Echo Cancellation

Abstract

An echo cancellation device (1) comprises a first adaptive filter (13) for producing a first echo cancellation signal (y1), a second adaptive filter (15) for producing a second echo cancellation signal (y2), and a post-processor (18) for suppressing any remaining echo. The first adaptive filter (13) and the second adaptive filter (15) are designed for canceling a first (e.g. direct) part of the echo impulse response and a second (e.g. diffuse) part of the echo impulse response respectively. The device (1) may be utilized in a mobile telephone.

Description

The present invention will further be explained below with reference to exemplary embodiments illustrated in the accompanying drawings, in which:

FIG. 1 schematically shows an echo cancellation device according to the Prior Art.

FIG. 2 schematically shows a first embodiment of an echo cancellation device according to the present invention.

FIG. 3 schematically shows a second embodiment of an echo cancellation device according to the present invention.

FIG. 4 schematically shows a third embodiment of an echo cancellation device according to the present invention.

FIG. 5 schematically shows a fourth embodiment of an echo cancellation device according to the present invention.

FIG. 6 schematically shows a first arrangement of filter sections according to the present invention.

FIG. 7 schematically shows a second arrangement of filter sections according to the present invention.

FIG. 8 schematically shows a mobile (cellular) telephone apparatus which incorporates a device according to the present invention.

FIG. 9 schematically shows a typical impulse response of a mobile (cellular) telephone.

FIG. 10 schematically shows the energy decay curve corresponding with the impulse response of FIG. 9.

The echo canceling device 1′ according to the Prior Art as shown in FIG. 1 comprises an input terminal 11 for receiving a far-end signal x, an output terminal 12 for supplying a (processed) residual signal r_p, an adaptive filter (AF) 13, a filter update unit 15, an (optional) amplifier 17, a post-processor 18 and a combination unit 19. The amplifier 17 is coupled to a first transducer (loudspeaker) 2 while the combination unit 19 is coupled to a second transducer (microphone) 3.

The loudspeaker 2 reproduces the far-end signal x that has been amplified by the amplifier 17. This reproduced far-end signal appears as an echo e at the microphone 3, together with the near-end signal s. The echo cancellation device 1′ attempts to remove the echo e from the microphone signal z.

The microphone signal z is combined with an echo cancellation signal y produced by the adaptive filter. The combination unit 19 is typically constituted by an adder having a negative or inverting input (−) to which the echo cancellation signal y is fed, resulting in a subtraction of said signal. The residual signal r produced by the combination unit 19 is therefore the difference signal of the echo cancellation signal y and the microphone signal z: r=z−y.

The filter update unit 15 typically determines the correlation of the far-end signal x and the residual signal r and controls the(coefficients of the) adaptive filter 13 in such a way that this correlation is minimized. It can be seen from FIG. 1 that the echo cancellation signal y is based upon the far-end signal x. This will later be explained in more detail with reference to FIG. 6.

A post-processor (PP) 18 processes the residual signal r so as to remove remaining echo components and produces the processed residual signal r_p. The echo cancellation signal y and the microphone signal z are fed to the post-processor 18 as auxiliary signals.

An echo canceling device of this type is disclosed in United States Patent Application US 2003/0031315 (Belt et al./Philips), the entire contents of which are herewith incorporated in this document.

Although the echo canceling device of FIG. 1 generally works well, there are cases in which it does not work satisfactorily and the near-end signal is distorted. An example is a (mobile) telephone operating in “hands-free” mode, where the echo signal e produced by the loudspeaker 2 in response to the far-end signal x is much greater than the near-end (e.g. speech) signal s produced by a user of the telephone. Neither the adaptive filter 13 nor the post-processor 18 is capable of removing such an echo signal without introducing signal distortion. The present invention seeks to solve this problem.

The echo canceling device 1 according to the present invention shown merely by way of non-limiting example in FIG. 2 comprises an input terminal 11, an output terminal 12, a first adaptive filter 13, a second adaptive filter 14, a first filter update unit 15, a second filter update unit 16, an (optional) amplifier 17, a post-processor 18, a first combination unit 19, a second combination unit 20, and a delay unit 21. It will be understood that in digital embodiments, the device 1 will further comprise suitable A/D (analog/digital) and D/A (digital/analog) converters which are not shown for the sake of clarity of the illustration.

The first adaptive filter 13 produces a first echo cancellation signal y₁ which is combined with the microphone signal z in the first combination unit 19 to produce a first residual signal r₁. Similarly, the second adaptive filter 14 produces a second echo cancellation signal y₂ which the second combination unit 20 combines with the first residual signal r₁ to produce a second residual signal r₂. The post-processor 18 receives this second residual signal r₂ and the echo cancellation signals y₁ and y₂ to produce a processed residual signal r_p from which the echoes are substantially completely removed.

In accordance with the present invention, the second adaptive filter 14 operates on the delayed far-end signal x_d produced by the delay unit 21, while the first adaptive filter 13 operates on the original far-end signal x. As a result, the first adaptive filter 13 cancels the echoes corresponding with the first part of the impulse response underlying the echo signal, while the second adaptive filter 14 cancels the echoes of a second, further part of the impulse response. This will be further explained with reference to FIGS. 9 and 10.

FIG. 9 shows an exemplary impulse response of a mobile (cellular) telephone in “hands-free” mode, while FIG. 10 shows the Energy Decay Curve (EDC) of the same signal. It will be clear to those skilled in the art that the impulse response is the signal received by the microphone (3 in FIG. 2) when a Dirac (or “spike”) signal is produced by the loudspeaker (2 in FIG. 2) and that the impulse response represents the acoustic behavior of the device and its surroundings.

The EDC is a measure of the remaining energy of the signal and may be expressed mathematically as

$\begin{array}{cc}\mathrm{EDC}\left(i\right)=10\log \sum _{m=i}^{\infty }h_m^2& \left(1\right)\end{array}$

where h_m is the amplitude of the m^th signal sample. In FIG. 10, the EDC of the signal of FIG. 9 is shown as a function of time, the moment in time being indicated by the sample number (n).

The amplitude A of the signal of FIG. 9 is initially zero. After a short period, in the example shown corresponding with approximately 20 samples (the time is indicated by the sample number n), the direct part of the signal arrives at the microphone. This causes a sharp drop in the EDC curve as the remaining energy of the signal rapidly decreases. After a time T, in the present example corresponding with 100 samples, the diffuse reverberations part of the signal begins. This is evidenced in FIG. 10, where the EDC between n=100 and n=800 is approximately a straight line, the slope of which is related to the reverberation time (often referred to as “T60”) of the room in which the device is located. As the EDC is a logarithmic measure, expressed in deciBel (dB), the impulse response signal of FIG. 9 decays exponentially from t=T (that is, from n=100).

The present inventors have realized that the direct part (from t=0 to t=T) and the diffuse part (from t=T) of the impulse response have different properties. The echo signal, which can be described as the convolution of the impulse response and the (amplified) far-end signal, should therefore be canceled using two (or more) echo cancellation signals based upon different parts of the impulse response. The operation of the post-processor may be significantly improved by using these different echo cancellation signals.

Returning to FIG. 2, it can be seen that the filter lengths of the filters 13 and 14, and the delay of the delay unit 21 are preferably chosen such that the first filter 13 produces an echo cancellation signal y₁ that corresponds with the first, direct part of the impulse response (from t=0 to t=T) while the second filter 14 produces an echo cancellation signal y₂ that corresponds with the second, diffuse part of the impulse response (from t=T). This may be accomplished, in the example shown, by both a filter length and a delay equal to 100 samples. This will cause the first echo cancellation signal y₁ to substantially completely cancel the impulse response during the first 100 samples, after which the second echo cancellation signal y₂ accomplishes the same during the second 100 samples.

The operation of the filter units 13 and 14 is controlled by the respective filter (coefficient) update units 15 and 16. These filter update units 15 and 16 determine the coefficients of the filter units 13 and 14 respectively using techniques which may be known per se, for example techniques based on the correlation of the far-end signal x and the first residual signal r₁ (filter update unit 15), and the correlation of the delayed far-end signal x_d and the second residual signal r₂ (filter update unit 16). In accordance with the present invention, both the first echo cancellation signal y₁ and the second echo cancellation signal y₂ are fed to the post-processor 18 to further process the (second) residual signal r₂ so as to substantially remove any remaining echoes. The post-processor of the present invention preferably utilizes spectral subtraction to remove remaining echoes. Accordingly, the absolute value |R_p| of the frequency spectrum R_p of the processed residual signal r_p may be calculated by:

|R_p|=|R₁|−γ₁ε.|Y₁|−γ₂.|Y₂|  (2)

where |R₁|, |Y₁| and |Y₂| are the absolute values (that is, magnitudes) of the frequency spectra of the (first) residual signal r₁, the first echo cancellation signal y₁ and the second echo cancellation signal y₂ respectively, γ₁ and γ₂ are a first and a second over-subtraction factor respectively, and ε is an estimate of the achieved echo return loss enhancement (ERLE) of the first adaptive filter 13. The product ε.|Y₁| is therefore an estimate of the residual direct echo signal after the first combination unit 19. The (complex) frequency spectrum R_p is determined using the absolute value |R_p| and the phase of the spectrum R₂ calculated from the second residual value r₂. Using an inverse Fourier transform, the processed residual signal r_p is determined from the complex spectrum R_p.

The ERLE factor ε can be estimated by the formula:

$\begin{array}{cc}\varepsilon =\frac{\stackrel{\_}{R_1}}{\stackrel{\_}{Z}}& \left(3\right)\end{array}$

where |R₁| is the absolute value of the frequency spectrum of the (first) residual signal r₁, |Z| is the absolute value of the frequency spectrum of the microphone signal z, and both |R₁| and |Z| are averaged over a (short) time period in which no near-end signal (s in FIG. 2) is present and the microphone signal accordingly consists of the echo signal only. The factor ε is therefore indicative of the extent to which the echo is dampened by the (first) adaptive filter: in the absence of a near-end signal, the residual signal r₁, and hence its spectrum R₁, is ideally equal to zero, and ε=0. If the adaptive filter produces no echo cancellation signal y₁, the residual signal r₁ is equal to the microphone signal z and ε=1. The ERLE factor ε may be determined by the post-processor 18, but may also be determined by a separate unit (not shown), external to the post-processor. The ERLE factor ε is typically approximately equal to 0.2, although other values are also found.

The over-subtraction factors γ₁ and γ₂ determine the weights of the respective spectra in the spectral subtraction of formula (2) and are typically slightly greater than or approximately equal to 1, for example 1.1, although values ranging from about 0.5 to about 2.0 may be used. It is preferred that γ₂ is greater than γ₁. The over-subtraction factors compensate for the fact that the amplitudes of the echo cancellation signals typically have some variance relative to their “ideal” values and can be too small to effectively cancel the echo signal.

It can be seen from formula (2) that the frequency spectra of both the first echo cancellation signal y₁ and the second echo cancellation signal y₂ are used by the post-processor 18. In addition, the frequency spectrum of the first residual signal r₁ is used in formula (2). Although FIG. 2 shows a direct connection for feeding the first residual signal r₁ to the post-processor 18, the post-processor may derive this signal from the signals r₂ and y₂. In that case, the said connection may be omitted, as is shown in FIG. 3.

The value |R_p| is typically determined for each frequency component separately and is therefore frequency-dependent. The value |R_p| may further be used to determine a gain factor G which may be defined by:

$\begin{array}{cc}G=\frac{\mathrm{Rp}}{R_2}& \left(4\right)\end{array}$

which gain factor G is then multiplied with the complex value R₂ of the residual signal r₂ to obtain the spectrum R_p of the processed residual signal r_p:

R_p=G.R₂  (5)

The original phase of the (complex) spectrum R₂ is used to produce R_p and, using an inverse Fourier transform, the signal r_p. It is noted that the gain factor G is preferably also frequency-dependent. Gain factors may advantageously be used to limit the amount of echo suppression in order to reduce signal distortion. To this end, the gain factor may be compared with a minimum gain, the gain factor having a maximum value equal to one.

It is further noted that the post-processor 18 preferably processes the signals z, y₁, y₂, . . . per frame or block (B), each block being subjected to a fast Fourier transform (FFT) to obtain the complex frequency spectrum. This complex spectrum is separated into a magnitude (absolute value) and a phase using well-known techniques.

By using two distinct echo cancellation signals (y₁ and y₂) in the post-processor, a much improved post-processing is achieved which results in a virtually complete echo cancellation without distorting the near-end signal.

Instead of using the first residual signal r₁ as in formula (2), the post-processor may utilize the microphone signal z to determine the processed residual signal r_p using the formula:

|R_p|=|Z|−γ₁.|Y₁|−γ₂.|Y₂|  (6)

where |Z| is the absolute value of the spectrum Z of the microphone signal z. This is schematically illustrated in FIG. 3, where the connection feeding r₁ to the post-processor has been omitted. As in the embodiment of FIG. 2, the value |R_p| may be used to calculate a gain factor G, as in formula (4). It is noted that formula (6) does not require the ERLE factor ε.

The exemplary device 1 of FIG. 4 contains the same components as the device 1 of FIG. 3. In addition, a tail estimation unit 22 is coupled to the post-processor 18. This tail estimation unit 22 provides an estimate of the “tail” of the signal, that is, the part of the signal that cannot be echo-compensated by the first and the second filters due to the limited filter lengths.

Referring to FIG. 9, it was mentioned above that the first filter preferably has a length corresponding to T (in the example given: 100 samples), while the second filter and the delay have similar lengths. In the example discussed above, this means that the echo in only the first 200 samples of the impulse response is compensated by the filters (the echo compensation amounts to approximately −30 dB, as indicated by FIG. 10). The tail estimation unit 22 cancels echo components in the remainder of the signal (that is, from 200 samples) by estimating this “tail”. The (absolute value of the) frequency spectrum of the output (processed residual) signal r_p may now be determined by:

|R_p|=|Z|−γ₁.|Y₁|−γ₂.|Y₂|−γ₃.|Y₃|  (7)

where γ₃ is the over-subtraction factor of the absolute value |Y₃| of the tail spectrum Y₃. As is the case with γ₁ and γ₂, the over-subtraction factor γ₃ is slightly greater than or approximately equal to 1, and γ₃>γ₂>γ₁. As can be seen, formula (7) is largely identical to formula (6), with the exception of an added tail term involving the tail estimation. The tail spectrum Y₃ may be estimated as follows:

|Y₃|=α.|Y₃|₋₁+|Y_2B|₋₁  (8)

where the subscript “−1” indicates a previous block or frame, and where α is a factor related to the reverberation time (often referred to as “T60 time” by those skilled in the art) of the room in which the device is located, α typically being smaller than one. In addition, Y_2B is the spectrum of the signal obtained by convolving the last B coefficients of the second filter 14 with the delayed far-end signal x_d:

$\begin{array}{cc}{y}_{2B}=\sum _{l=l0}^{l1}h_2\left(l\right)·x_d\left(n-l\right)& \left(9\right)\end{array}$

where l0=N₂−B, l1=N₂−1, the block size B is a number approximately equal to 80, and h₂(1) are the coefficients of the second filter 14. This will later be explained in more detail with reference to FIGS. 6 and 7.

It is noted that the tail estimation unit 22 of FIG. 4 may also be added to the embodiments of FIGS. 2 and 5.

The embodiment of FIG. 5 comprises a single adaptive filter 13′ and an associated filter update unit 15. The adaptive filter 13′ of FIG. 5 produces an echo cancellation signal y that consists of a combination of the echo cancellation signals y₁ and y₂ of FIGS. 2-4: the first part of the echo cancellation signal y consists of y₁ while the second part is identical to y₂. Typically but not necessarily, the two parts have equal lengths, in which case the first echo cancellation signal y₁ constitutes the first half of the composite echo cancellation signal y, while the second echo cancellation signal y₂ forms its second half. The residual signal r, the first echo cancellation signal y₁ and the second echo cancellation signal y₂ are fed to the post-processor 18 to produce a processed residual signal r_p. In the embodiment shown in FIG. 5, the microphone signal z is also fed to the post-processor 18. The processing of these signals may be carried out in accordance with the formulae given above.

The structure of the filter 13′ of FIG. 5 will be explained with reference to FIGS. 6 and 7. FIG. 6 schematically shows digital filter sections 13 and 14 and a delay unit 21 in accordance with FIGS. 2-4. The first filter section 13 comprises a number of filter cells 40₀-40₁, each coupled to an associated filter coefficient unit 41₀-41₁. The filter coefficients of these units 40_i (i=0 . . . 1) are determined by the filter update unit 15. The first filter cell 40₀ receives samples of the far-end signal x, which are shifted through the subsequent cells (to the right in FIGS. 6 and 7). The sample value of each filter cell 40_i is multiplied by the filter coefficient of the respective coefficient unit 41_i so as to produce a weighted sample value. Summation unit 42 sums all weighted values of the filter 13 to produce the first echo cancellation signal y₁. It is noted that the repeated addition of the weighted sample values amounts to a convolution of the signal x and the filter coefficients.

Similarly, the filter section 14 has a number of filter cells 40₁₊₁-40_m and associated filter coefficient units 41₁₊₁-41_m which produce weighted sample values that are summed by a summation unit 43 to produce a second echo cancellation signal y₂. These echo cancellation signals y₁ and y₂ may be added in an adder 44 to produce a combined echo cancellation signal y.

The length of a data block B is typically less than or equal to the length of a filter section so as to allow an entire block of samples to be filtered simultaneously. The block length B is used in formula (9) above. It can be seen that the signal y_2B is equal to the signal y₂ if the filter length is equal to the block length B. If the block length B is smaller than the filter length, a separate summation unit 43′ (not shown) may be used to produce the signal y_2B of formula (9).

The second filter section 14 receives delayed signal samples from the delay unit 21. It is preferred that the delay imposed by delay unit 21 is approximately equal to the delay caused by the filter units 40_i of the first filter section 13 so that the filter sections 13 and 14 process contiguous sets of samples.

The combined filter unit 13′ of FIG. 7 also comprises a first filter section 13 and a second filter section 14. However, these filter sections are not separate but are concatenated: samples are shifted from the last filter cell 40₁ of the first filter section 13 into the first filter cell 40₁₊₁ of the second filter section 14. Each filter section has a dedicated summation unit 42, 43 for producing a respective echo cancellation signal y₁, y₂. An adder 44 combines these individual echo cancellation signals into a combined echo cancellation signal y.

It is noted that no delay unit 21 is required in the embodiment of FIG. 7. Instead, the first filter section 13 provides a suitable delay for the second filter section 14. This delay corresponds exactly with the filter length of the first filter section 13.

The echo cancellation device of the present invention may be incorporated into various other devices, for example consumer devices such as mobile (cellular) telephone apparatus. An exemplary telephone apparatus 80 incorporating the echo cancellation device of the present invention is shown in FIG. 8. The telephone apparatus 80 benefits from improved echo cancellation and reduced signal distortion, in particular in “hands-free” mode.

The preferred embodiment of the present invention may be summarized as an echo cancellation device for canceling any echo in a microphone signal, the device comprising:

• • a first adaptive filter for producing a first echo cancellation signal,
  • a first combination device for combining the first echo cancellation signal and the microphone signal so as to produce a first residual signal,
  • a second adaptive filter for producing a second echo cancellation signal,
  • a second combination device for combining the first residual signal and the echo second cancellation signal so as to produce a second residual signal, and
  • a post-processor for suppressing any remaining echo in the second residual signal,wherein the first adaptive filter and the second adaptive filter are designed for canceling a first part of the echo impulse response and a second part of the echo impulse response respectively.

As mentioned above, it has been shown that the two adaptive filters may be replaced with a single adaptive filter having two concatenated sections, each section producing an individual echo cancellation signal, to obtain the same benefits.

The present invention is based upon the insight that an echo cancellation device may advantageously comprise two filters which each operate on a different part of the acoustic impulse response and each produce an individual echo cancellation signal. The present invention benefits from the further insight that two or more individual echo cancellation signals may advantageously be used by a post-processor to remove any remaining echoes from the residual signal.

It is noted that any terms used in this document should not be construed so as to limit the scope of the present invention. In particular, the words “comprise(s)” and “comprising” are not meant to exclude any elements not specifically stated. Single (circuit) elements may be substituted with multiple (circuit) elements or with their equivalents.

It will be understood by those skilled in the art that the present invention is not limited to the embodiments illustrated above and that many modifications and additions may be made without departing from the scope of the invention as defined in the appending claims.

Claims

1. An echo cancellation device (1) for canceling an echo in a microphone signal (z) in response to a far-end signal (x), the device comprising: a first adaptive filter section (13) for filtering the far-end signal (x) so as to produce a first echo cancellation signal (y1),a second adaptive filter section (14) for filtering a delayed far-end signal (xd) so as to produce a second echo cancellation signal (y2),at least one combination unit (19, 20) for combining an echo cancellation signal (y1, y2; y) with the microphone signal (z) so as to produce a residual signal (r1, r2; r), anda post-processor (18) for substantially removing any remaining echoes from the residual signal (r1, r2; r),

2. The device according to claim 1, wherein the post-processor (18) is arranged for utilizing the microphone signal (z).

3. The device according to claim 1, wherein the filter sections (13, 14) each constitute a filter unit having an individual associated filter update unit (15, 16), and wherein a delay unit (21) is provided for delaying the far-end signal (xd).

4. The device according to claim 3, wherein the delay unit (21) is arranged for providing a delay substantially corresponding to the filter length of the first filter unit (13).

5. The device according to claim 3, wherein the post-processor (18) is arranged for utilizing both a first residual signal (r1) produced by a first combination unit (19) and a second residual signal (r2) produced by a second combination unit (20), each combination unit (19, 20) preferably being coupled to a respective filter unit (13, 14).

6. The device according to claim 1, further comprising a tail estimation unit (22) for estimating the echo tail of the microphone signal (z), wherein the post-processor (18) is arranged for utilizing the estimated echo tail.

7. The device according to claim 1, wherein the filter sections (13, 14) together constitute a single adaptive filter unit (13′) arranged for producing a combined echo cancellation signal (y) comprising a combination of the first (y1) and the second (y2) echo cancellation signal.

8. The device according to claim 1, wherein the post-processor (18) is arranged for spectral subtraction.

9. The device according to claim 8, wherein the spectral subtraction involves over-subtraction factors (γ1, γ2, γ3).

10. An audio system, comprising an echo cancellation device (1) according to claim 1.

11. A mobile telephone device, comprising an echo cancellation device (1) according to claim 1.

12. A method of canceling an echo in a microphone signal (z) in response to a far-end signal (x), the method comprising the steps of: filtering the far-end signal (x) so as to produce a first echo cancellation signal (y1),filtering a delayed far-end signal (xd) so as to produce a second echo cancellation signal (y2),combining an echo cancellation signal (y1, y2; y) with the microphone signal (z) so as to produce a residual signal (r1, r2; r), andpost-processing the residual signal (r1, r2; r) so as to substantially remove any remaining echoes,

13. The method according to claim 12, wherein the step of post-processing involves spectral subtraction.

14. The method according to claim 12, further comprising the step of estimating the tail of the microphone signal (z), wherein the step of post-processing involves utilizing the estimated signal tail.

15. A computer program product for carrying out the method according to claim 12.