Noise-level adaptive residual echo suppressor

• US 6,959,167 B1
• Filed: 08/10/2001
• Issued: 10/25/2005
• Est. Priority Date: 08/11/2000
• Status: Expired due to Term

• Current Assignee: Cerence Inc., Cerence Operating Company (Cerence Inc.), Lernout & Hauspie Speech Products NV (Intel Corporation), General Motors Corporation (Motors Liquidation Co. GUC Trust)
• Original Assignee: ScanSoft, Inc. n/k/a Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Dehandschutter, Wouter
• Primary Examiner(s): D AGOSTA, STEPHEN M
• Application Number:
  US09/927,780
• Time in Patent Office: 1,537 Days
• Field of Search: 455/10, 455 631- 65, 455 6711- 677, 455/423, 455501-506, 455/570, 370286-292, 379/3, 37940601-40616, 375229-236, 708300-304, 381/71.1
• US Class Current: 455/1
• CPC Class Codes: H04B 2203/5483 using coupling circuits
• CPC Class Codes: H04B 2203/5491 using filtering and bypassing
• CPC Class Codes: H04B 3/23 using a replica of transmitted signal in the time domain, e.g. echo cancellers
• CPC Class Codes: H04B 3/56 Circuits for coupling, blocking, or by-passing of signals

Abstract

An improved echo control system has an adaptive controlled echo canceling filter arrangement in which a background filter updates the filter coefficients of an echo shaping filter responsive to a normalized least mean square (NLMS) algorithm. An adaptive control module determines a weighted reference signal for the background filter in which the weight is proportional to a far signal reference and an estimate of the norm of an echo canceller error vector, and inversely proportional to en estimate of a residue of the echo canceller. And a non-linear normalized convergence term is used in the NLMS algorithm.

Claims

1. An improved echo control system including:

an echo-containing near signal input;

an echo canceller, coupled to a far signal reference, producing an echo estimate signal output representative of the echo contained in the near signal;

a signal coupling node, coupled to the near signal input and the echo estimate signal output, producing an echo-canceled signal output having an echo residue;

an echo shaping filter, coupled to the echo-canceled signal output, reducing the echo residue and providing an echo-suppressed signal output, the echo shaping filter having a spectral response determined by filter coefficients; and

a background filter, coupled to;

(a) an error signal representative of the difference between;

(i) the echo canceled signal, and

(ii) a signal representative of background filter spectral response, and

(b) an adaptive control module producing a reference signal output that is a weighted sum of;

(i) the echo-containing signal, and

(ii) the echo canceled signal,

the background filter updating the filter coefficients of the echo shaping filter responsive to a normalized least mean square (NLMS) algorithm;

wherein the improvement comprises;

determining, in the adaptive control module, a reference signal weight for the weighted sum, the weight being proportional to the far signal reference; and

an estimate of the norm of an echo canceller error vector, and inversely proportional to en estimate of a residue of the echo canceller; and

using a non-linear normalized convergence term in the NLMS algorithm.

2. An improved echo control system according to claim 1, wherein the echo canceller is a finite impulse response (FIR) filter.

3. An improved echo control system according to claim 1, wherein the echo shaping filter is a finite impulse response (FIR) filter.

4. An improved echo control system according to claim 1, wherein the background filter is a finite impulse response (FIR) filter.

5. An improved echo control system according to claim 1, wherein the echo canceller error vector is determined as:

Δ

w(k)=w_ep−

w(k)

where Δ

w(k) represents the echo canceller error vector, w_ep represents a physical echo path identified by the echo canceller, and w(k) the echo canceller response.

6. An improved echo control system according to claim 1, wherein the reference signal weight is determined as:

$\mathrm{\alpha <br><br>}<br><br>\left(k\right)=\frac{\mathrm{\beta <br><br>}<br><br><br><br>\mathrm{\Delta <br><br>}<br><br><br><br>w<br><br>\left(k\right)<br><br><br><br>{\stackrel{\_}{x}}_s<br><br>\left(k\right)}{{\stackrel{\_}{e}}_s<br><br>\left(k\right)}$

where α

(k) represents the reference signal weight, β

represents a constant normalizing term, ∥

Δ

w(k)∥

represents an estimate of the norm of the echo canceller error vector, {overscore (x)}_s(k) represents a short-term average magnitude of the far signal reference, and ē

_s(k) represents a short-term average magnitude of the echo canceller residue.

7. An improved echo control system according to claim 6, wherein the echo canceller error vector is determined as:

$\frac{N+N_T}{N_T}<br><br>\underset{i=1}{\overset{N_T}{\sum <br><br>}}<br><br><br><br>w_i<br><br>\left(k\right)<br><br>.$

8. An improved echo control system according to claim 1, wherein the NLMS update algorithm is:

$h<br><br>\left(k+1\right)=h<br><br>\left(k\right)+\frac{\mathrm{\mu <br><br>}}{\mathrm{\zeta <br><br>}+{z<br><br>\left(k\right)}^T<br><br>z<br><br>\left(k\right)}<br><br>z<br><br>\left(k\right)<br><br>e_h<br><br>\left(k\right)$

where h(k) represents the echo shaping filter having an order L_H, z(k) represents a vector representing the L_H most recent values of the reference signal output, e_h (k) represents the error signal, ζ

represents a non-negative constant, and $\frac{\mathrm{\mu <br><br>}}{\mathrm{\zeta <br><br>}+{z<br><br>\left(k\right)}^T<br><br>z<br><br>\left(k\right)}$

represents a normalized convergence coefficient.

9. An improved method of echo control including:

providing an echo-containing near signal input;

producing, with an echo canceller coupled to a far signal reference, an echo estimate signal output representative of the echo contained in the near signal;

producing, with a signal coupling node coupled to the near signal input and the echo estimate signal output, an echo-canceled signal output having an echo residue;

reducing, with an echo shaping filter coupled to the echo-canceled signal output, the echo residue and providing an echo-suppressed signal output, the echo shaping filter having a spectral response determined by filter coefficients; and

providing a background filter, coupled to;

(a) an error signal representative of the difference between;

(i) the echo canceled signal, and

(ii) a signal representative of back ground filter spectral response, and

(b) an adaptive control module producing a reference signal output that is a weighted sum of;

(i) the echo-containing signal, and

(ii) the echo canceled signal,

the background filter updating the filter coefficients of the echo shaping filter responsive to a normalized least mean square (NLMS) algorithm;

wherein the improvement comprises;

determining, in the adaptive control module, a reference signal weight for the weighted sum, the weight being proportional to the far signal reference; and

an estimate of the norm of an echo canceller error vector, and inversely proportional to en estimate of a residue of the echo canceller; and

using a non-linear normalized convergence term in the NLMS algorithm.

10. An improved echo control method according to claim 9, wherein the echo canceller is a finite impulse response (FIR) filter.

11. An improved echo control method according to claim 9, wherein the echo shaping filter is a finite impulse response (FIR) filter.

12. An improved echo control method according to claim 9, wherein the background filter is a finite impulse response (FIR) filter.

13. An improved echo control method according to claim 9, wherein the echo canceller error vector is determined as:

Δ

w(k)=w_ep−

w(k)

where Δ

w(k) represents the echo canceller error vector, w_ep represents a physical echo path identified by the echo canceller, and w(k) the echo canceller response.

14. An improved echo control method according to claim 9, wherein the reference signal weight is determined as:

$\mathrm{\alpha <br><br>}<br><br>\left(k\right)=\frac{\mathrm{\beta <br><br>}<br><br><br><br>\mathrm{\Delta <br><br>}<br><br><br><br>w<br><br>\left(k\right)<br><br><br><br>{\stackrel{\_}{x}}_s<br><br>\left(k\right)}{{\stackrel{\_}{e}}_s<br><br>\left(k\right)}$

where α

(k) represents the reference signal weight, β

represents a constant normalizing term, ∥

Δ

w(k)∥

represents an estimate of the norm of the echo canceller error vector, {overscore (x)}_s(k) represents a short-term average magnitude of the far signal reference, and ē

_s(k) represents a short-term average magnitude of the echo canceller residue.

15. An improved echo control method according to claim 14, wherein the echo canceller error vector is determined as:

$\frac{N+N_T}{N_T}<br><br>\underset{i=1}{\overset{N_T}{\sum <br><br>}}<br><br><br><br>w_i<br><br>\left(k\right)<br><br>.$

.

16. An improved echo control method according to claim 9, wherein the NLMS update algorithm is:

$h<br><br>\left(k+1\right)=h<br><br>\left(k\right)+\frac{\mathrm{\mu <br><br>}}{\mathrm{\zeta <br><br>}+{z<br><br>\left(k\right)}^T<br><br>z<br><br>\left(k\right)}<br><br>z<br><br>\left(k\right)<br><br>e_h<br><br>\left(k\right)$

where h(k) represents the echo shaping filter having an order L_H, z(k) represents a vector representing the L_H most recent values of the reference signal output, e_h (k) represents the error signal, ζ

represents a non-negative constant, and $\frac{\mathrm{\mu <br><br>}}{\mathrm{\zeta <br><br>}+{z<br><br>\left(k\right)}^T<br><br>z<br><br>\left(k\right)}$

represents a normalized convergence coefficient.