Telephonic communications – Echo cancellation or suppression – Residual echo cancellation
Reexamination Certificate
1999-09-17
2004-09-14
Tieu, Binh (Department: 2643)
Telephonic communications
Echo cancellation or suppression
Residual echo cancellation
C379S406010, C379S406020, C379S406090, C370S286000, C370S289000
Reexamination Certificate
active
06792106
ABSTRACT:
BACKGROUND OF THE INVENTION
Echo cancellation is a very good way to remove undesirable echo in communication systems.
FIG. 1
illustrates a block diagram of a transmission network employing a conventional echo canceller. The echo canceller
200
is connected to a digital network
100
and a hybrid
230
via unidirectional paths
203
and
205
. The dashed lines in connections
203
and
205
are used to indicate that the connections may be of a sufficient length to cause an echo signal to be subjectively annoying. Such an echo originates at hybrid
230
which is connected to a telephone S
2
, via path
202
.
A widely used adaptation algorithm for updating the coefficients vector in the echo canceller
200
is illustrated in FIG.
2
. This adaptation algorithm is a normalized LMS (NLMS) algorithm
210
given by:
h
k
+
1
=
h
k
+
a
&LeftDoubleBracketingBar;
x
k
&RightDoubleBracketingBar;
2
·
e
k
·
x
k
,
with
(
1
)
e
k
=
x
k
T
·
(
g
-
h
k
)
+
n
k
(
2
)
d
k
=
x
k
T
·
g
+
n
k
,
(
3
)
where h
k
is the echo canceller coefficient-vector
212
, x
k
is the input vector, e
k
is the residual echo, d
k
is the reference echo, 0<&agr;<2 is a scalar that controls the stability and convergence rate, ∥x
k
∥
2
is the norm of the input vector x
k
, g is the actual echo path coefficient vector
214
, and n
k
is additive noise (or near-end speech in the case of double talk).
The choice of &agr; is critical for good performance of the echo canceller
200
. A small &agr; value will ensure small misadjustment in the steady state and is needed for noise insensitivity. However, a small &agr; value brings about low convergence speed. A large &agr; value will in general provide faster convergence and better tracking capabilities at the cost of higher excess mean-squared error in the steady state. As a result, the appropriate choice of &agr; is a major issue for suitable adaptation in the NLMS algorithm
210
.
In prior literature, great efforts have been made to control automatic adjustment of the convergence rate (including echo path change and double talk detection). In D. L. Duttweiler, “A twelve-channel digital echo canceller,” IEEE Trans. On Communications, Vol. COM-26, No. 5, pp. 647-653, May 1978, the Geigel algorithm is presented for double talk detection. The Geigel algorithm compares the magnitude of the current sample of d
k
with a current value of x
max,k
. If the magnitude of d
k
is least −6 dB higher than x
max,k
, then double talk is determined to be present. The Geigel algorithm is simple and fast. However, when the magnitude of d
k
is −6 dB lower than x
max,k
during double talk, the Geigel algorithm fails to detect the double talk. The Geigel algorithm is also sensitive to near-end noise interference.
Other than the Geigel algorithm, many other methods based on correlation are known initiated for automatic adaptation and double talk detection/echo path change detection. In F. Casco et al., “A variable step size NLMS algorithm,” IEICE Trans. Fundamentals, Vol. E78-A, No. 8, pp. 1004-1009, August 1995, a variable step size LMS FIR adaptive filter algorithm is proposed. In this algorithm, the step size adjustment &agr; is controlled by the correlation between the residual echo e
k
and the reference echo d
k
. However, the effectiveness of this algorithm has only been proven with a white noise input. K. Fujii and J. Ohga, “Double-talk detection method with detecting echo path fluctuation,” Electronics and Communications in Japan, Part 3, Vol. 78, No. 4, pp. 82-93, 1996, proposes using the normalized cross correlation between the reference echo d
k
and the estimated echo
y
k
(
x
k
T
·
h
k
)
(
4
)
for discriminating double talk from the echo path change. With this method, it is assumed that in case of echo path change, the normalized cross correlation between d
k
and y
k
approaches zero while in case of double talk, the corresponding cross correlation approaches one. By observing the cross correlation behavior, double talk can be discriminated from the echo path change. There are, however, two problems with this method. First, when the echo path changes proportionally in magnitude, the normalized cross correlation may approach one or be larger than one, instead of zero. Second, in the case of double talk, the normalized correlation between the near end speech and the estimated echo might approach one (instead of the assumed zero) in a certain sample period, which makes it even harder to use the corresponding cross correlation for differentiation of double talk from echo path change.
As discussed above, there have been numerous attempts to design an effective echo canceller utilizing cross correlation. However, despite numerous efforts to perform adaptive filtering based on cross correlation information, there is no evidence that these adaptive filtering techniques may be successfully applied to methods for an NLMS echo cancellation purposes.
SUMMARY OF THE INVENTION
The present invention is directed to a simple and fast technique which does not use cross correlation information.
The basic idea of the invention is to alternate adaptation in the NLMS algorithm between slow mode, aggressive mode and inhibition mode according to different echo canceller status. During initial conversation, the NLMS algorithm is in an aggressive mode to ensure fast convergence. After convergence, the NLMS algorithm switches to a slow mode for lower residual echo return. Whenever an increase of residual echo caused by either double talk or echo path change is detected, the current adaptive filter coefficients are retained and then updated in the aggressive mode. The decision as to which echo estimation (the response of the retained filter or the response of the aggressively adapted filter) is subtracted from the reference echo for echo cancellation within a short detection delay period is determined according to the Geigel algorithm. If double talk is detected according to the Geigel algorithm within the short period, then the retained filter taps are applied to yield the residual echo. Otherwise, the aggressively adapted filter taps are applied to yield the residual echo. In order to decide more precisely whether double talk occurs or echo path change occurs, within the short period of Geigel detection, the current residual echo produced by the aggressively adapted filter is compared with the residual echo produced by retained filter. If the short-term averaged residual echo yielded by the aggressively adapted filter is always lower than the retained filter output, then it is determined that echo path change has occurred. The system is then updated with the aggressive mode output and switches to aggressive mode adaptation until a new convergence is achieved. Otherwise, it is determined that the residual echo increase is caused by double talk. The current filter taps are then updated with the retained filter taps and the inhibition (adaptation-frozen) mode is chosen and maintained until the error signal (residual echo) again decreases within a range of the pre-change value.
REFERENCES:
patent: 5307405 (1994-04-01), Sih
patent: 5559881 (1996-09-01), Sih
patent: 5646991 (1997-07-01), Sih
patent: 5687229 (1997-11-01), Sih
patent: 5937009 (1999-08-01), Wong et al.
patent: 5951626 (1999-09-01), Duttweiler
patent: 6028929 (2000-02-01), Laberteaux
patent: 6029126 (2000-02-01), Malvar
patent: 6058362 (2000-05-01), Malvar
patent: 6115689 (2000-09-01), Malvar
patent: 6181753 (2001-01-01), Takada et al.
patent: 6181794 (2001-01-01), Park et al.
patent: 6198819 (2001-03-01), Farrell et al.
patent: 6201866 (2001-03-01), Ariyama et al.
patent: 6219418 (2001-04-01), Eriksson et al.
patent: 6507652 (2003-01-01), Laberteaux
patent: 6516063 (2003-02-01), Farrell et al.
patent: 6526141 (2003-02-01), Benesty et al.
patent: 6700978 (2004-03-01), Bershad et al.
D.L. Duttweiler, “A Twelve-Channel Digital Echo Canceler,” IEEE Trans. on Comm., vol. COM-26, No. 5, pp. 647-653, May 1978.
F. Casco et al., “A Variable Step S
Agere Systems Inc.
Tieu Binh
LandOfFree
Echo canceller and method of echo cancellation using an NLMS... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Echo canceller and method of echo cancellation using an NLMS..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Echo canceller and method of echo cancellation using an NLMS... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3192205