Multi-channel echo cancellation with round robin regularization

• US 7,352,858 B2
• Filed: 06/30/2004
• Issued: 04/01/2008
• Est. Priority Date: 06/30/2004
• Status: Expired due to Fees

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Platt, John C., Stokes, Jack W.
• Primary Examiner(s): Kuntz; Curtis A
• Assistant Examiner(s): Jamal; Alexander
• Application Number:
  US10/883,128
• Publication Number:
  US 20060002547A1
• Time in Patent Office: 1,371 Days
• Field of Search: 381/66, 381/93, 381/94.1, 379/406.06, 379/406.08
• US Class Current: 379/406.08
• CPC Class Codes: H04M 9/082 using echo cancellers echo cancellers per se H04B3/23

Abstract

A method and system of multi-channel echo cancellation using round robin regularization. The multi-channel round robin regularization echo cancellation method includes applying a plurality of adaptive filters, each having an inverse correlation matrix, to the multi-channel playback signal. Each of the plurality of adaptive filters is selected in a round robin sequence, so that every round each of the filters is selected. The inverse correlation matrix associated with each selected adaptive filter then is regularized as needed. The regularized adaptive filter then is used to remove the echo of the multi-channel playback signal from a captured signal. Regularization is implemented in a round robin manner to ensure that each subband is selected so that the adaptive filter for that subband can be examined. Other features of the multi-channel echo cancellation system and method include dynamic switching between monaural and multi-channel echo cancellation and mixed processing for lower and upper subbands.

Claims

1. A method for canceling echo of a plurality of playback signals in a captured signal, comprising:

decomposing the plurality of playback signals into a plurality of subband signals;

applying a plurality of adaptive filters each having an inverse correlation matrix to the plurality of subband signals;

selecting each of the plurality of adaptive filters in a round robin sequence;

regularizing the inverse correlation matrix of each selected adaptive filter, regularizing the inverse correlation matrix further comprising;

inverting the inverse correlation matrix to produce a correlation matrix;

regularizing the correlation matrix to produce a regularized correlation matrix;

obtaining a regularized inverse correlation matrix by inverting the regularized correlation matrix; and

reducing the echo in the captured signal using the outputs of the adaptive filters.

2. The method as set forth in claim 1, wherein the plurality of adaptive filters comprises recursive least squares (RLS) adaptive filters.

3. The method as set forth in claim 2, further comprising applying the plurality of RLS adaptive filters in parallel.

4. The method as set forth in claim 2, further comprising having a length of the RLS adaptive filters longer in lower subband signals that contain lower frequencies of the plurality of subband signals and shorter in upper subband signals that contain higher frequencies of the plurality of subband signals.

5. The method as set forth in claim 1, wherein regularizing further comprises ensuring that the plurality of adaptive filters are stable.

6. The method as set forth in claim 1, wherein regularizing the inverse correlation matrix further comprises processing the inverse correlation matrix such that a determinant of the correlation matrix is not close to zero.

7. The method as set forth in claim 1, wherein regularizing the correlation matrix further comprises adding a small positive value to diagonal elements of the correlation matrix.

8. The method as set forth in claim 1, wherein regularizing the correlation matrix further comprises:

defining a threshold value;

comparing each diagonal element of the correlation matrix to the threshold value; and

setting the diagonal element equal to the threshold value if the correlation matrix diagonal element is smaller than the threshold value.

9. The method as set forth in claim 1, wherein applying a plurality of adaptive filters comprises:

adapting a plurality of weight vectors corresponding to each of the plurality of adaptive filters;

de-aggregating each of the plurality of weight vectors to produce a plurality of single-channel filter coefficient vectors such that each playback channel has one corresponding single-channel filter coefficient vector; and

applying the plurality of single-channel filter coefficient vectors to the plurality of subband signals.

10. The method as set forth in claim 1, wherein the captured signal comprises an acoustic signal.

11. The method as set forth in claim 1 wherein the captured signal comprises an electrical signal.

12. The method as set forth in claim 1, wherein the plurality of playback signals is a stereo signal.

13. The method as set forth in claim 1, wherein applying a plurality of adaptive filters further comprises identifying,a transfer function of a room in which the echo is present.

14. A computer-readable medium having computer-executable instructions for performing the method recited in claim 1.

15. A computer-implemented method for eliminating echoes of a plurality of playback signals in a captured signal, comprising:

decomposing the plurality of playback signals into subbands;

processing the subbands using a recursive least squares (RLS) adaptive filter for each of the subbands;

selecting each of the RLS adaptive filters in round robin scheme;

regularizing the selected RLS adaptive filter to ensure stability, regularizing the selected RLS adaptive filters further comprising;

computing a correlation matrix of the RLS adaptive filter by inverting an inverse correlation matrix;

regularizing a correlation matrix of the RLS adaptive filter to obtain a regularized correlation matrix;

computing a regularized inverse correlation matrix of the RLS adaptive filter from the regularized correlation matrix; and

reducing echoes in the captured signal using the outputs of the RLS adaptive filters.

16. The computer-implemented method of claim 15, wherein the captured signal comprises an acoustic signal.

17. The computer-implemented method of claim 15, wherein the captured signal comprises an electrical signal.

18. The computer-implemented method of claim 15, wherein regularizing the selected RLS adaptive filter further comprises examining the correlation matrix of the RLS adaptive filter to determine whether regularization is necessary.

19. The computer-implemented method of claim 15, wherein regularizing the correlation matrix further comprises adding a small positive number to each diagonal element of the correlation matrix.

20. The computer-implemented method of claim 15, wherein regularizing the correlation matrix of the RLS adaptive filter further comprises:

defining a threshold value; and

setting a diagonal element of the correlation matrix equal to the threshold value if the diagonal element is smaller than the threshold value.

21. The computer-implemented method of claim 15, wherein computing the regularized inverse correlation matrix comprises inverting the regularized correlation matrix to obtain the regularized inverse correlation matrix.

22. The computer-implemented method of claim 15, wherein the round robin scheme comprises selecting a single RLS adaptive filter for every frame of audio data in the captured signal.

23. The computer-implemented method of claim 15, wherein the round robin scheme comprises selecting a single RLS adaptive filter every several frames of audio data in the captured signal.

24. The computer-implemented method of claim 15, wherein the round robin scheme comprises selecting multiple RLS adaptive filters for every frame of audio data in the captured signal.

25. A method for automatically switching between two different types of echo cancellation processes when processing one or more playback signals and a captured signal containing echo, comprising:

detecting whether the one or more playback signals is monaural;

processing the captured signal and the one or more playback signals using a monaural echo cancellation process if the one or more playback signals are monaural;

processing the captured signal and the one of more playback signals using a multi-channel round robin regularization echo cancellation process if the one or more playback signals are not monaural;

using an estimation technique to estimate a correlation matrix for the multi-channel round robin regularization echo cancellation process, using the estimation technique further comprising one of;

(a) solving a normal equation in a statistical manner using a method of steepest descent when dynamically switching from the multi-channel round robin regularization echo cancellation process to the monaural echo cancellation process;

(b) solving the normal equation deterministically using a least squares process when dynamically switching from the monaural echo cancellation process to the multi-channel round robin regularization echo cancellation process; and

dynamically switching between the monaural echo cancellation process and the multi-channel round robin regularization echo cancellation process whenever the detection changes from one type to another.

26. The method as set forth in claim 25, wherein the monaural echo cancellation process is normalized least mean squares (NLMS) echo cancellation.

27. The method as set forth in claim 25, wherein the monaural echo cancellation process is least mean squares (LMS) echo cancellation.

28. The method as set forth in claim 25, wherein the monaural echo cancellation process and the multi-channel round robin regularization echo cancellation process share state variables.

29. The method as set forth in claim 25, wherein the captured signal contains lower captured subbands and upper captured subbands and further comprising processing the upper captured subband using a normalized least mean squares (NLMS) process.

30. A process for canceling echoes of a plurality of playback signals in a captured signal, the captured signal comprising lower captured subbands containing lower frequencies of the captured signal and higher captured subbands containing higher frequencies of the captured signal and the plurality of playback signals comprising lower playback subbands containing lower frequencies of the plurality of playback signals and higher playback subbands containing higher frequencies of the plurality of playback signals, the process comprising:

using a first echo cancellation process to cancel echoes from the upper captured subbands; and

using a multi-channel round robin regularization echo cancellation process to cancel the echoes in the lower captured subbands, the multi-channel round robin regularization echo cancellation process further comprising;

applying a plurality of adaptive filters, which are recursive least square (RLS) adaptive filters and each of the plurality of adaptive filters comprising an inverse correlation matrix, to each of the lower playback subbands;

selecting each of the plurality of adaptive filters in a round robin sequence;

regularizing the inverse correlation matrix of each selected adaptive filter;

computing each correlation matrix from an inverse correlation matrix.

31. The process of claim 30, wherein the first echo cancellation process is a normalized least mean squares (NLMS) process.

32. The process of claim 30, wherein regularizing further comprises adding a small, positive value to diagonal elements of the correlation matrix.

33. The process of claim 30, wherein regularizing the correlation matrix further comprises:

defining a threshold value;

comparing each diagonal element of the correlation matrix to the threshold value; and

setting the diagonal element equal to the threshold value if the correlation matrix diagonal element is smaller than the threshold value.

34. The process of claim 30, wherein regularizing the correlation matrix further comprises ensuring that the plurality of adaptive filters remain stable.