Printer Friendly

Design and implementation of a novel echo-canceller for reduction of acoustic noise from environmental system.


Nowadays the acoustic echo problems related to different communication applications is increased. The presence of acoustic echo creates disturbance in communication systems and degrade the voice quality during the voice communication. The Acoustic Echo is a type of noise signal which is reflected signal or delayed form of simple original signal with minimized amplitude. This eventually creates the communication disturbance for many applications i.e. video-conferencing, tele-conferencing and etc. To overcome these acoustic echo problems, Echo Cancellation system must be used in communication applications. Acoustic Echo Cancellation [2] and [6] eliminates or minimize unwanted echo noise signal and improves the quality of audio communication systems. Figure 1 elaborates the echo generation during the conversation.

Most of the advance echo cancellation system uses adaptive filtering techniques. Digital adaptive filter are also used for echo cancellation for maximum noise reduction in spite of environmental changes. In adaptive filtering two types of digital filter structures can be used [5]: Infinite Impulse Response (IIR) [9]--[11] and Finite Impulse Response (FIR) structures [12]-[15]. The performance of these echo cancellation system is totally depends upon the effectiveness of the adaptive filtering algorithm and its parameters. In the case of acoustic echo cancellation, the optimal output of the adaptive filter is equal in value to the unwanted echoed signal. When the adaptive filter output is equal to desired signal the error signal will become zero and echoed signal would be completely cancelled and the far user would not hear any echo noise during communication.

The adaptive filter self-adjusts its weights according to an optimization algorithm affected by an error signal. Adaptive filter algorithms try to find the optimal value for time-variant signals using feedback in the form of an error signal to refine its weights. Basic block diagram of adaptive Echo Canceller is presented in figure 2. In this paper a detail analysis is perform on LMS algorithm [1] based adaptive filter, by varying its parameters for different inputs such as voice, noise, echo and their combinations.

There are many adaptive algorithms that could be used in echo cancellation for adaptive filtering including wiener filter, steepest descent method, least mean square, normalized least mean square and recursive least squares [7] and [8], However, a generic block diagram of any adaptive filtering is given as figure 3.

In this diagram x (n) is the input signal, d (n) is the desired signal, y(n) is the output of the filter ande (n) is the error signal which is the difference of the d (n)&y(n) i.e.

e (n) = d (n)- y(n) (1)

The input signal x[n] = (x (n), ... ... ... ... , x(n--l)) produces output signal as:

y (n) = x (n) * w(n) (2)

Where, w[n] = ([w.sub.0] (n),... ... ... ..., [w.sub.l] (n)) are filter co- efficient or weights.

Wiener stated the following condition for finding the optimal weights of the filter:

[nabla]= 0 (3)

i.e. the gradient of estimated square error equal to zero. The optimal weight will be:

[w.sub.0] (n) = p[R.sup.-1] (4)

Here, p and R are the cross and auto co-relation matrices, their values are;



p = [[p(0) ... p(1 -m)].sup.h] (6)

The challenge of attaining the values of auto correlation and cross correlation can be subsided by steepest descent method. This algorithm is based on the local iterative descent method, which starts from an initial guess of filter weight w(0) and generates a sequence of weight vector w (1) , w (2) ,... ... ... ... by iterating the algorithm and reduce the function J (w), as;

J(w(n +1))< J(w(n)) (7)

Where w (n) the previous value of the weight vector and w(n + 1) is its updated value.

The steepest decent algorithm is described as,

w(n + 1) = w(n)- 1/2 [micro]g(n) (8)

Where ndenotes the iteration, [micro]is called as the step-size parameter.

So, the wiener solution for filter weight by use of steepest descent method produces: w (n + 1) = w (n) + [mu][p - Rw(n)] (9)

In (9), the adjustments [delta]w(n) applied to the tap- weight vector at timen + 1 is equal to [mu] [p - Rw (n)].


Wiener and steepest descent can give optimal solution if (i) the exact measurements of the gradient Vector [nabla]J at each iterationnis known and (ii) the exact suitable value of the step-size parameter [mu] is known. However, practically exact measurements of the gradient vector are not possible, since it would require prior knowledge of both the correlation matrixfland cross-correlation vectorp.

Least mean square Algorithm provides an effective method to compute the optimal solution in terms of least mean square. The LMS algorithm uses an iterative search for the Wiener solution on a quadratic surface with a simplified gradient. It uses instantaneous error [e.sup.2](n) instead of estimated error E[[e.sup.2](n)] as the cost function. Now, the coefficient or weight w(n) updating equation for LMS algorithm can be defined as: [??] (n + 1) = [??] (n) + [mu]e* (n) x(n) (10)

For adaptation of the weight vectorw(n + 1), LMS algorithm requires knowledge of most recent weight vector w(n), input vector x(n) and instantaneous error signal e(n).


For the simulation purpose the three different input models are developed which is used as a input to adaptive acoustic Echo Canceller system. The three input simulation models are voice and noise, Voice and echo, voice-noise and echo which are illustrated in figures 5, 6 and 7 respectively.

For generation of figure 5 model uniform white noise is added to the original voice signal, Echo signal in figure 6 model is generated by using delay element and then adds it with voice signal and in last model, echo signal and uniform white noise is added with original voice signal.

2.1 Applying LMS A Igorithm to AEC System:

The block diagram of the LMS algorithm based AEC Control System is presented in Figure 8. It basically comprises of two processes, filtering process and adaptive process. For the simulation and detail analysis three different sets of input are applied. In first stage input signal goes to digital filter to produce its outputy(n), this output signal combined with desired output d(n) and generates an error signal, this is called process of filtering. The input signal x(n) and the error signal e(n) are then fed to the least mean square algorithm block for updating of weight coefficients of digital filter, this is called adaptation process.


For thorough investigation of LMS algorithm in acoustic echo cancellation application the important parameters of LMS algorithm i.e filter length, step size [3] and sampling time are varied and analyzed for different set of input signal which are original voice signal, Voice with echo signal, Voice with Noise signal and Voice with noise-echo signal. For the simulation purpose uniform white noise with amplitude of 2 is used as a noise signal. These all input signals applied to LMS based AEC system are shown in figures 9, 10, 11 & 12 respectively.

3.1 Varying the Parameters of LMS for AEC Application:

All three input models voice--noise, voice--echo and voice-echo-noise input signals are applied on LMS based AEC system one by one. Each input model is analyzed by varying three parameters that are filter length, step size and the sampling time for achieving the most optimal outcome. Table 1 shows the set of values selected for each parameter of the models.

3.2 Varying the parameters of Voice -Noise input model:

For experiment purpose first Voice-Noise input model is applied on LMS based AEC system. Different values of its parameters are tested to check the behavior of the error signal between the outputs of discrete FIR system and the LMS block. Experiments are performed and output signal are analyzed for different values of filter lengths, different step size (H.C. Huang, 2012) and for different sampling time, figure 13 is the output error signal for filter length 16 and 32.

The output error signal has been observed for various step sizes. Some sample simulations are presented in figures 14 for step size 0.01, 0.015 and 0.016.

Similarly, figure 15 presents the output error signal simulations for sampling time of 8K, 8.8K, and 10K From these simulation results, it can be observed that voice--noise input model produces most optimum results at parameters, filter length = 32, step size = 0.015 and sampling time is 10k.

3.3 Varying the parameters of Voice- Echo input model:

Like voice and noise input module experiments is performed for Voice and Echo input by changing the LMS algorithm parameters. The output signal for filter lengths 32 and 64 are shown in figure21 and 22, outputs for different step sizes are shown in figures 23 to 25 and error signals for different sampling times can be seen in figures 26 to 28. The best results for voice and echo input model are find for filter length = 32, step size = 0.05 and sampling time is 8.8k.

3.4 Varying the parameters of Echo-Voice-Noise input model:

The third model for LMS is of voice and noise; we put different values and check the error signal between the outputs of discrete FIR system and the LMS block same as previous described manner. Different outputs error signal for different filter lengths, step sizes and sampling time are shown in figures from 29 to 36. For voice, echo and noise input model the best results were achieved when set filter length = 32, step size = 0.025 and sampling time is 8.8k.

3. Conclusions:

It has been observed that for designing of environmental friendly echo-canceller if different parameters are applied to all attributes, classification accuracy yields best results. The proposed method is tested with differently correlated input data sets such as voice-noise, voice-echo and voice-echo-noise datasets, showing effectiveness of the system in controlled simulation environment. Meanwhile, collected results may be assessed and perhaps managed to generate a generic echo-canceller that can work with various noise and echo levels in order to create a database that may lead to development of novel environmental friendly echo-noise canceller.


Article history:

Received 12 November 2014

Received in revised form 31 December 2014

Accepted 22 January 2015

Available online 25 February 2015


[1] Bershad, N.J., J. Carlos, M. Bermudez and J. Tourneret, 2008. Stochastic Analysis of the LMS Algorithm for System Identification With Subspace Inputs, IEEE Transaction on Signal Processing, 56(3): 1018-1027.

[2] Krishna, E.H., M. Raghuram, K. Venu Madhav and K. Ashoka Redd, 2010. Acoustic Echo Cancellation using a Computationally Efficient Transform Domain LMS Adaptive Filter, 10th International Conference on Information Science, Signal Processing and their Applications (ISSPA 2010): 409-412.

[3] Homana, I., M. Topa, K. Botond, C. Contan, 2011. Technical University of Cluj-Napoca, Str. C. Daicoviciu Nr. 15, 400020 Cluj-Napoca, Romania,2011 .Behavior of Variable Step Size NLMS Algorithms for Acoustic Echo Cancellation, IEEE 10th International Symposium on Signals, Circuits and Systems (ISSCS 2011): 1-4.

[4] Huang, H.C. and J. Lee, 2012. A New Variable Step-Size NLMS Algorithm and Its Performance Analysis, Signal Processing, 60(4): 2055-2060.

[5] Haque, M.A., A.K.M. K. Islam, M.I. Islam, 2009. Evaluation of Performances of Digital Adaptive Filters in Acoustic Echo Cancellation, Proceedings of 2009 12th International Conference on Computer and Information Technology (ICCIT 2009): 215-219.

[6] Hwang, S.S., S.C. Kim and C.D. Lee, 2010. Real Time Acoustic Echo Cancellation Technique Based on Normalized Least Mean Square Method, Proceeding of 2010 2nd International Conference on Computer Engineering and Technology (ICCET 2010): V6-12--V6-16.

[7] Fu, J. and W.P. Zhu, 2008. A Nonlinear Acoustic Echo Canceller Using Sigmoid Transform in Conjunction with RLS Algorithm, IEEE Transaction on Circuit and Systems-II: Express Briefs, 55(10): 1056-1060.

[8] Paleologu, C., J. Benesty and S. Ciochin, 2008. A Robust Variable Forgetting Factor Recursive LeastSquares Algorithm for System Identification, IEEE Signal Processing Letters, 15: 597- 600.

[9] Azim, C.F., A.R. Memon and S.F. Ahmed, 2011. On the optimal design of IIR filter with its state-space realization, Indian J. Science & Technology, 4(9): 1021-1024.

[10] Yongfeng, Z., F.Panguo, Z. Jun, 2013. Adaptive IIR filtering based on balanced-realization, Journal of System Engineering and Electronics, 24(1): 19-25.

[11] Ferro, H.F., L.F.O. Chamon, C.G. Lopes, 2014. FIR-IIR adaptive filters hybrid combination, Electronics Letters, 50(7): 501-503.

[12] Ahmed, S.F., C.F. Azim and A.R. Memon, 2011. Minimization of the excessive noise for broad-band active noise feed-forward control system, International Journal of Academic Research, 3(2): 1034-1039.

[13] Mehima, A., A.N. Willson, 2015. Optimal factoring of FIR filters, IEEE Transactions on Signal Processing, 63(3): 647-661.

[14] Ahmed, S.F., C.F. Azim, A.R. Memon and H. Desa, 2011. Global optimal solution for active noise control Problem, Indian J. Science & Technology, 4(9): 1015-1020.

[15] Lai, X., Z. Lin, 2014. Optimal design of Constrained FIR filters without phase response specifications, IEEE Transactions on Signal Processing, 62(17): 4532-4546.

(1) Choudhry Fahad Azim, (2) Syed Faiz Ahmed, (3) Mohd.Izhar A. Bakar, (2) Kushairy Abdul Kadir

(1) Hamdard University, HamdardInstitute of Engineering and Technology, Karachi- -74600, Pakistan

(2) University Kuala Lumpur, British Malaysian Institute, (UniKL BMI), Batu 8, JlnSg. Pusu, Selangor 53100, Malaysia

Corresponding Author: Dr. Choudhry Fahad Azim, Hamdard University, Hamdard Institute of Engineering and Technology, Faculty of Engineering Science and Technology, Karachi Pakistan.


Table 1: LMS parameters value set.

                                                 Voice, Noise
LMS Parameters    Voice & Noise   Voice & Echo     and Echo

Filter Length          16              32             16
                       32              64             32

Step Size             0.01           0.045          0.020
                      0.015          0.050          0.025
                      0.016          0.055          0.030

Sampling Times        8 KHz          8 KHz          8 KHz
                     8.8KHz          8.8KHz         8.8KHz
                     10 KHz          10 KHz         10 KHz
COPYRIGHT 2015 American-Eurasian Network for Scientific Information
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2015 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Azim, Choudhry Fahad; Ahmed, Syed Faiz; Bakar, Mohd. Izhar A.; Kadir, Kushairy Abdul
Publication:Advances in Environmental Biology
Article Type:Report
Date:Feb 1, 2015
Previous Article:Consanguinity and homogamous marriages among Chaouis, a Berber population from Eastern Algeria.
Next Article:"The evaluation of developing techniques of altruism and martyrdom culture".

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters