IJCER (www.ijceronline.com) International Journal of computational Engineering research

International Journal Of Computational Engineering Research (ijceronline.com) Vol. 2 Issue. 4

Echo Cancellation by Adaptive Combination of Nsafs Adaptedbystocastic
Gradient Method
Rekha Saroha [1], Sonia Malik [2], Rohit Anand [3]
[1] [2]
Department of Electronics and Communication Engineering, NCCE, Panipat, INDIA
[3]
Assistant Professor, Electronics and Communication Engineering, NCCE, Panipat, INDIA

Abstract
In acoustic echo cancellation the highly correlated speech input signal and very large impulse response path of echo signal will
slow down the convergence rate of adaptive filters if fullband adaptive filter is used. To solve these problems subband adaptive
filters are used. Adaptive combination methods provide an interesting way to improve adaptive filter’s performance. A tradeoff
between fast convergence rate and small steady state mean square error (MSE) in adaptive combination is achieved by stochastic
gradient algorithm. The individual each filter is independently adapted by it’s own error signal while combination is adapted by
sum of squared subband errors as the cost function. Adaptive combination of normalized sub band adaptive filters is used. In our
proposed method, adaptive combination of sub band signals before going to the adaptive filters. Experimental results show that
the combination method can obtain both fast convergence rate and small steady state MSE by using less number of adaptive
filters.

General Terms: Acoustic Echo Cancellation, subband adaptive filters, echo return loss enhancement.
Keywords: Round Trip Delay (RTD), Acoustic Echo Cancellation (AEC), Normalized Least Mean Square Algorithm (NLMS),
Mean Square Error (MSE), Subband Adaptive Filters (SAF), Adaptive Combination Normalized Subband Adaptive Filters
(NSAF) ,signal to noise ratio (SNR).

1. Introduction
The history of echo cancellation begins on 10th July 1962.In telephones and teleconferencing; a reflection can occur where there
is an impedance mismatch. If the reflected signal reaches the far end subscriber with a RTD of a few milliseconds then it is
perceived as reverberator. If RTD exceeds a few tens of milliseconds the reflection is known as distinct echo [1].
Echo suppressor was used to remove echo which introduces a very large transmission loss in return path. A new technique that
did not interrupt the echo return path called echo cancellation. The AEC estimates the characteristics of the echo path and
generates a replica of the echo. The echo is then subtracted from the received signal. Adaptive filters are used as echo canceller.
The normalized least-mean-square (NLMS) algorithm is one of the most popular adaptive filters. Speech input signal of the
adaptive filter is highly correlated and the impulse response of the acoustic echo path is very long.These two characteristics will
slow down the convergence rate of the acoustic echo canceller. So sub band adaptive filtering is used to solve these problems. In
SAFs, each sub band uses an individual adaptive sub filters. Recently, an adaptive combination of fullband adaptive filters has
been proposed in [8], and its mean-square performance has been analyzed in [9]. More recently, a combination of SAFs for AEC
has been proposed [10], which is based on a conventional sub band structure. In this paper we propose a new scheme for adaptive
combination of subband adaptive filters deal with the tradeoff problem encountered in AEC which are implemented by NSAFs.
The NSAF can be viewed as a subband generalization of the NLMS based adaptive filter. In the proposed combination, mixing
parameter that controls the combination is adapted by means of a stochastic gradient algorithm which employs the sum of squared
subband errors as the cost function.
Section II represents the Simulink model for fullband adaptive filter acts as acoustic echo cancellation [2].Section III represents
sub band adaptive filter and adaptive combination of normalized sub band adaptive filters. Section IV improved adaptive
combination of normalized subband adaptive filters. Section v represents Experiment results.
Notation: Symbols in uppercase letters are used for matrices and in lower cases are used for vectors. Other notations are as
follows: (.) T represents transpose,E(.) is expectation, || ⋅|| represents the Euclidean vector, and diag(�) stands for a diagonal
matrix.
2. Simulink Model Of Fullband Adaptive Filter Acts As Acoustic Echo Cancellation
A block diagram of AEC is shown in Fig.1 [2].

Issn 2250-3005(online) August| 2012 Page 1079


Fig. 1: Block diagram of fullband adaptive filter acts as AEC
Speech signal originating from loudspeaker is received by microphone passing through acoustic echo path. The acoustic echo is
removed by adaptive filters. The d (n) signal contains the speech signal and noise signal. The goal of the adaptive filter w (n) is to
produce a replica of the echo signal y (n). y(n) can be used to cancel the echo by subtracting it from the microphone signal d(n)
resulting in error free signal e(n)[3].Algorithm for AEC is as follows [2]:
Adjustable tap weights can be expressed as:
..(1)

Input signal can be expressed as

U (n) = …. (2)

The output signal y(n) of adaptive filter is the multiplication of w(n) and u(n).

Y (n) = = …. (3)

The error signals difference between the desired response d (n) and filter response y (n) is expressed as:

e (n)=d(n)- u(n)……..(4)

3. Simulink Model of Nsaf and Its Adaptive Combination
3.1 Simulink Model of NSAF
The block diagram of normalized subband adaptive filter is shown in fig. 2[5,6].

Fig. 2: Block diagram of NSAF



The input speech signal u (n) and desired output d(n) are decomposed into N spectral bands using analysis filters. Analysis
filtering is then performed in these sub-bands by a set of independent filters (h0 (n), h1 (n),…, hM-1(n))[2].The sub band signals
are further processed by individual adaptive sub filters Wi(z).Each sub band is computing error signal e(n). By updating the tap
weights, minimizes the sub band error signal. The full band error signal e(n) is finally obtained by interpolating and recombining
all the subband error signals using a synthesis filter bank. The updating equation of NSAFs is written as follows [3]:
+ …..(5)

Where i=1,2,…N-1. Where
,M is the length of the adaptive filter wˆ (k) ,μ is the step-size, and δ is the
regularization parameter[3].

3.2 Simulink Model of Adaptive Combination of NSAFs
The block diagram of adaptive combination of normalized subband adaptive filters is shown in fig.3[9].

Fig. 3: Adaptive combination of NSAFs

A large step size yields a fast convergence rate but also a large steady state MSE [7].To achieve fast convergence rate and small
steady state MSE, adaptive combination of subband adaptive filters is done .So that large step sizes adaptive filters give fast
convergence rate and small step sizes adaptive filters give small steady state MSE.So idea becomes to adapt different step sizes
filters independently and combination is carried out by using a mixing parameter lambda.

The input signal is U (n)= , weight vectors are .
So output becomes y(n)= . The update eq. of sub band adaptive filter is given in eq. 5.

+ Where

= d(n)- (n) and = d(n)- (n) and

, . Consider , then adaptive filter has faster convergence rate and large steady
state MSE whereas has slower faster convergence rate but small steady state MSE.So our purpose is to get large
convergence rate and small steady state MSE ,So combine both adaptive filters. The output of overall filter is:
Y(n)= (n)+[1- ] (n)……….(6) where is mixing parameter. The overall filter with tap weight factor of the form is:
w(n)= (n)+ [1- ] (n).For adaptation of mixing parameter ,use stocastic gradient method to minimize error of
overall filter .



However instead of directly adjusting (n), we will adapt a variable that defines (n) as a sigmoidal function. (n)=sgm
[α(n)]= …….(7)

The update eq. for α(n) is given as:

α(n+1)=α(n)+µ e(n) [ (n)- (n)] (n)[1- (n)]………..(8)

where µ is the step size for adapting α(n).since the mixing parameter is defined by the sigmoidal function[15],which insists the
mixing parameter to lie exactly inside the interval (0,1), this combination is convex combination.[14][3].

4. Improved Adaptive Combination Of Normalized Subband Adaptive Filters
In improved adaptive combination of normalized subband adaptive filters,we take the following assumption.

1) d(n) and u(n) are related by a linear regression model d(n)= (n)u(n)+

for some unknown weight vector of length M and where is an independent distributed noise.

2) The initial condition (0), (0) and a(0) are independent of u(n),d(n) or all n.

3)

A large convergence rate and small steady state MSE can be greatly achieved by using less amount of adaptive filters as
comparison to the previous adaptive combination method. This can be achieved by using the idea of adaptive combination of
speech input signal before going to the NSAF.

5. Experimental Results
The full-band and sub-band systems, adaptive combination of subband adaptive filters and its improvement were modeled in
Matlab Simulink and many simulations for different inputs and number of sub-bands were performed. For the adaptive algorithm
several different algorithms can be used, but the most common one is the normalized least mean squares (NLMS). The order of
the NLMS filters was chosen from N=64 to N=2 . The designs were made in Matlab-Simulink environment and the simulations
were run for 5000 samples for Gaussian noise and sine wave input, respective 12*104 samples in the case of speech input. A
reverberating effect was added to the input by an artificial Schroeder I reverberator which contained four comb filters in parallel
and two all-passes filters series connected. The first estimation of a system capability is represented by the (output error-voice
input), but in order to measure its potential, Echo Return Loss Enhancement (ERLE) should be computed; it is defined as the ratio
of the power of the desired signal over the power of the residual signal. Comparison between full band, sub bands, adaptive
combination, and their improvement is done based on SNR and MSE and by using output error- voice input and ERLE.

6. Conclusion
The NSAF is a good candidate for implementing acoustic echo cancellers because of its fast convergence rate. However, it
requires a tradeoff between fast convergence rate and small steady-state MSE. This paper presented an adaptive convex
combination of two NSAFs to solve this problem. In addition to the conventional coupling update method for component filters,
we also proposed a coupling update mechanism which requires less number of adaptive filters as than used in conventional
method. To verify the effectiveness of the proposed scheme, simulations using different input signals as well as system noises
with different SNRs were performed. The experimental results demonstrated that the proposed scheme can obtain improved
performance as compared to the conventional NSAF.
7. Acknowledgments
With a deep sense of gratitude and heartiest honor, I would like to express my immense thanks to Mr. Rohit Anand, Assistant
Professor, N.C College of Engineering, Panipat for their valuable and sustained guidance, constant encouragement and careful
supervision during the entire course which made the project successful. I proudly acknowledge my sincere and heartfelt thanks to



Ms. Sonia Malik, Department of N.C College of Engineering, Panipat for their valuable help, constant encouragement and
careful supervision during the entire course which made the project successful.
References
[1] M. M. Sondhi, “The history of echo cancellation,” IEEE Signal Process.Mag., vol. 23, no. 5, pp. 95–98, Sep. 2006.
[2] K. A. Lee, W. S. Gan, and S. M. Kuo, Subband Adaptive Filtering: Theory and Implementation. Hoboken, NJ: Wiley,
2009.
[3] Jingen Ni and Feng Li ,”adaptive combination of subband adaptive filters for acoustic echo cancellation.” IEEE
Transactions on Consumer Electronics, Vol. 56, No. 3, August 2010
[4] Irina Dornean, Marina Topa, Botond Sandor Kirei, Marius Neag, “Sub-band adaptive filtering for acoustic echo
cancellation.”2009.
[5] B. Farhang-Boroujeny, Adaptive Filters: Theory and Applications, New York: John Wiley & Sons, Inc., 1998.
[6] J. J. Shynk, ‘Frequency domain and multirate adaptive filtering filtering’, IEEE Signal Processing Mag., 9, January 1992,
14–37.
[7] N. J. Bershad, J. C. M. Bermudez and J.-Y. Tourneret, “An affine combination of two LMS adaptive filtersTransient
mean-square analysis,” IEEE Trans. Signal Process., vol. 56, no. 5, pp. 1853–1864, May 2008.
[8] J. Arenas-García, M. Martínez-Ramón, A. Navia-Vázquez, and A. R. Figueiras-Vidal, “Plant identification via adaptive
combination of transversal filters,” Signal Process., vol. 86, pp. 2430–2438, 2006.
[9] J. Arenas-Garcia, A. R. Figueiras-Vidal, and A. H. Sayed, “Mean-square performance of a convex combination of two
adaptive filters,” IEEE Trans. Signal Process., vol. 54, no. 3, pp. 1078–1090, Mar. 2006.
[10] L. A. Azpicueta-Ruiz, A. R. Figueiras-Vidal, and J. Arenas-García, “Acoustic echo cancellation in frequency domain
using combinations of filters,” in 19th Int. Congress on Acoustics (ICA), Madrid, Sep. 2007.
[11] J. Arenas-García, V. Gómez-Verdejo, M. Martínez-Ramón, and A. R. Figueiras-Vidal, “Separate-variable adaptive
combination of LMS adaptive filters for plant identification,” in Proc. 13th IEEE Int. Workshop Neural Networks Signal
Processing, Toulouse, France, 2003, pp. 239–248.
[12] J. Arenas-García, M. Martínez-Ramón, V. Gómez-Verdejo, and A. R. Figueiras-Vidal, “Multiple plant identifier via
adaptive LMS convex combination,” in Proc. IEEE Int. Symp. Intel. Signal Processing, Budapest, Hungary, 2003, pp.
137–142.
[13] A. H. Sayed, Fundamentals of Adaptive Filtering. New York: Wiley,2003.
[14] N. J. Bershad, J. C. M. Bermudez and J.-Y. Tourneret, “An affine combination of two LMS adaptive filters —Transient
mean-square analysis,” IEEE Trans. Signal Process., vol. 56, no. 5, pp. 1853–1864, May 2008.
[15] K. A. Lee and W. S. Gan, “Improving convergence of the NLMS algorithm using constrained subband updates,” IEEE
Signal Process. Lett., vol. 11, no. 9, pp. 736–739 , Sep. 2004.
[16] P. P. Vaidyanathan, Multirate Systems and Filter Banks. Englewood Cliffs, NJ: Prentice-Hall, 1993.
[17] S. Haykin, Adaptive Filter Theory. Upper Saddle River, NJ: Prentice- Hall, 2002.


IJCER (www.ijceronline.com) International Journal of computational Engineering research

Related slideshows

More Related Content

IJCER (www.ijceronline.com) International Journal of computational Engineering research