in the generation of said echo estimate, the presence of said second signal allowir.g said echo canceller to converge at a higher rate.
As embodied and broadly described herein the invention also provides a method for reducing a magnitude of an echo occurring in a return channel of a communication system, said method including the steps of:
a) generating an echo estimate, comprising the steps of:
- providing a first signal that is a precursor of the echo in the return channel of the commllnication system;
- processing said first signal to obtain CA 0223160~ 1998-03-10 therefrom a second signal that is more decorrelated than said first signal;
- providing an error signal that is representative o~ a divergence between said echo estimate and an actual echo propagating on the return channel of the communication system;
- processing said first, second and third signals to produce said echo estimate;
b) subtracting from a signal propagating through said return channel said echo estimate.
As embodied and broadly described herein the invention also provides an adaptive filtering system having a high convergence rate, said adaptive filtering system comprising:
- a filtering stage including a predetermined number of tap coefficients that determine the transfer function of th~e filter, said filtering stage including an input for receiving a first signal to be conditioned by said filtering stage;
- 25 - a processing stage to compute said tap coefficients, said processing stage including:
- an input for receiving a second signal that is a CA 0223160~ 1998-03-10 W O 98/0618~ PCT/CA96/00771 2~
component o~ said first signal, said second signal being more decorrelated than said first signal; and - an input for receiving an error signal representative of a difference between an actual output of said filtering stage and a desired output, whereby said processing stage determines at least partially on the basis ~ of said second signal and said error signal said tap coefficients.
B~IEF l~ C~TPT:~:ON OF THE DRAWINGS
Figure 1 is a block diagram of communication link utilizing a hybrid between a four wire section and a two wire section of the llnk;
Figure 2 is a diagram of a typical speech waveform;
Figure 3 is a block diagram of a voice coder circuit;
Figure 4 is a block diagram of a voice decoder for speech regeneration;
Figure 5 is a block d1agram of an echo canceller circuit showing the relationship with the voice decoder depicted in Figure 4;
Figure 6 is a block diagram for re-generating excitation from the decoded speech signal for use in the adaptive filter of the echo canceller, Figure 7 is more detailed block diagram of the echo CA 0223160~ 1998-03-10 W O 98106~85 PCTICA~5~-7/1 canceller circuit;
Figure 8 is a block diagram illustrating the trans~er function of the algorithm operating the adaptive filter of the echo canceller depicted in Figure 7;
Figure 9 is a block diagram illustrating the transfer function o~ the power estimator block shown in the diagram at Figure 7; and Figure lO is block diagram illustrating the transfer function o~ the block generating the coefficients pj(k) shown in Figure 7.
DES~L- l~N OF A ~K~K~V EMBODDMEN~
The present invention provides an improved echo canceller system that is particularly well suited for use in ~mmllnications networks where an echo is susceptible to occur either due to electrical interactions inside the c~mmlln;cation link (electrical echo) or as a result of an acoustic feedback taking place at one end of the comm'ln;cation link (acoustical echo). For simplicity the example of the invention described below will be made with reference to a cellular telephone network where echo occurs as a result of an impedance mismatch. It should be understood, however, that the scope of the invention should CA 0223160~ 1998-03-lO
not be limited to this specific c~mmllnication network environment nor to the particular type of echo encountered (electrical echo).
A typical wireless communication network is depicted by the block diagram in Figure 1. The network includes a wireless telephone 10 that can exchange coded speech signals with a land based station 20. ~he station includes a transmission channel 22 and a reception channel 24 connected to a hybrid 26 that constitutes an interface between the four wire reception/transmission channels ~ section and the two wire local loop.
During a typical call the speech waveform of the type shown in the graph at ~1gure 2, is digitized and the consecutive bytes are grouped in successive frames that are coded and transmitted individually to the reception port 24. For the purpose of the example let's consider a single frame of the digitized speech waveform. The digitized data is processed by a voice coder shown in Figure 3 that is located in the telephone 10, which extracts perceptually significant features of speech from the time waveform. In essence, the voice coder 28 analyzes the waveform to CA 0223l60~ l998-03-l0 W O 98/06185 PCTICA96/OOMl produce a time-varying model of the vocal tract excitation and transfer function. During a given frame, however, the vocal tract is assumed to represent a linear time-invariant process.
Thus, for each frame of the speech waveform the voice coder 28 generates an excitatlon signal and a set of predictor coefficients that characterize the transfer function of the model. The excitation signal and the set of predictor coefficients are modulated and transmitted towards the reception channel 24. The same process is repeated for each frame of the speech waveform.
At the reception channel 24 the coded signal is demodulated and the digitized speech wave~orm re-constructed by using a decoder 30 whose transfer functionis illustrated in figure 4. In essence, the decoder 30 generates the current speech sample by a linear combination of previous speech samples and the excitation. Post processing, illustrated by the block 31 is then applied to improve speech quality. The output signal x(k)of the decoder 30 which is the reconstructed digitized waveform is expressed by the following equation:
CA 0223l605 l998-03-lO
x (k) =~, cix (lc-i ) +n (k) Where x(k) is the kth output sample n(k) is the excitation signal;
C1,c2, cMare the predictor coefficients for the given frame;
M is the order of the model. Typically, M
has a value of 10.
It should be noted that n(k) is not exactly the same as the excitation signal received by the voice decoder that re-constructs the speech waveform. The difference is due primarily to the presence of the post-processing stage 31 that eliminates some components of the re-constructed speech signal. To distinguish these two quantities n~k) will be referred to in the following description as re-constructed excitation.
In theory all the energy of the reconstructed speech signal x(k) should be transmitted entirely to the local loop through the hybrid. However, the hybrid will reflect a small fraction of the incoming signal energy due to CA 0223160~ 1998-03-10 PC~/CA96/00771 impedance mismatch problems. As a result, the reflected signal is returned to the source as an echo through the transmitter port. The subjective effects of the echo to the human ear depends primarily upon the round-trip delay (echo with respect to original signal). When the delay exceeds approximately 40 milliseconds the echo signals become clearly perceptible and annoying during a conversation.
This threshold is exceeded in most telecommunications systems which necessitates the use of an echo canceller.
The improved echo canceller 32 constructed in accordance with the present invention is depicted in figure 5. The echo canceller 32 is connected to the speech decoder 30 and receives the reconstructed speech signal x(k). On the basis of this input the echo canceller 32 synthesizes an echo signal that is added with the composite signal Z(k) that is constituted by the speech signal originating from the local loop and the echo. In the situation where the subscriber at the local loop is not talking then Z(k) is only the echo signal. If the echo canceller models the true echo path with accuracy, the resultant signal is echo free.
CA 0223160~ 1998-03-10 The echo canceller includes an adaptive filter that alters the input signal x(k~ to bring it as close as possible to the echo. A feedback loop 36 is provided allowing the filter 34 to alter its transfer function and converge to nullify or at least reduce the error signal e(k~. A novel aspect of the present invention resides in the provision of a decorrelation filter 38 that supplies a frequency rich input to the adaptive filter 34 which enables that fllter to converge faster. The decorrelation filter 38 uses the predictor coefficients from the speech decoder 30 to process the speech signal x(k) to extract there~rom the excitation information n(k). The excitation signal n(k) is much more decorrelated than the reconstructed speech signal and this allows the adaptive ~ilter 34 to reach a small error condition much ~aster.
The block diagram of the decorrelation filter 38 is shown at figure 6. It will be noted that the decorrelation filter is set as the inverse of the speech decoder. The output of the decorrelation filter is expressed by the following e~uation:
n (k) =x (k) -~ cix (k-i ) CA 0223160~ 1998-03-10 W O 98/06185 PCT/CA9~ 7/1 Where cl,cq, cM are the predictor coefficients for the given frame The predictor coefficients are updated at the same time as those of the speech decoder. Since the speech decoder 30 is constant within a frame, the decorrelation filter 38 needs to be modified only after each frame and then it is kept unchanged. The updating procedure of the decorrelation filter 38 is effected simply by copying the predictor coefficients from the speech decoder 30 by any convenient means.
Typically, the order of the decorrelation filter 38 would be 10 (order of the filter or value M) which is typical of most commercially available speech decoders.
Figure 7 is a more detailed block diagram of the echo canceller 32. In addition to the decorrelation filter block 38 and the adaptive filter 34 shown in figure 5 the more complete diagram of figure 7 also indicates that the echo canceller includes two additional function blocks namely a generator o~ coefficients 40 and a power estimator 42 WO 98106185 PCT/CA9~ C7~1 that process individually the re-constructed excitation signal n(k) to supply additional parameters to a novel adaptive algorithm that calculates the coefficients of the adaptive filter. Note that for the purpose of simplicity, the adaptive filter 34 has been shown at figure 7 as two separate blocks namely the adaptive algorithm 44 and the filter core 46 which effect the actual signal processing.
In the most preferred embodiment of the present invention the following modified least mean square algorithm has been implemented:
ho (k+l ) =ho (k) +~ Z ( ) Po (k) hl (k+l ) =hl (k) +1l ~Z (k) Pl (k) hN(k+l ) =h(N l, (k) +u ~ (k) PN-1 (k) ~here hi is the ith coefficient of the filter CA 0223160~ 1998-03-lO
W O 98/0618~ PCT/C~96/00771 core 46;
e(k) is the error signal;
Pi(k) is a ith coefficient (see below);
,u is a positive constant; and ~~2n (k) is the power estimate of the re-constructed excitation signal n(k) The block diagram of the circuit for calculating the coe~ficients of the filter core 46 is shown in figure 8.
The error e(k) is divided by the power estimate and multiplied by the coe~ficient Pi(k). The resultant N values are scaled by a positive constant u and added to the coefficients for adaptation. Once the parameters have been established they are transferred to the filter core 46.
Preferably the positive constant ,u should be less than l/3 N and most preferably it should be set to a value in the range 1/4 N to 1/5 N.
The power estimate is the exponential running average of n2(k). It can be established by using the following equation. The transfer function is graphically illustrated by the block diagram in figure 9.
~ n (k) = (1---N) &Z~ (k-1~ + Nn~ (k) Finally, the coefficients Pi(k) are generated from the excitation signal as follows:
pO (k) =n (k) pi(k)=p~ ~(k-1)-c,(k) n( k) lsisM
Pi (k) =P~i l, (k-l) M+l~;i<N-1 Where: ci is the ith coefficient transferred from the voice decoderi N-1 is the order of the echo canceller;
M is the order of the voice decoder The block diagram for generating the coefficients is shown in figure 10.
Typically, N has a value in the range from 300 to 2000. In the most preferred embodiment of the invention N has a value of 384.
W O 98/06185 . PCT/CA~5.~~7/1 The above description of the invention should not be interpreted in any limiting manner since variations and refinements of the preferred embodiment are possible without departing from the spirit of the invention. The scope of the invention is defined in the appended claims and their equivalents.
As embodied and bro-adly described herein the invention provides an echo canceller for reducing a magnitude of an CA 0223160~ 1998-03-10 W O9~06185 PCT/CA96/00771 echo occurring in a return channel of a c~mmnnication system, said echo canceller including:
- first means for generating an echo estimate, said first means including a first input for receiving a first signal that is a precursor of the echo in the return channel of the comm--ni cation system and a second input for receiving a second signal related to said first signal and ~eing more decorrelated than said first signal; and - second means in operative relationship with said first means, said second means being responsive to said echo estimate for conditioning a signal propagating on the return channel of the c~mmlln-cation system to inhibit echo therein.
The present inventor has made the unexpected discovery that the echo canceller can be caused to converge faster by utilizing in the adaptation process a component of the source signal (the signal from which the echo estimate is generated) that is more decorrelated than the source signal. This additional signal element allows to reduce the learning period of the echo canceller very significantly.
CA 0223160~ 1998-03-lO
W 098/0618~ PCT/CA96/00771 In a most preferred embodiment of the invention the echo canceller includes an adaptive filter provided with a decorrelation unit whose function is to whiten the re-constructed speech input generated by the voice decoder.
The decorrelated data along with the error signal e(k) are used to change the adaptive filter characteristics in order to keep the error signal at minimum. The decorrelation ~ilter employs the same parameters used by the low bit-rate speech decoder so no additional processing is required to tune the decorrelation filter. Typically, the information extracted by the decorrelation filter is closely related to excitation signal contained in each frame of the coded speech signal. One characteristic o~ the excitation signal is that it is highly decorrelated. The decorrelated nature of the excitation signal allows the adaptive filter to converge faster toward the minimllm error by comparison to prior art systems where the input of the adaptive filter is only a correlated speech signal.
For the purposes of the present specification the expression "decorre~ated" and "correlated" will be defined with relation to one another. A "decorrelated" signal will designate a signal that is characterized by a more uniform CA 0223160~ 1998-03-lO
energy distribution across the frequency range of interest than a "correlated" signal. Thus, for a given frequency range divided in a predetermined number of bands the energy content of a decorreLated signal will typically be distributed more uniformly in the frequency range and among a larger number of bands than in the case of a correlated signal. The correlated nature of a given signal can also be expressed in terms of spectral density or spectral energy distribution which is defined as the power carried by the signal within some interval o~ frequency. A
correlated signal will exhibit a substantially less uniform spectral density over the same frequency range than a decorrelated signal. As an example, a totally random or unpredictable signal in the time domain produces a perfectly decorrelated signal that has a frequency spectrum which is flat across the frequency range of interest. On the other hand, a speech waveform exhibits a much higher level of correlation.
~ he definition above as to what constitutes a correlated signal and a decorrelated signal applies to long-term measurements or observations, not to short term occurrences. Indeed, a totally random signal may over a CA 0223160~ 1998-03-lO
W O 98/06185 PCTlCA9''~C7/l short period of time appear highly correlated (the energy is concentrated in a narrow frequency band) while over a longer term period the signal is decorrelated because the energy load "walks" across the entire frequency range of interest. In contrast, in a correlated signal most of the energy will systematically be found to occur in a narrow frequency band within the frequency range. For example, in speech coding applications "long term" could refer to a period of time in the order of 20 - ~0 milliseconds.
In a preferred embodiment of the invention summarized above, the decorrelation filter is updated at the same time as the voice decoder which, typically occurs at each frame of the coded speech signal. The updating procedure of the decorrelation filter simply consist of copying the coefficient of the voice decoder.
In a most preferred embodiment the echo canceller uses a novel least-mean-square type algorithm to update the parameters of the adaptive filter in order to maintain the ~ 25 error signal low. This algorithm uses the three following inputs:
CA 0223160~ 1998-03-lO
A) The error signal e(k);
B) A power estimate signal that is the exponential running average of the output of the decorrelation filter;
C) A number of sub parameters established on the basis of the output of the decorrelated signal.
The adaptive algorithm processes those signals to calculate a set of coefficients that are communicated to the adaptive filter of the echo canceller.
As embodied and broadly described herein the invention further provides a commllnication system including:
- voice decoder for receiving coefficients data and excitation information to re-construct a speech signal;
- a return channel in which an echo is susceptible to occur;
- an echo canceller for reducing a m~-agnitude of the echo in said return channel,-said echo canceller comprising;
CA 0223160~ 1998-03-10 a) first means for generating an estimate of the echo in said return channel, including I) a ~irst input in operative relationship with said voice decoder for receiving said re-constructed speech signal;
II) a second input for receiving a component of said re-constructed speech signal, said component being more decorrelated than said re-constructed speech signal;
III) a third input for receiving an error signal which is representative of a divergence between said echo estimate a real echo propagating on said return channel;
IV) means for processing the signals at said first, second and third inputs and for outputting an echo estimate signal;
b) second means in operative relationship with said first means, said second means being responsive to said echo estimate for conditioning a signal propagating on said return channel to inhibit echo therein;
CA 0223160~ 1998-03-10 c) a divergence detector means for calculating said error signal between said echo estimate and an actual echo propagating on said return channel of the c~m~l~nication system, said divergence detector being operatively associated with said third input to supply thereto said error signal.
As embodied and broadly described herein the invention also provides a combination that includes:
- a voice decoder for generating a re-constructed speech signal;
- means ~or generating an estimate of an echo of said re-constructed speech signal, comprising:
- an adaptive filter in operative relationship with said voice decoder to condition said re-constructed speech signal and in turn generate said estimate of an echo, said adaptive filter having a transfer function determined by a plurality of coefficients:
- processing means including:
a) a first input for receiving a signal derivable from said re-CA 0223160~ 1998-03-10 W O98/06185 PCTICAg6/00771 constructed speech signal and being more decorrelated than said re-constructed speech signal;
b) a second input for receiving a error signal indicative of a difference between said estimate of an echo and a desired output, wherein said processing means determines at least partially from the signal at said first input and said error signal said plurality of coefficients.
As embodied and broadly de'scribed hereln the invention yet provides a method of increasing a convergence rate of an echo canceller that reduces the magnitude of an echo occurring in a return channel of a communication system, said echo canceller including:
- first means for generating an echo estimate, said first means including an input for receiving a first signal that is a precursor of the echo in the return channel of the comm-1nication system;
- second means in~operative relationship with said first means, said second means being responsive to CA 0223160~ 1998-03-lO
said echo estimate for conditioning a signal propagating on the return channel of the communication system to inhibit echo therein, said method comprising the steps of:
- providing to said first means a second signal derivable ~rom said first signal, said second signal being more decorrelated than said first signal, said first and second signals influencing said first means in the generation of said echo estimate, the presence of said second signal allowir.g said echo canceller to converge at a higher rate.
As embodied and broadly described herein the invention also provides a method for reducing a magnitude of an echo occurring in a return channel of a communication system, said method including the steps of:
a) generating an echo estimate, comprising the steps of:
- providing a first signal that is a precursor of the echo in the return channel of the commllnication system;
- processing said first signal to obtain CA 0223160~ 1998-03-10 therefrom a second signal that is more decorrelated than said first signal;
- providing an error signal that is representative o~ a divergence between said echo estimate and an actual echo propagating on the return channel of the communication system;
- processing said first, second and third signals to produce said echo estimate;
b) subtracting from a signal propagating through said return channel said echo estimate.
As embodied and broadly described herein the invention also provides an adaptive filtering system having a high convergence rate, said adaptive filtering system comprising:
- a filtering stage including a predetermined number of tap coefficients that determine the transfer function of th~e filter, said filtering stage including an input for receiving a first signal to be conditioned by said filtering stage;
- 25 - a processing stage to compute said tap coefficients, said processing stage including:
- an input for receiving a second signal that is a CA 0223160~ 1998-03-10 W O 98/0618~ PCT/CA96/00771 2~
component o~ said first signal, said second signal being more decorrelated than said first signal; and - an input for receiving an error signal representative of a difference between an actual output of said filtering stage and a desired output, whereby said processing stage determines at least partially on the basis ~ of said second signal and said error signal said tap coefficients.
B~IEF l~ C~TPT:~:ON OF THE DRAWINGS
Figure 1 is a block diagram of communication link utilizing a hybrid between a four wire section and a two wire section of the llnk;
Figure 2 is a diagram of a typical speech waveform;
Figure 3 is a block diagram of a voice coder circuit;
Figure 4 is a block diagram of a voice decoder for speech regeneration;
Figure 5 is a block d1agram of an echo canceller circuit showing the relationship with the voice decoder depicted in Figure 4;
Figure 6 is a block diagram for re-generating excitation from the decoded speech signal for use in the adaptive filter of the echo canceller, Figure 7 is more detailed block diagram of the echo CA 0223160~ 1998-03-10 W O 98106~85 PCTICA~5~-7/1 canceller circuit;
Figure 8 is a block diagram illustrating the trans~er function of the algorithm operating the adaptive filter of the echo canceller depicted in Figure 7;
Figure 9 is a block diagram illustrating the transfer function o~ the power estimator block shown in the diagram at Figure 7; and Figure lO is block diagram illustrating the transfer function o~ the block generating the coefficients pj(k) shown in Figure 7.
DES~L- l~N OF A ~K~K~V EMBODDMEN~
The present invention provides an improved echo canceller system that is particularly well suited for use in ~mmllnications networks where an echo is susceptible to occur either due to electrical interactions inside the c~mmlln;cation link (electrical echo) or as a result of an acoustic feedback taking place at one end of the comm'ln;cation link (acoustical echo). For simplicity the example of the invention described below will be made with reference to a cellular telephone network where echo occurs as a result of an impedance mismatch. It should be understood, however, that the scope of the invention should CA 0223160~ 1998-03-lO
not be limited to this specific c~mmllnication network environment nor to the particular type of echo encountered (electrical echo).
A typical wireless communication network is depicted by the block diagram in Figure 1. The network includes a wireless telephone 10 that can exchange coded speech signals with a land based station 20. ~he station includes a transmission channel 22 and a reception channel 24 connected to a hybrid 26 that constitutes an interface between the four wire reception/transmission channels ~ section and the two wire local loop.
During a typical call the speech waveform of the type shown in the graph at ~1gure 2, is digitized and the consecutive bytes are grouped in successive frames that are coded and transmitted individually to the reception port 24. For the purpose of the example let's consider a single frame of the digitized speech waveform. The digitized data is processed by a voice coder shown in Figure 3 that is located in the telephone 10, which extracts perceptually significant features of speech from the time waveform. In essence, the voice coder 28 analyzes the waveform to CA 0223l60~ l998-03-l0 W O 98/06185 PCTICA96/OOMl produce a time-varying model of the vocal tract excitation and transfer function. During a given frame, however, the vocal tract is assumed to represent a linear time-invariant process.
Thus, for each frame of the speech waveform the voice coder 28 generates an excitatlon signal and a set of predictor coefficients that characterize the transfer function of the model. The excitation signal and the set of predictor coefficients are modulated and transmitted towards the reception channel 24. The same process is repeated for each frame of the speech waveform.
At the reception channel 24 the coded signal is demodulated and the digitized speech wave~orm re-constructed by using a decoder 30 whose transfer functionis illustrated in figure 4. In essence, the decoder 30 generates the current speech sample by a linear combination of previous speech samples and the excitation. Post processing, illustrated by the block 31 is then applied to improve speech quality. The output signal x(k)of the decoder 30 which is the reconstructed digitized waveform is expressed by the following equation:
CA 0223l605 l998-03-lO
x (k) =~, cix (lc-i ) +n (k) Where x(k) is the kth output sample n(k) is the excitation signal;
C1,c2, cMare the predictor coefficients for the given frame;
M is the order of the model. Typically, M
has a value of 10.
It should be noted that n(k) is not exactly the same as the excitation signal received by the voice decoder that re-constructs the speech waveform. The difference is due primarily to the presence of the post-processing stage 31 that eliminates some components of the re-constructed speech signal. To distinguish these two quantities n~k) will be referred to in the following description as re-constructed excitation.
In theory all the energy of the reconstructed speech signal x(k) should be transmitted entirely to the local loop through the hybrid. However, the hybrid will reflect a small fraction of the incoming signal energy due to CA 0223160~ 1998-03-10 PC~/CA96/00771 impedance mismatch problems. As a result, the reflected signal is returned to the source as an echo through the transmitter port. The subjective effects of the echo to the human ear depends primarily upon the round-trip delay (echo with respect to original signal). When the delay exceeds approximately 40 milliseconds the echo signals become clearly perceptible and annoying during a conversation.
This threshold is exceeded in most telecommunications systems which necessitates the use of an echo canceller.
The improved echo canceller 32 constructed in accordance with the present invention is depicted in figure 5. The echo canceller 32 is connected to the speech decoder 30 and receives the reconstructed speech signal x(k). On the basis of this input the echo canceller 32 synthesizes an echo signal that is added with the composite signal Z(k) that is constituted by the speech signal originating from the local loop and the echo. In the situation where the subscriber at the local loop is not talking then Z(k) is only the echo signal. If the echo canceller models the true echo path with accuracy, the resultant signal is echo free.
CA 0223160~ 1998-03-10 The echo canceller includes an adaptive filter that alters the input signal x(k~ to bring it as close as possible to the echo. A feedback loop 36 is provided allowing the filter 34 to alter its transfer function and converge to nullify or at least reduce the error signal e(k~. A novel aspect of the present invention resides in the provision of a decorrelation filter 38 that supplies a frequency rich input to the adaptive filter 34 which enables that fllter to converge faster. The decorrelation filter 38 uses the predictor coefficients from the speech decoder 30 to process the speech signal x(k) to extract there~rom the excitation information n(k). The excitation signal n(k) is much more decorrelated than the reconstructed speech signal and this allows the adaptive ~ilter 34 to reach a small error condition much ~aster.
The block diagram of the decorrelation filter 38 is shown at figure 6. It will be noted that the decorrelation filter is set as the inverse of the speech decoder. The output of the decorrelation filter is expressed by the following e~uation:
n (k) =x (k) -~ cix (k-i ) CA 0223160~ 1998-03-10 W O 98/06185 PCT/CA9~ 7/1 Where cl,cq, cM are the predictor coefficients for the given frame The predictor coefficients are updated at the same time as those of the speech decoder. Since the speech decoder 30 is constant within a frame, the decorrelation filter 38 needs to be modified only after each frame and then it is kept unchanged. The updating procedure of the decorrelation filter 38 is effected simply by copying the predictor coefficients from the speech decoder 30 by any convenient means.
Typically, the order of the decorrelation filter 38 would be 10 (order of the filter or value M) which is typical of most commercially available speech decoders.
Figure 7 is a more detailed block diagram of the echo canceller 32. In addition to the decorrelation filter block 38 and the adaptive filter 34 shown in figure 5 the more complete diagram of figure 7 also indicates that the echo canceller includes two additional function blocks namely a generator o~ coefficients 40 and a power estimator 42 WO 98106185 PCT/CA9~ C7~1 that process individually the re-constructed excitation signal n(k) to supply additional parameters to a novel adaptive algorithm that calculates the coefficients of the adaptive filter. Note that for the purpose of simplicity, the adaptive filter 34 has been shown at figure 7 as two separate blocks namely the adaptive algorithm 44 and the filter core 46 which effect the actual signal processing.
In the most preferred embodiment of the present invention the following modified least mean square algorithm has been implemented:
ho (k+l ) =ho (k) +~ Z ( ) Po (k) hl (k+l ) =hl (k) +1l ~Z (k) Pl (k) hN(k+l ) =h(N l, (k) +u ~ (k) PN-1 (k) ~here hi is the ith coefficient of the filter CA 0223160~ 1998-03-lO
W O 98/0618~ PCT/C~96/00771 core 46;
e(k) is the error signal;
Pi(k) is a ith coefficient (see below);
,u is a positive constant; and ~~2n (k) is the power estimate of the re-constructed excitation signal n(k) The block diagram of the circuit for calculating the coe~ficients of the filter core 46 is shown in figure 8.
The error e(k) is divided by the power estimate and multiplied by the coe~ficient Pi(k). The resultant N values are scaled by a positive constant u and added to the coefficients for adaptation. Once the parameters have been established they are transferred to the filter core 46.
Preferably the positive constant ,u should be less than l/3 N and most preferably it should be set to a value in the range 1/4 N to 1/5 N.
The power estimate is the exponential running average of n2(k). It can be established by using the following equation. The transfer function is graphically illustrated by the block diagram in figure 9.
~ n (k) = (1---N) &Z~ (k-1~ + Nn~ (k) Finally, the coefficients Pi(k) are generated from the excitation signal as follows:
pO (k) =n (k) pi(k)=p~ ~(k-1)-c,(k) n( k) lsisM
Pi (k) =P~i l, (k-l) M+l~;i<N-1 Where: ci is the ith coefficient transferred from the voice decoderi N-1 is the order of the echo canceller;
M is the order of the voice decoder The block diagram for generating the coefficients is shown in figure 10.
Typically, N has a value in the range from 300 to 2000. In the most preferred embodiment of the invention N has a value of 384.
W O 98/06185 . PCT/CA~5.~~7/1 The above description of the invention should not be interpreted in any limiting manner since variations and refinements of the preferred embodiment are possible without departing from the spirit of the invention. The scope of the invention is defined in the appended claims and their equivalents.
Claims (18)
1) An adaptive echo canceller for reducing a magnitude of an echo occurring in a return channel of a communication system, the communication system including a forward channel through which propagates a first signal that is a precursor of the echo in the return channel of the communication system, the first signal being a re-constructed speech signal, said echo canceller including:
- first means for generating an echo estimate, said first means including:
a) a first input for receiving the first signal that is a precursor of the echo in the return channel of the communication system;
b) a second input for receiving a set of first coefficients of a voice decoder capable of generating the first signal;
c) first processing means for processing the first set of coefficients and the first signal to generate an excitation signal that is more decorelated than the first signal d) second processing means for processing the first set of coefficients and the excitation signal to generate a second set of coefficients;
e) an adaptive filter receiving the second. set of coefficients for processing the first signal in accordance with a transfer function determined at least in part by the second set of coefficients to generate the echo estimate - second means in operative relationship with said first means, said second means being responsive to the echo estimate for conditioning a signal propagating on the return channel of the communication system to inhibit echo therein.
- first means for generating an echo estimate, said first means including:
a) a first input for receiving the first signal that is a precursor of the echo in the return channel of the communication system;
b) a second input for receiving a set of first coefficients of a voice decoder capable of generating the first signal;
c) first processing means for processing the first set of coefficients and the first signal to generate an excitation signal that is more decorelated than the first signal d) second processing means for processing the first set of coefficients and the excitation signal to generate a second set of coefficients;
e) an adaptive filter receiving the second. set of coefficients for processing the first signal in accordance with a transfer function determined at least in part by the second set of coefficients to generate the echo estimate - second means in operative relationship with said first means, said second means being responsive to the echo estimate for conditioning a signal propagating on the return channel of the communication system to inhibit echo therein.
2) An echo canceller as defined in claim 1, including a divergence detector means for calculating an error signal between said echo estimate and an actual echo propagating on the return channel of the communication system, said second processing means including an input for receiving the error signal.
3) An echo canceller as defined in claim 2, wherein said second processing means calculates a power estimate of the excitation signal and utilizes the power estimate in the computation of the second set of coefficients.
4) An echo canceller as defined in claim 3, wherein said second means includes a subtractor that subtracts the echo estimate from a signal propagating in the return channel.
5) An echo canceller as defined in claim 4, wherein said second set of coefficients are computed by the following equations:
Where h i is the ith coefficient of the second set of coefficients that includes N
coefficients;
a (k) is the error signal;
p i(k) is an ith coefficient of a third set of coefficients that are correlated to the first set of coefficients;
µ is a positive constant; and ~2n(k) is a power estimate of the excitation signal.
Where h i is the ith coefficient of the second set of coefficients that includes N
coefficients;
a (k) is the error signal;
p i(k) is an ith coefficient of a third set of coefficients that are correlated to the first set of coefficients;
µ is a positive constant; and ~2n(k) is a power estimate of the excitation signal.
6) An echo canceller as defined in claim 5, wherein the value ~2n(k) is computed by using the following equation:
Where n(k) is the excitation signal.
Where n(k) is the excitation signal.
7) An echo canceller as defined in claim 6, wherein the coefficients p i(k) are computed as follows:
Where: c i is the ith coefficient of the first set of coefficients;
M is the number of coefficients in the first set of coefficients.
Where: c i is the ith coefficient of the first set of coefficients;
M is the number of coefficients in the first set of coefficients.
8) An echo canceller as defined in claim 7, wherein has a value in the range from 1/4 N to 1/5 N.
9) An echo canceller as defined in claim 8, wherein N
has a value in the range from 300 to 2000.
has a value in the range from 300 to 2000.
10) A method for reducing a magnitude of an echo occurring in a return channel of a communication system, the communication system including a forward channel through which propagates a first signal that is a precursor of the echo in the return channel of the communication system, the first signal being a re-constructed speech signal, said method comprising the steps of:
- providing a first set of coefficients of a voice decoder capable of generating the first signal;
- processing the first set of coefficients and the first signal to generate an excitation signal that is more decorelated than the first signal;
- processing the first set of coefficients and the excitation signal to generate a second set of coefficients;
- supplying the second set of coefficients to an adaptive filter having a transfer function determined at least in part by the second set of coefficients;
- processing the first signal by said adaptive filter to generate the echo estimate;
- utilizing said echo estimate to inhibit echo propagating in the return channel of the communication system.
- providing a first set of coefficients of a voice decoder capable of generating the first signal;
- processing the first set of coefficients and the first signal to generate an excitation signal that is more decorelated than the first signal;
- processing the first set of coefficients and the excitation signal to generate a second set of coefficients;
- supplying the second set of coefficients to an adaptive filter having a transfer function determined at least in part by the second set of coefficients;
- processing the first signal by said adaptive filter to generate the echo estimate;
- utilizing said echo estimate to inhibit echo propagating in the return channel of the communication system.
11) A method as defined in claim 10, comprising the step of calculating an error signal between said echo estimate and an actual echo propagating on the return channel of the communication system, and utilizing said error signal to generate said second set of coefficients.
12) A method as defined in claim 11, comprising the step of computing a power estimate of the excitation signal and utilizing said power estimate in the computation of the second set of coefficients.
13) A method as defined in claim 12, comprising the step of subtracting said echo estimate from a signal propagating in the return channel.
14) A method as defined in claim 11, comprising the step of computing said second set of coefficients by the following equations.
Where h i is the ith coefficient of the second set of coefficients that includes N
coefficients;
e(k) is the error signal;
p i(k) is an ith coefficient of a third set of coefficients that are correlated to the first set of coefficients;
µ is a positive constant; and is a power estimate of the excitation signal.
Where h i is the ith coefficient of the second set of coefficients that includes N
coefficients;
e(k) is the error signal;
p i(k) is an ith coefficient of a third set of coefficients that are correlated to the first set of coefficients;
µ is a positive constant; and is a power estimate of the excitation signal.
15) A method as defined in claim 14, wherein the value is computed by using the following equation:
Where n(k) is the excitation signal.
Where n(k) is the excitation signal.
16) A method as defined in claim 15, wherein the coefficients p i(k) are computed as follows:
Where: c i is the ith coefficient of the first coefficients;
M is the number of coefficients in the first set of coefficients.
Where: c i is the ith coefficient of the first coefficients;
M is the number of coefficients in the first set of coefficients.
17) A method as defined in claim 16, wherein µ has a value in the range from 1/4 N to 1/5 N.
18) A method as defined in claim 17, wherein N has a value in the range from 300 to 2000.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US69091496A | 1996-08-01 | 1996-08-01 | |
| US08/690,914 | 1996-08-01 | ||
| PCT/CA1996/000771 WO1998006185A1 (en) | 1996-08-01 | 1996-11-25 | Echo cancelling system for digital telephony applications |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2231605A1 CA2231605A1 (en) | 1998-02-12 |
| CA2231605C true CA2231605C (en) | 2001-11-06 |
Family
ID=24774476
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002231605A Expired - Fee Related CA2231605C (en) | 1996-08-01 | 1996-11-25 | An enhanced echo canceller for digital cellular application |
Country Status (8)
| Country | Link |
|---|---|
| US (1) | US5912966A (en) |
| EP (1) | EP0853844B1 (en) |
| JP (1) | JPH11513230A (en) |
| CN (1) | CN1111973C (en) |
| CA (1) | CA2231605C (en) |
| DE (1) | DE69627359T2 (en) |
| HK (1) | HK1017526A1 (en) |
| WO (1) | WO1998006185A1 (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1569357A1 (en) * | 1997-09-16 | 2005-08-31 | Sanyo Electric Co., Ltd. | Voice switch |
| EP1022865B1 (en) * | 1997-09-16 | 2011-03-30 | Sanyo Electric Co., Ltd. | Echo canceller and echo cancelling method |
| US6181791B1 (en) * | 1998-01-06 | 2001-01-30 | Stmicroelectronics, Inc. | Apparatus and method for reducing local interference in subscriber loop communication system |
| GB2336279A (en) * | 1998-03-30 | 1999-10-13 | Motorola Inc | Echo canceller and method for device having a decoder |
| FI981091A (en) | 1998-05-15 | 1999-11-16 | Nokia Networks Oy | Method and apparatus for echo cancellation in a digital mobile communication system |
| US6560332B1 (en) | 1999-05-18 | 2003-05-06 | Telefonaktiebolaget Lm Ericsson (Publ) | Methods and apparatus for improving echo suppression in bi-directional communications systems |
| US6526140B1 (en) | 1999-11-03 | 2003-02-25 | Tellabs Operations, Inc. | Consolidated voice activity detection and noise estimation |
| US6804350B1 (en) * | 2000-12-21 | 2004-10-12 | Cisco Technology, Inc. | Method and apparatus for improving echo cancellation in non-voip systems |
| JP3771812B2 (en) * | 2001-05-28 | 2006-04-26 | インターナショナル・ビジネス・マシーンズ・コーポレーション | Robot and control method thereof |
| KR100431965B1 (en) * | 2001-06-23 | 2004-07-21 | (주) 에스제이-테크 | Apparatus and method for removing echo-audio signal using time-varying algorithm with time-varying step size |
| US6957240B2 (en) * | 2001-08-08 | 2005-10-18 | Octasic Inc. | Method and apparatus for providing an error characterization estimate of an impulse response derived using least squares |
| US6965640B2 (en) * | 2001-08-08 | 2005-11-15 | Octasic Inc. | Method and apparatus for generating a set of filter coefficients providing adaptive noise reduction |
| US6999509B2 (en) * | 2001-08-08 | 2006-02-14 | Octasic Inc. | Method and apparatus for generating a set of filter coefficients for a time updated adaptive filter |
| US6970896B2 (en) | 2001-08-08 | 2005-11-29 | Octasic Inc. | Method and apparatus for generating a set of filter coefficients |
| US6751313B2 (en) * | 2001-10-31 | 2004-06-15 | Conexant Systems, Inc. | Method and apparatus for an improved echo canceller |
| KR100466567B1 (en) * | 2002-11-26 | 2005-01-15 | 현대모비스 주식회사 | Echo signal removing device |
| EP1718103B1 (en) | 2005-04-29 | 2009-12-02 | Harman Becker Automotive Systems GmbH | Compensation of reverberation and feedback |
| CN101312372B (en) * | 2008-05-12 | 2013-01-02 | 北京创毅视讯科技有限公司 | Echo eliminator and echo eliminating method |
| GB2465047B (en) * | 2009-09-03 | 2010-09-22 | Peter Graham Craven | Prediction of signals |
| CN103795473B (en) * | 2012-11-02 | 2017-04-12 | 华为技术有限公司 | Method and system for eliminating power-frequency interference |
| US10468020B2 (en) * | 2017-06-06 | 2019-11-05 | Cypress Semiconductor Corporation | Systems and methods for removing interference for audio pattern recognition |
| US10382145B2 (en) * | 2017-07-13 | 2019-08-13 | Benjamin J. Egg | System and method for improving wireless data links |
| CN109040499B (en) * | 2018-08-14 | 2020-12-01 | 西南交通大学 | Adaptive echo cancellation method for resisting impact interference |
Family Cites Families (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4468641A (en) * | 1982-06-28 | 1984-08-28 | At&T Bell Laboratories | Adaptive filter update normalization |
| US4712235A (en) * | 1984-11-19 | 1987-12-08 | International Business Machines Corporation | Method and apparatus for improved control and time sharing of an echo canceller |
| US4584411A (en) * | 1985-02-25 | 1986-04-22 | Shell Oil Company | Hydroformylation process |
| EP0199879B1 (en) * | 1985-04-30 | 1990-03-21 | International Business Machines Corporation | A process and system for improving echo cancellation within a transmission network |
| US4697261A (en) * | 1986-09-05 | 1987-09-29 | M/A-Com Government Systems, Inc. | Linear predictive echo canceller integrated with RELP vocoder |
| US5062102A (en) * | 1988-12-01 | 1991-10-29 | Nec Corporation | Echo canceller with means for determining filter coefficients from autocorrelation and cross-correlation coefficients |
| JPH0748681B2 (en) * | 1989-02-23 | 1995-05-24 | 日本電気株式会社 | Echo canceller coefficient control method |
| JP2842607B2 (en) * | 1989-03-13 | 1999-01-06 | 株式会社日立製作所 | Echo canceller, communication device including the same, and signal processing method |
| US5305309A (en) * | 1989-12-06 | 1994-04-19 | Fujitsu Limited | Echo canceller |
| JP2518433B2 (en) * | 1990-01-24 | 1996-07-24 | 日本電気株式会社 | Double talk detection circuit |
| WO1991012683A1 (en) * | 1990-02-16 | 1991-08-22 | Fujitsu Limited | Speaking device having hand-free conversational function |
| JPH03262939A (en) * | 1990-03-14 | 1991-11-22 | Fujitsu Ltd | Method and device for detecting echo path variation |
| US5305307A (en) * | 1991-01-04 | 1994-04-19 | Picturetel Corporation | Adaptive acoustic echo canceller having means for reducing or eliminating echo in a plurality of signal bandwidths |
| US5263019A (en) * | 1991-01-04 | 1993-11-16 | Picturetel Corporation | Method and apparatus for estimating the level of acoustic feedback between a loudspeaker and microphone |
| JP2836277B2 (en) * | 1991-03-14 | 1998-12-14 | 国際電信電話株式会社 | Echo cancellation device |
| EP0543568A2 (en) * | 1991-11-22 | 1993-05-26 | AT&T Corp. | High resolution filtering using low resolution processors |
| CA2060667C (en) * | 1992-02-05 | 1998-12-08 | Paul Marc Yatrou | Adaptive sparse echo canceller using a sub-rate filter for active tap selection |
| JP2853455B2 (en) * | 1992-05-15 | 1999-02-03 | ケイディディ株式会社 | Echo canceller |
| US5307405A (en) * | 1992-09-25 | 1994-04-26 | Qualcomm Incorporated | Network echo canceller |
| US5374705A (en) * | 1993-12-27 | 1994-12-20 | The Dow Chemical Company | Process for the preparation of polyether polyols with a reduced unsaturation content |
| US5475731A (en) * | 1994-01-07 | 1995-12-12 | Ericsson Inc. | Echo-canceling system and method using echo estimate to modify error signal |
| DE69532394T2 (en) * | 1994-02-10 | 2004-11-04 | Nippon Telegraph And Telephone Corp. | Method and device for echo cancellation using the "fast projection scheme" |
| US5592548A (en) * | 1995-05-31 | 1997-01-07 | Qualcomm Incorporated | System and method for avoiding false convergence in the presence of tones in a time-domain echo cancellation process |
| US5613899A (en) * | 1995-06-05 | 1997-03-25 | Southern Carbide Specialists, Inc. | Centerless ceramic ferrule grinder |
-
1996
- 1996-11-25 CN CN96198688A patent/CN1111973C/en not_active Expired - Fee Related
- 1996-11-25 CA CA002231605A patent/CA2231605C/en not_active Expired - Fee Related
- 1996-11-25 JP JP10507405A patent/JPH11513230A/en not_active Ceased
- 1996-11-25 EP EP96938873A patent/EP0853844B1/en not_active Expired - Lifetime
- 1996-11-25 WO PCT/CA1996/000771 patent/WO1998006185A1/en active IP Right Grant
- 1996-11-25 DE DE69627359T patent/DE69627359T2/en not_active Expired - Fee Related
-
1997
- 1997-09-19 US US08/933,383 patent/US5912966A/en not_active Expired - Fee Related
-
1999
- 1999-06-07 HK HK99102481A patent/HK1017526A1/en not_active IP Right Cessation
Also Published As
| Publication number | Publication date |
|---|---|
| WO1998006185A1 (en) | 1998-02-12 |
| DE69627359D1 (en) | 2003-05-15 |
| CA2231605A1 (en) | 1998-02-12 |
| CN1203711A (en) | 1998-12-30 |
| HK1017526A1 (en) | 1999-11-19 |
| US5912966A (en) | 1999-06-15 |
| JPH11513230A (en) | 1999-11-09 |
| EP0853844A1 (en) | 1998-07-22 |
| DE69627359T2 (en) | 2003-10-30 |
| EP0853844B1 (en) | 2003-04-09 |
| CN1111973C (en) | 2003-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2231605C (en) | An enhanced echo canceller for digital cellular application | |
| CN103391381B (en) | Method and device for canceling echo | |
| US7039181B2 (en) | Consolidated voice activity detection and noise estimation | |
| US4764955A (en) | Process for determining an echo path flat delay and echo canceler using said process | |
| EP1208689B1 (en) | Acoustical echo cancellation device | |
| EP1152547A2 (en) | Network echo canceller | |
| US4697261A (en) | Linear predictive echo canceller integrated with RELP vocoder | |
| CA2162571C (en) | Echo canceler and method for learning for the same | |
| EP1012996A2 (en) | Method and apparatus for echo estimation and suppression | |
| JPH11331046A (en) | Method and device for suppressing echo and computer readable storage medium storing echo suppressing program | |
| WO2002062046A1 (en) | Line echo canceller scalable to multiple voice channel/ports | |
| EP1393461B1 (en) | Echo canceller and a method of cancelling echo | |
| US7711107B1 (en) | Perceptual masking of residual echo | |
| JP3970746B2 (en) | Echo canceller performance evaluation test equipment | |
| MXPA98002468A (en) | Echo cancelling system for digital telephony applications | |
| KR20000057739A (en) | Subband echo canceller and method therefor | |
| Periakarruppan | Packet based echo cancellation for voice over Internet protocol | |
| JPH08213939A (en) | Echo canceler | |
| KANG et al. | A new post-filtering algorithm for residual acoustic echo cancellation in hands-free mobile application | |
| JPH098709A (en) | Sub-band echo erasing method | |
| J Jones | Echo control techniques in public switched telephone networks | |
| EP1341365A1 (en) | Method and arrangement for processing a speech signal | |
| Tangwongsan et al. | Echo cancellation in VoIP with improved least square lattice method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed |