Telephonic communications – Echo cancellation or suppression – Using digital signal processing
Reexamination Certificate
1999-08-24
2004-02-03
Tieu, Binh (Department: 2747)
Telephonic communications
Echo cancellation or suppression
Using digital signal processing
C379S406110, C370S290000
Reexamination Certificate
active
06687373
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates generally to telecommunications. More specifically, the invention relates to echo cancellation in data, voice and other communication systems.
2. Description of the Related Art
The phenomenon of“echo” in telephone and voice carrier systems results from leakage back to the source of electrical signals as they approach their destination. Such echo is due to impedance mismatches between source and destination and/or reflections of the signals from components such as speakers. One classic occurrence of such echo is the “line echo” experienced in the conversion of four-wire systems (public networks) to two-wire systems (such as telephone wiring in a home).
For example, as shown in
FIG. 1
, two terminals
110
and
120
, such as telephones, which subsist on two-wire networks connect to each other over a four-wire network
130
. To facilitate the needed conversion, an analog device known as a hybrid is inserted between a four-wire network and the two-wire network. Thus, a hybrid
115
interconnects terminal
110
with the four-wire network
130
and a hybrid
125
interconnects terminal
120
with the four-wire network
130
. Typically, hybrids
115
and
125
will consist of a pair of induction coupled transformers that a duplex signal into two simplex signals. Owing to the leakage current in each hybrid, a part of the signal energy that is transmitted will b reflected back to the source. Thus, terminal
110
would hear an echo of its own outgoing transmission (such as a voice echo) reflecting from its interception at hybrid
125
(which is attached to terminal
115
).
To improve the quality of such transmissions, line echo cancellers are usually placed in the network. As shown in
FIG. 2
, a Line Echo Canceller (LEC)
200
is placed as close as possible to the source of the echo (in terms of the echo received by terminal
210
) proximate to a hybrid
225
. Terminal
210
sends a signal that is received as y(n), and serves as one input to LEC
200
. A composite signal y(n) entering the network as a result of a terminal
220
(and its hybrid
225
) is the sum of an echo “r(n)” of the x(n) signal and the transmitted signal t(n) from terminal
220
. In a typical line echo canceller, such as LEC
200
, an adaptive transversal FIR (Finite Impulse Response) filter is used to predict the echo based on previous samples. LEC
200
, for instance, outputs an estimated echo signal r_est(n).
The filter is optimized by measuring and feeding back an error signal e(n)=y(n)-r_est(n). Based upon the goal of reducing the mean-squared error, the filter's coefficients are updated. One such updating scheme is known as LMS (Least Mean Square) which can be represented in discrete terms by:
a
k
(
i
+
1
)
=
a
k
(
i
)
+
2
β
*
1
M
∑
m
=
0
M
-
1
e
(
i
-
,
)
*
y
(
i
-
m
-
k
)
,
where “M” is the number of discrete samples, “ak” is a coefficient of the filter.
One critical coefficient in the above formula is the Beta (&bgr;). The Beta is the loop gain factor that determines the convergence performance over time of the adaptive filter. The larger the Beta factor, in general, the more rapidly will the performance of the filter converge toward an optimum cancellation result. However, if the Beta is chosen too large then the filter performance may begin diverging at a certain point (FIG.
3
(
a
)). A lower Beta may not converge until after a long time, as shown in FIG.
3
(
b
), but will have less asymptotic error and thus, can be prevented from diverging off. Additionally, an input signal with a low power may cause slower convergence. In such a case the Beta factor can be normalized by the average power of the input signal converting the LMS algorithm into NLMS (Normalized Least Mean Square). Even with NLMS, the choice of Beta is difficult to resolve.
Thus, there is a need for a Beta discovery scheme that permits an adaptive filter to most optimally converge without risk of divergence or slowing due to low power in signals.
One other issue in the design of the filter is in choosing the number of taps used. When calculating the estimate of the echo signal, the filter coefficient at each tap is multiplied by the reference signal when the reference signal passes the tap. These products are accumulated together and represent the estimated echo. Therefore, the greater the number of taps, the more multiply-and-accumulate instructions need to be processed or implemented. This increases the cost and decreases the speed of the filter in estimating the echo and, consequently, in canceling
Additionally, there is a need for determining the most optimal number of filter taps so that excess computation is reduced.
SUMMARY OF THE INVENTION
What is disclosed is a method, apparatus and system employing such for adapting the Beta factor in a echo cancellation adaptive filter by using echo loss return enhancement (ERLE) as a metric to determine how large or small the Beta should be set to. In general, a high ERLE indicates that a low Beta factor should be used, while a low ERLE indicates that a high Beta factor should be used.
Also disclosed is a method and apparatus for determining the optimum number of filter taps used in an adaptive filter by dividing a filter array into blocks of taps, finding for each tap the sum of squares of the coefficients therein and then comparing the sum to an average block energy. Based on this comparison, a block is deemed to either contribute significantly to the operation of the filter or not. Starting with the tail end block of the array, the optimum filter order is found by determining when the first contributing block is found.
REFERENCES:
patent: 5473686 (1995-12-01), Virdee
patent: 5631899 (1997-05-01), Duttweiler
patent: 5631900 (1997-05-01), McCaslin et al.
patent: 5664011 (1997-09-01), Crochiere et al.
patent: 5687229 (1997-11-01), Sih
patent: 5757906 (1998-05-01), Unno
patent: 5764753 (1998-06-01), McCaslin et al.
patent: 5805696 (1998-09-01), Im et al.
patent: 5812537 (1998-09-01), Betts et al.
patent: 5905717 (1999-05-01), Hasegawa
patent: 5923749 (1999-07-01), Gustafsson et al.
patent: 5933797 (1999-08-01), Haakansson et al.
patent: 5953658 (1999-09-01), Scott
patent: 5987143 (1999-11-01), Okuno et al.
patent: 6195430 (2001-02-01), Eriksson et al.
patent: 6212273 (2001-04-01), Hemkumar et al.
patent: 6282176 (2001-08-01), Hemkumar
patent: 6434110 (2002-08-01), Hemkumar
Song Wei-Jei
Yeh Chiang
Blakely , Sokoloff, Taylor & Zafman LLP
Tieu Binh
LandOfFree
Heusristics for optimum beta factor and filter order... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Heusristics for optimum beta factor and filter order..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Heusristics for optimum beta factor and filter order... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3336416