IRJET-New Technique for Recursive Least Square Adaptive Algorithm for Acoustic Echo Cancellation of Speech signal in an auditorium
Content
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
New Technique for Recursive Least Square Adaptive Algorithm
for Acoustic Echo Cancellation of Speech signal in an auditorium
Praveen.N 1, S.Ranjitha2 Dr .H. N. Suresh3
Research scholar, Dept. Of E&I, BIT, Under VTU, Belgaum India,
BE(ECE), Bangalore Institute of Technology,vv puram ,Bangalore-04
Professor, Bangalore Institute of Technology, Dept.of Elecronics and Instrumentation Engg., Bangalore–04
1
2Final
3
Abstract: In today’s technological society, human
computer interactions are ever increasing. In many new
systems, voice recognition platforms are implemented
to give users more convenient ways of operating
equipment and systems. To improve the audibility of the
speech, the noise and acoustic echo must be removed
from the speech signal. In this paper, we presented a
new adaptive algorithm in the frequency domain for
acoustic echo cancellation of speech signal in an
auditorium. The RLS algorithm, the forgetting factor
remains constant, which is utilized for the stability of
the adaptive algorithm. However, the constant value of
the forgetting factor will not support for the sensitive
system. The value of the forgetting factor depends on
the echo and reverberation. In an auditorium speech,
the echo and reverberation signals are not in a stable
manner since the constant value of forgetting factor is
not a perfect solution for the removing the echo and
reverberation. In order to solve this problem we
presented average recursive least square adaptive
algorithm, which produces the flexible forgetting factor
in a min-max manner. The estimated echo values are
constructed with the aid of combined feature of the minmax manner, which leads to increase the quality of the
speech signal. Finally, our proposed algorithm is
implemented using MATLAB and the experimental
results showed that the proposed ARLS algorithm
outperformed than the existing RLS algorithm.
Keywords: frequency domain for acoustic echo
cancellation, adaptive filter, recursive least
square, average recursive least square,
reverberation.
1.
INTRODUCTION
The acoustic echo, which is well-known as a
“multipath echo”, is formed by poor voice coupling
between the earpiece and microphone in handsets and
© 2015, IRJET
hands-free gadgets. Additional voice degradation is caused
as voice-compressing and encoding/decoding devices
process the voice paths within the handsets and in
wireless networks. This gives returned echo signals with
highly variable properties. At the point when compounded
with inherent digital transmission delays, call quality is
incredibly reduced for the wireline caller. Acoustic
coupling is because of the reflection of the loudspeaker’s
sound waves from walls, door, ceiling, windows and other
different objects back to the microphone. The aftereffect of
the reflections is the formation of a multipath echo and
multiple harmonics of echoes, which are transmitted back
to the far-end and are heard by the talker as an echo unless
wiped out. Adaptive cancellation of such acoustic echoes
has turned out to be critical in hands-free communication
systems such as teleconference or video conference
systems [1-11].
Echo signal is the delayed type of original speaker
signal. That implies, echo signal can be expected as a noise
in speaker signal. The eliminating of noise from the
speaker signal cannot be executed by classical filters,
which suppress certain frequency parts and pass the
others. This is the reason that, filter design used to
eliminate echo is the subject of optimal filter design. The
essential reason for the optimal filter design is to minimize
the dissimilarity between desired response and actual
response of the filter. Filter response does not just rely on
the statistical information; because physical signal’s
statistical information has usually a changing nature.
Consequently, a filter structure, which adjusted its
response, according to the change of the error signal, is
essential to adapt filter coefficients in a manner to
minimize error signal [8]. Adaptive filter is the answer to
this issue. Adaptive filter is a filter with coefficients, which
are adjusted periodically keeping in mind the end goal to
attempt meeting some performance criterion, which is
normally in the form of some error or cost function
minimization [9, 11]. An adaptive filter is a digital filter
that can alter its coefficients to give the best match to a
ISO 9001:2008 Certified Journal
Page 1065
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
given desired signal. At the point when an adaptive filter
works in a changeable environment, the filter coefficients
can adapt in response to changes in the applied input
signals. The main task of the adaptive filter is to estimate
the characteristics of the echo path, creating the echo and
compensate for it. To do this the echo path is viewed as an
unknown system with some impulse response and the
adaptive filter must mimic this response. Adaptive filters
have been utilized as a part of different parts of signal
processing in recent years. Among the possible
applications is the Acoustic Echo Cancellation [11,12].
recursive Bayesian estimator that takes the form of an
adaptive Kalman algorithm in the discrete Fourier
transform (DFT) domain, has been derived. The paper has
also demonstrated that such a recursive estimator
acknowledged by means of a stable and structurally
proficient multichannel state-space frequency-domain
adaptive filter. The paper has additionally shown the
proposed algorithm, which comes from a contained
structure, gave successful nonlinear echo cancellation in
the vicinity of continuous double-talk, fluctuating degree of
nonlinear distortion, and changes in the echo path.
Adaptive Filters are usually actualized in the time
domain, which functions admirably in many scenarios on
the other hand; in numerous applications, the impulse
response turns out to be too long, increasing the
complexity of the filter beyond a level where it can no
longer be implemented efficiently in the time domain.
Then again, there exists an alternate solution and that is to
actualize the filters in the frequency domain. The Discrete
Fourier Transform or more precisely the Fast Fourier
Transform (FFT) permits the conversion of signals from
the time domain to the frequency domain in an efficient
manner [12,13].
Luis A et al. [15] have introduced a new method
for nonlinear acoustic echo cancellation based on adaptive
Volterra Filters with linear and quadratic kernels, that
mechanically choosed those diagonals contributing most
to the output of the quadratic kernel with the objective of
minimizing the overall mean-square error. In the echo
cancellation scenarios, not all coefficients were similarly
relevant for the modeling of the nonlinear echo, but
coefficients close to the main diagonal of the second-order
kernel depict the majority of the nonlinear echo
distortions, such that not all diagonals need to be executed.
Then again, that was hard to choose the most suitable
number of diagonals apriori, since there have numerous
elements that effect the decision, for example, the energy
of the nonlinear echo, the shape of the room impulse
response, or the step size utilized for the adjustment of
kernel coefficients. The proposed method includes
adaptive scaling components that control the impact of
every group of adjacent diagonals contributing to the
quadratic kernel output. Zoran M. Šari´c et al. [16] have
proposed a computationally proficient form of the
partitioned block frequency domain adaptive filter with
many iterations on current data block. The algorithm
executed as a cascade of two adaptive filters. The first filter
minimized the Least Square (LS) criteria leading to
unbiased estimate of a room response. The second filter
accelerates the convergence rate utilizing many iterations
to minimize adjusted LS criterion. Coefficients upgrades
computed in a single step substitute for several iterations
and cut computational costs. The difficulty of the algorithm
is o(log2(R)), where R had a number of iterations. The
proposed algorithm has been tested in a simulated room
and a real reverberant room. Luis A. Azpicueta-Ruiz et al.
[17] have presented an AEC based on combination of
filters in discrete Fourier transform domain. Considering
that both the input signal and the cancellation scenario
make the performance of adaptive filters was frequency
dependent, the proposed method have exploited the
combination capabilities employing different mixing
parameters to separately combine
2. RELATED WORKS:
Yüksel Özbay et al. [11] have presented an
algorithm for the determination of optimal adaptation rate
(μ) for the least-mean-square (LMS) adaptation algorithm
that has been utilized in the adaptive filter. The efficiency
of their optimal μ value determination algorithm has been
demonstrated on a single direction voice conference
application with one speaker.
A DSP card
(TMS320C6713), a Laptop computer, an amplifier, a
loudspeaker and two microphones in the two applications
has been utilized. In the first application, two microphones
had placed close to the loudspeaker, while in the other
application, one microphone had placed close to
loudspeaker and speech trial had been implemented in the
far-end microphone. Output of the adaptive filter has been
observed for μ values of 0, 0.1, 100 and optimal (a value
between 0.01 and 100). The best outcomes in the adaptive
filter had been achieved from optimal μ value.
Sarmad Malik and Gerald Enzner [14] have
discussed about the adaptive acoustic echo cancellation in
the vicinity of an unknown memory less nonlinearity
preceding the echo path. Through absorbed the
coefficients of the nonlinear expansion into the unknown
echo path, the cascade observation model had been altered
into an equal multichannel structure, which further
increased with a multichannel first-order Markov model.
For the subsequent multichannel state-space model, a
© 2015, IRJET
ISO 9001:2008 Certified Journal
Page 1066
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
proposed effective acoustic echo cancellation methodology
for an auditorium.
The figure 3.1 is represented the overall block
diagram of the proposed acoustic echo canceller. In this
figure 1, the acoustic echo cancellation in an auditorium is
illustrated. The speech signal with the reverberation of
voice and auditorium noise is collected by the microphone
and collected speech signal is passed to the speaker. The
problem in this audio setup is that the passed voice signal
is played
Figure 3.1: Block Diagram for acoustic echo canceller
independent spectral regions of two frequencydomain adaptive filters with different step sizes. Thusly,
the proposed method outclassed recent algorithms where
only a single combining parameter mixes the overall
outputs of two frequency-domain adaptive filters. These
advantages were shown by means of realistic experiments.
3. PROPOSED METHODOLOGY:
3.1 Acoustic echo cancelation in auditorium
An echo is a reflection of sound, arriving at the
listener sometime after the direct sound. Echo is the
reflected replica of the voice heard eventually later and
deferred version of the original. Echo cancellation is the
procedure that eliminates unwanted echoes from the
original signal. It incorporates first recognizing the
originally transmitted signal that re-shows up, with some
deferral, in the speech signal. When the echo is accepted, it
can be removed by 'subtracting' it from the speech signal.
Numerous reflections in acoustic enclosures and
transmission delay affect the sound quality, which on
account of a teleconferencing system lead to a poor
understanding of the conversation.
3.2 Acoustic problems in auditorium
The assembly room, as a spot for listening created
from the classical open-air theaters. The outline of
different sorts of auditoriums has turn into a mind
boggling issue, because in addition to its different,
sometimes conflicting, aesthetics, functional, technical,
artistic and economical requirements, an auditorium
regularly needs to suit a remarkably large audience. In a
few ways, even the largest hall is same as the smaller
rooms, the essential acoustic criteria are the same. On the
other hand, the primary defects of the auditorium
conferencing are reverberation and echo. Keeping in mind
the end goal to take care of this issue, in this paper, we
© 2015, IRJET
through loudspeaker and its reflections of the
room boundaries will also collected by the microphones
and passed to the speaker. This makes listener hear the
repeated voice with delayed reflections of the auditorium
walls. The presence of acoustic echo in the auditorium
makes the listeners feel that they are being interrupted
with the repeated voice, forcing them to stop speaking
until the echo faded away and the process is repeated over
and over again. This acoustic echo and reverberation
degrades the quality of the communication considerably.
3.3 Adaptive filters for acoustic echo cancellers in
frequency domain adaptive filter
The fundamental function of the AEC is to
estimate the acoustic transfer function from the speakers
to the microphone including the reflections way. Filtering
the incoming voice signal through the evaluated acoustic
transmission function delivers an estimate of the echo
signal y(n). Subtracting this evaluated echo from the
microphone signal results in the echo free signal e(n)=
d(n)-y(n) which is send to speaker rather than the
microphone signal d(n). Acoustic echo cancellers typically
utilize adaptive finite impulse response filters to assess the
acoustic echo path. The FIR coefficients are adjusted
utilizing an adaptive algorithm to minimize the error
signal. The figure 3.2 represents the block diagram of the
adaptive filter. The adaptive filter is indicated in the dotted
box of the figure 2, which contains the two portions
specifically filter part and update part. The function of the
filter part is to compute the convolution of the input signal
Sout and the filter coefficients resulting in the filter output y
(n). The set of filter coefficients are constantly adjusted by
the update part. The update part is additionally called as
adaptive algorithm, which is responsible for updating the
filter coefficients so that the filter output y (n) turns out to
be as close as possible to the desired signal d (n). In most
cases update part changes the filter coefficients in small
steps to minimize a certain function of the error signal
e(n). The error signal e(n) represents the difference
between the desired signal d(n) and the filter output i.e.
e(n) = d(n)-y(n).
ISO 9001:2008 Certified Journal
Page 1067
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
Figure3.2: Block Diagram of Average RLS Filter
S
out
Adjustabl
e filter
Adaptive
algorithm
y
(n)
e
(n
)
+ d
(n
)
Frequency-domain adaptive filtering is an attractive
solution to deal with this difficult problem. There are two
principal
advantages
to
frequencydomain
implementations of adaptive filters. First the amount of
computation can be greatly reduced by repla- cing timedomain convolution and/or correlation by fast transform
domain block-convolution and/or block cor- relation
based either on the fast Fourier transform (wr). The
second advantage comes from the decorrelating property
of the discrete Fourier transform and the possibility of
using different step sizes for each transform domain
adaptive weight, which results in a quasi-optimal
convergence rate, even in the presence of large variations
in the input power spectrum (a situation where timedomain LMS-type algorithms perform very poorly).
In frequency domain adapative filter, both
filtering and coefficient update can be performed sampleper-sample or n blocks of sample. A block of L sample are
collected in a buffer and the adaptive filter function is
called to process the whole buffer resluting in L output
samples and updating all the filter coefficient every bufferfull samples. In block processing case,it is possible to
perform the filtering and coefficient update functions
entirely in frequency-domain. This is achieved by first
applying the fourier transformation on the data buffer and
performing the filtering and update by complex element
ise multipplication in the frequency domain. The result is
then converted back to time domain using the inverse
fourier transform. This procedure results in a very efficient
implementation of large adaptive filters, suchas those
commonly used in acoustic echo canccellers.
4. PROPOSED AVERAGE RECURSIVE LEAST SQUARES
METHOD
4.1 Recursive least squares adaptive filter
The Recursive least squares (RLS) is an adaptive filter
which recursively finds the coefficients that minimize a
weighted linear least squares cost function relating to the
© 2015, IRJET
e-ISSN: 2395-0056
p-ISSN: 2395-0072
input signals. This is rather than different calculations, for
example, the least mean squares (LMS) that goal is to
decrease the mean square error. In the derivation of the
RLS, the input signals are considered deterministic, while
for the LMS and similar algorithm they are viewed as
stochastic. Contrasted with most of its competitors, the
RLS exhibits extremely fast convergence. As specified the
previously the memory of the RLS algorithm is restricted
to a limited number of values, relating to the order of the
filter tap weight vector. Firstly, two factors of the RLS
implementation ought to be noted: the first is that in spite
of the fact that matrix inversion is crucial to the derivation
of the RLS algorithm, no matrix inversion calculations are
needed for the execution, hence significantly reducing the
amount of computational complexity of the algorithm.
Secondly, unlike the LMS based algorithms, current
variables are updated within the iteration they are to be
utilized, utilizing values from the previous iteration. To
implement the RLS algorithm, the following steps are
executed in the following order.
1.
The filter output is calculated using the filter tap
weights from the previous iteration and the
current input vector.
y n1 (n) w t (n 1) x(n) …
2.
(1)
The intermediate gain vector is calculated using
eq. (2).
1
u(n) (n 1) X (n)
1
k ( n)
u ( n) …
T
X (n)u (n)
(2)
3.
The estimation error value is calculated using eq.
(3).
en 1 (n) d (n) y n 1 (n) …
(3)
4.
The filter tap weight vector is updated using eq.
(4) and the gain vector is calculated in eq. (2).
w(n) w T (n 1) k (n)en1 (n) …
5.
(4)
The inverse matrix is calculated using eq. (5).
1 (n) 1 ( 1 (n 1) k (n)[ X T (n) 1 (n 1)])
(5)
From the above description of the RLS algorithm,
where the forgetting factor remains constant value which
is laid between zero and one. Selecting the value of the
ISO 9001:2008 Certified Journal
Page 1068
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
forgetting factor is a based on the following condition. The
smaller value of the forgetting factor is, the smaller
contribution of previous samples. This makes the filter
more sensitive to recent samples, which means more
fluctuations in the filter co-efficients. The case is referred
to as the growing window RLS algorithm. In practice, is
usually chosen between 0.98 and 1. The constant value of
the forgetting factor is used for stability of the adaptive
algorithm. However, the constant value of the forgetting
factor will not support for the sensitive system. The value
of the forgetting factor is depends on the echo and
reverberation. In an auditorium speech, the echo and
reverberation signals are not in a stable manner since the
constant value of forgetting factor is not suitable for this
application. This problem motivated us to design a RLS
algorithm with flexible forgetting factor.
4.2 Average RLS estimation
In our proposed methodology, we introduce the
average recursive least square adaptive algorithm in the
frequency domain for effective acoustic echo cancellation.
In a standard RLS algorithm, the value of forgetting factor
placed remains constant. In the case of error of the signal
is larger sensitivity of the adaptive algorithm needs to be
increase. The sensitivity of the RLS algorithm depends on
the forgetting factor. By decreasing the value of the
forgetting factor, the sensitivity of the RLS adaptive
algorithm is increased. In our research, we discuss the
problem acoustic echo cancellation in an auditorium. In
the auditorium, the speech signal is affected by the both
echo and reverberation signal since the error value
become larger. In order to remove the larger error in this
paper, we designed novel average recursive lease square
(ARLS) adaptive algorithm.
From the above figure 1, represents the proposed
method of acoustic echo cancellation, from which the
adaptive filer (i) and adaptive filter (j) are presented
where ‘i’ and ‘j’ has the minimum and maximum value of
forgetting factor values respectively. Here, we used the
average recursive least square as adaptive filter. According
to that, the adaptive filters process the input signal mixed
with the echo and reverberation and produces the
estimated echo. Our proposed algorithm selects the
combined estimated echo of the both adaptive filer (i) and
adaptive filter (j) which, leads to reduce the error value.
e-ISSN: 2395-0056
p-ISSN: 2395-0072
1
1 (n) 2 ( 21 (n 1) k (n)[ X T (n) 21 (n 1)])
(7)
2
Calculate Average filter value for RLS filter,
1 (n) = Avg ( 1 (n) , 1 (n) );
Avg
1
(8)
2
5. SIMULATION AND RESULTS
This paper presents a details sketch of an Acoustic
Echo canceller, (AEC). The software simulation and the
results of simulation of the ARLS-AEC algorithm, which
was performed in MATLAB, are discussed. The proposed
Average recursive least square adaptive algorithm in
frequency domain is implemented in MATLAB Version
8.1.0.604 (R2013a). The system on which the technique
was simulated was having 4 GB RAM with 64 bit operating
systems having i5 Processor. For assessment of the
proposed method, randomly generated signals has been
used.
In order to evaluate the quality of the echo
cancellation algorithm the measure of ERLE was used.
ERLE, measured in dB is defined as the ratio of the
instantaneous power of the signal, d(n), and the
instantaneous power of the residual error signal, e(n),
immediately after cancellation. ERLE measures the amount
of loss introduced by the adaptive filter alone.
Mathematically it can be stated as
ERLE = 10log Pd (n) /Pe (n) = 10log E[d (n)]^2/ E[e (n)]^2
For a good echo canceller circuit, an ERLE in the
range of 30 dB – 40dB is considered to be ideal. The Table
1 shows comparison between existing and proposed
methods ERLE values in the range 30-40 dB
Signals
1
2
ERLE
Existing
Existing
method with method with
forgetting
forgetting
factor 0.98
factor 0.90
0.5638
0.5612
13.0075
12.7446
Proposed
method with
average
forgetting
factor
0.0884
0.0861
Table: 5.1 ERLE comparison
a.Comparison between original signal and
estimated signals in RLS Algorithm in max
forgetting factor
Let’s consider two system in parallel, based on forgetting
factor
1
1 (n) 1 ( 11 (n 1) k (n)[ X T (n) 11 (n 1)])
(6)
1
© 2015, IRJET
ISO 9001:2008 Certified Journal
Page 1069
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
Figure 5.6: Estimated error
Figure 5.1: Input near end and far end signal for
echo cancelation Fs = 8000
Figure 5.7: Weightage Curve for existing system for
forgetting factor 0.90
Figure 5.2: Actual and Estimated output for input
signal
Figure 5.3: Estimated error
Figure 5.4: Weightage Curve for existing system for
forgetting factor 0.98
Comparisons graphs from figure 5.1-5.4:The
Estimated curves are obtained by varying forgetting factor
lamda. The Echo and reverberation both are maintain in
stable condition. The comparative curves have plotted
between actual and estimated signal. Figure 5.2 gives
estimated curves for stable echo and reverberation in
standard time limit. Figure 5.3 gives estimated error in
employed auditorium for sample N = 8000. From
analysing the results, we can infer that all the cases gave
good results. Among the Weightage curves, the distance
between the curves are high in stable situation it
represents this forgetting factor lamda gives only dilute
weightage values.
a. Comparison between original signal and
estimated signals in RLS Algorithm in min
forgetting factor
Figure 5.5: Actual and Estimated output for input
signal
© 2015, IRJET
Comparisons graphs from figure 5.5- 5.7: The
Estimated curves are obtained by varying forgetting factor
lamda into minimum (0.90). The Echo and reverberation
both are maintain in stable condition. The comparative
curves have plotted between actual and estimated signal.
Figure 5.6 gives estimated curves for stable echo and
reverberation in standard time limit and minimum lamda
= 0.90. Figure 5.7 gives estimated error in employed
auditorium for sample N = 8000 and lamda = 0.90. From
analysing the results, we can infer that all the cases gave
worst results. Among the Weightage curves, the distance
between the curves are high in stable situation it
represents this forgetting factor lamda gives only dilute
weightage values. Moreover the estimated and actual
signals are not matched in any condition.
b. Comparison between RLS Algorithm in minmax forgetting factor
Figure 5.9: Estimated error
Figure 5.10: Weightage Curve for proposed system
forgetting factor as a average
ISO 9001:2008 Certified Journal
Page 1070
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
6. CONCLUSION
A new algorithm was proposed for an acoustic
echo canceller with average RLS algorithm. Its
performance was studied in comparison with conventional
algorithms in a simulation. Good performance was
confirmed with the proposed algorithm. Furthermore, a
parallel echo cancelling architecture suitable for hardware
implementation by frequency domain transfer processing.
Near end signal, Far end signal echo and reverberation in
auditorium was gradually optimized using average RLS
filters by changing forgetting factor. The proposed system
is stable, when echo and reverberation is high. Finally, the
relationship between the echo cancellation algorithm the
measure ERLE and signal and Weightage of error signal
characteristics was clarified.
7 .REFERENCES
[1] Moura, Jose MF, James L. Flanagan, and Nikil S. Jayant,
"The Discipline of Signal Processing [Reflections]", IEEE
Signal Processing Magazine, Vol. 30, No. 6, pp. 174-176,
2013.
[2] Breining, C, Dreiscitel, P, Hänsler, E, Mader, A, Nitsch, B,
Puder, H, & Tilp, J, “Acoustic echo control. An application of
very-high-order adaptive filters”, IEEE Signal Processing
Magazine, Vol.16, No.4, pp.42-69, 1999.
[3] F.S. Hallack, M.R. Petraglia, Performance comparison of
adaptive algorithms applied to acoustic echo cancelling, In
the proceedings of 2003 IEEE International Symposium on,
vol. 2, pp. 1147–1150, June 2003.
[4] E. Haensler, “The hands-free telephone problem: an
annotated
bibliography
update”,
Annales
Des
Télécommunications, Vol. 49, No. 7-8, pp. 360-367, 1994.
[5] C. Breining, P. Dreiscitel, E. Hansler, A. Mader, B. Nitsch,
H. Puder, T. Schertler, G. Schmidt, J. Tilp, “Acoustic echo
control. An application of very-highorder adaptive”, IEEE
Signal Process. Mag., Vol. 16, No. 4, pp. 42–69, 1999.
[6] G. Enzner and P. Vary, “A soft-partitioned frequencydomain adaptive filter for acoustic echo cancellation”, In
the proceedings of IEEE International Conference on
Acoustics, Speech, and Signal Processing, Vol. 5, pp. 393–
396, 2003.
[7] K. Eneman, M. Moonen, “Iterated partitioned block
frequency-domain adaptive filtering for acoustic echo
cancellation”, IEEE Transaction on Speech Audio Process,
Vol. 11, No. 2, pp. 143-158, 2003.
[8] Sheng Wu, Xiaojun Qiu and Ming Wu, "Stereo Acoustic
Echo
Cancellation
Employing
Frequency-Domain
Preprocessing and Adaptive Filter", IEEE Transactions on
© 2015, IRJET
e-ISSN: 2395-0056
p-ISSN: 2395-0072
Audio, Speech, and Language Processing, vol. 19, no. 3,
pp.614-623, 2010.
[9] Fukui, M, Shimauchi, S., Hioka, Y., Nakagawa, A., and
Haneda, Y., "Double-talk robust acoustic echo cancellation
for CD-quality hands-free videoconferencing system", IEEE
Transactions on Consumer Electronics, IEEE Transactions
onConsumer Electronics, Vol. 60, no. 3, 2014.
[10] Bekrani, M, Khong, A.W.H. and Lotfizad, M., "A
Clipping-Based Selective-Tap Adaptive Filtering Approach
to Stereophonic Acoustic Echo Cancellation", IEEE
Transactions on Audio, Speech, and Language Processing,
Vol. 19 , No. 6, 2011.
[11] Yüksel Özbay, AhmetRe¸sit Kavsao˘ glu, “An optimum
algorithm for adaptive filtering on acoustic echo
cancellation using TMS320C6713 DSP”, Digital Signal
Processing, Vol. 20, No. 1, pp. 133-148, 2010.
[12] K. G Gunale, S. N. Motade, S.L. Nalbalwar, S. B.
Deosarkar, “Frequency Domain Adaptive Filter using FFT
Algorithm for Acoustic Echo Cancellation”, In the
proceedings of 3rd International Conference on Emerging
Trends in Engineering and Technology (ICETET), pp. 582587, 2010.
[13] Gil-Cacho, J.M, van Waterschoot, T. , Moonen, M. ,
Jensen, S.H, "A Frequency-Domain Adaptive Filter (FDAF)
Prediction Error Method (PEM) Framework for DoubleTalk-Robust Acoustic Echo Cancellation", IEEE/ACM
Transactions on Audio, Speech, and Language Processing,
Vol. 22 , No. 12, 2014.
[14] Sarmad Malik and Gerald Enzner, "State-Space
Frequency-Domain Adaptive Filtering for Nonlinear
Acoustic Echo Cancellation", IEEE Transactions on audio,
speech, and language processing, Vol. 20, No. 7, pp. 20652079, September 2012.
[15] Luis A. Azpicueta-Ruiz, Aníbal R. Figueiras-Vidal,
Aníbal R. Figueiras-Vidal, Jerónimo Arenas-García,
"Enhanced Adaptive Volterra Filtering by Automatic
Attenuation of Memory Regions and Its Application to
Acoustic Echo Cancellation", IEEE transactions on signal
processing, Vol. 61, No. 11, pp. 2745- 2750, June 2013.
[16] Zoran M. Šari´c, IstvanI.Papp, DraganD.Kukolj, Ivan
Veliki´c, Gordana Veliki´c, "Partitioned block frequency
domain acoustic echo canceller with fast multiple
iterations", Digital Signal Processing, Vol. 27, pp. 119–128,
2014.
[17] Luis A. Azpicueta-Ruiz, Anibal R. Figuieras-Vidal and
Jeronimo Arenas-Garcia, "Acoustic echo cancellation in
discrete Fourier transform domain based on adaptive
combination of adaptive filters", In the proceedings of
Meetings on Acoustics, Vol. 19, 2013.
ISO 9001:2008 Certified Journal
Page 1071
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
New Technique for Recursive Least Square Adaptive Algorithm
for Acoustic Echo Cancellation of Speech signal in an auditorium
Praveen.N 1, S.Ranjitha2 Dr .H. N. Suresh3
Research scholar, Dept. Of E&I, BIT, Under VTU, Belgaum India,
BE(ECE), Bangalore Institute of Technology,vv puram ,Bangalore-04
Professor, Bangalore Institute of Technology, Dept.of Elecronics and Instrumentation Engg., Bangalore–04
1
2Final
3
Abstract: In today’s technological society, human
computer interactions are ever increasing. In many new
systems, voice recognition platforms are implemented
to give users more convenient ways of operating
equipment and systems. To improve the audibility of the
speech, the noise and acoustic echo must be removed
from the speech signal. In this paper, we presented a
new adaptive algorithm in the frequency domain for
acoustic echo cancellation of speech signal in an
auditorium. The RLS algorithm, the forgetting factor
remains constant, which is utilized for the stability of
the adaptive algorithm. However, the constant value of
the forgetting factor will not support for the sensitive
system. The value of the forgetting factor depends on
the echo and reverberation. In an auditorium speech,
the echo and reverberation signals are not in a stable
manner since the constant value of forgetting factor is
not a perfect solution for the removing the echo and
reverberation. In order to solve this problem we
presented average recursive least square adaptive
algorithm, which produces the flexible forgetting factor
in a min-max manner. The estimated echo values are
constructed with the aid of combined feature of the minmax manner, which leads to increase the quality of the
speech signal. Finally, our proposed algorithm is
implemented using MATLAB and the experimental
results showed that the proposed ARLS algorithm
outperformed than the existing RLS algorithm.
Keywords: frequency domain for acoustic echo
cancellation, adaptive filter, recursive least
square, average recursive least square,
reverberation.
1.
INTRODUCTION
The acoustic echo, which is well-known as a
“multipath echo”, is formed by poor voice coupling
between the earpiece and microphone in handsets and
© 2015, IRJET
hands-free gadgets. Additional voice degradation is caused
as voice-compressing and encoding/decoding devices
process the voice paths within the handsets and in
wireless networks. This gives returned echo signals with
highly variable properties. At the point when compounded
with inherent digital transmission delays, call quality is
incredibly reduced for the wireline caller. Acoustic
coupling is because of the reflection of the loudspeaker’s
sound waves from walls, door, ceiling, windows and other
different objects back to the microphone. The aftereffect of
the reflections is the formation of a multipath echo and
multiple harmonics of echoes, which are transmitted back
to the far-end and are heard by the talker as an echo unless
wiped out. Adaptive cancellation of such acoustic echoes
has turned out to be critical in hands-free communication
systems such as teleconference or video conference
systems [1-11].
Echo signal is the delayed type of original speaker
signal. That implies, echo signal can be expected as a noise
in speaker signal. The eliminating of noise from the
speaker signal cannot be executed by classical filters,
which suppress certain frequency parts and pass the
others. This is the reason that, filter design used to
eliminate echo is the subject of optimal filter design. The
essential reason for the optimal filter design is to minimize
the dissimilarity between desired response and actual
response of the filter. Filter response does not just rely on
the statistical information; because physical signal’s
statistical information has usually a changing nature.
Consequently, a filter structure, which adjusted its
response, according to the change of the error signal, is
essential to adapt filter coefficients in a manner to
minimize error signal [8]. Adaptive filter is the answer to
this issue. Adaptive filter is a filter with coefficients, which
are adjusted periodically keeping in mind the end goal to
attempt meeting some performance criterion, which is
normally in the form of some error or cost function
minimization [9, 11]. An adaptive filter is a digital filter
that can alter its coefficients to give the best match to a
ISO 9001:2008 Certified Journal
Page 1065
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
given desired signal. At the point when an adaptive filter
works in a changeable environment, the filter coefficients
can adapt in response to changes in the applied input
signals. The main task of the adaptive filter is to estimate
the characteristics of the echo path, creating the echo and
compensate for it. To do this the echo path is viewed as an
unknown system with some impulse response and the
adaptive filter must mimic this response. Adaptive filters
have been utilized as a part of different parts of signal
processing in recent years. Among the possible
applications is the Acoustic Echo Cancellation [11,12].
recursive Bayesian estimator that takes the form of an
adaptive Kalman algorithm in the discrete Fourier
transform (DFT) domain, has been derived. The paper has
also demonstrated that such a recursive estimator
acknowledged by means of a stable and structurally
proficient multichannel state-space frequency-domain
adaptive filter. The paper has additionally shown the
proposed algorithm, which comes from a contained
structure, gave successful nonlinear echo cancellation in
the vicinity of continuous double-talk, fluctuating degree of
nonlinear distortion, and changes in the echo path.
Adaptive Filters are usually actualized in the time
domain, which functions admirably in many scenarios on
the other hand; in numerous applications, the impulse
response turns out to be too long, increasing the
complexity of the filter beyond a level where it can no
longer be implemented efficiently in the time domain.
Then again, there exists an alternate solution and that is to
actualize the filters in the frequency domain. The Discrete
Fourier Transform or more precisely the Fast Fourier
Transform (FFT) permits the conversion of signals from
the time domain to the frequency domain in an efficient
manner [12,13].
Luis A et al. [15] have introduced a new method
for nonlinear acoustic echo cancellation based on adaptive
Volterra Filters with linear and quadratic kernels, that
mechanically choosed those diagonals contributing most
to the output of the quadratic kernel with the objective of
minimizing the overall mean-square error. In the echo
cancellation scenarios, not all coefficients were similarly
relevant for the modeling of the nonlinear echo, but
coefficients close to the main diagonal of the second-order
kernel depict the majority of the nonlinear echo
distortions, such that not all diagonals need to be executed.
Then again, that was hard to choose the most suitable
number of diagonals apriori, since there have numerous
elements that effect the decision, for example, the energy
of the nonlinear echo, the shape of the room impulse
response, or the step size utilized for the adjustment of
kernel coefficients. The proposed method includes
adaptive scaling components that control the impact of
every group of adjacent diagonals contributing to the
quadratic kernel output. Zoran M. Šari´c et al. [16] have
proposed a computationally proficient form of the
partitioned block frequency domain adaptive filter with
many iterations on current data block. The algorithm
executed as a cascade of two adaptive filters. The first filter
minimized the Least Square (LS) criteria leading to
unbiased estimate of a room response. The second filter
accelerates the convergence rate utilizing many iterations
to minimize adjusted LS criterion. Coefficients upgrades
computed in a single step substitute for several iterations
and cut computational costs. The difficulty of the algorithm
is o(log2(R)), where R had a number of iterations. The
proposed algorithm has been tested in a simulated room
and a real reverberant room. Luis A. Azpicueta-Ruiz et al.
[17] have presented an AEC based on combination of
filters in discrete Fourier transform domain. Considering
that both the input signal and the cancellation scenario
make the performance of adaptive filters was frequency
dependent, the proposed method have exploited the
combination capabilities employing different mixing
parameters to separately combine
2. RELATED WORKS:
Yüksel Özbay et al. [11] have presented an
algorithm for the determination of optimal adaptation rate
(μ) for the least-mean-square (LMS) adaptation algorithm
that has been utilized in the adaptive filter. The efficiency
of their optimal μ value determination algorithm has been
demonstrated on a single direction voice conference
application with one speaker.
A DSP card
(TMS320C6713), a Laptop computer, an amplifier, a
loudspeaker and two microphones in the two applications
has been utilized. In the first application, two microphones
had placed close to the loudspeaker, while in the other
application, one microphone had placed close to
loudspeaker and speech trial had been implemented in the
far-end microphone. Output of the adaptive filter has been
observed for μ values of 0, 0.1, 100 and optimal (a value
between 0.01 and 100). The best outcomes in the adaptive
filter had been achieved from optimal μ value.
Sarmad Malik and Gerald Enzner [14] have
discussed about the adaptive acoustic echo cancellation in
the vicinity of an unknown memory less nonlinearity
preceding the echo path. Through absorbed the
coefficients of the nonlinear expansion into the unknown
echo path, the cascade observation model had been altered
into an equal multichannel structure, which further
increased with a multichannel first-order Markov model.
For the subsequent multichannel state-space model, a
© 2015, IRJET
ISO 9001:2008 Certified Journal
Page 1066
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
proposed effective acoustic echo cancellation methodology
for an auditorium.
The figure 3.1 is represented the overall block
diagram of the proposed acoustic echo canceller. In this
figure 1, the acoustic echo cancellation in an auditorium is
illustrated. The speech signal with the reverberation of
voice and auditorium noise is collected by the microphone
and collected speech signal is passed to the speaker. The
problem in this audio setup is that the passed voice signal
is played
Figure 3.1: Block Diagram for acoustic echo canceller
independent spectral regions of two frequencydomain adaptive filters with different step sizes. Thusly,
the proposed method outclassed recent algorithms where
only a single combining parameter mixes the overall
outputs of two frequency-domain adaptive filters. These
advantages were shown by means of realistic experiments.
3. PROPOSED METHODOLOGY:
3.1 Acoustic echo cancelation in auditorium
An echo is a reflection of sound, arriving at the
listener sometime after the direct sound. Echo is the
reflected replica of the voice heard eventually later and
deferred version of the original. Echo cancellation is the
procedure that eliminates unwanted echoes from the
original signal. It incorporates first recognizing the
originally transmitted signal that re-shows up, with some
deferral, in the speech signal. When the echo is accepted, it
can be removed by 'subtracting' it from the speech signal.
Numerous reflections in acoustic enclosures and
transmission delay affect the sound quality, which on
account of a teleconferencing system lead to a poor
understanding of the conversation.
3.2 Acoustic problems in auditorium
The assembly room, as a spot for listening created
from the classical open-air theaters. The outline of
different sorts of auditoriums has turn into a mind
boggling issue, because in addition to its different,
sometimes conflicting, aesthetics, functional, technical,
artistic and economical requirements, an auditorium
regularly needs to suit a remarkably large audience. In a
few ways, even the largest hall is same as the smaller
rooms, the essential acoustic criteria are the same. On the
other hand, the primary defects of the auditorium
conferencing are reverberation and echo. Keeping in mind
the end goal to take care of this issue, in this paper, we
© 2015, IRJET
through loudspeaker and its reflections of the
room boundaries will also collected by the microphones
and passed to the speaker. This makes listener hear the
repeated voice with delayed reflections of the auditorium
walls. The presence of acoustic echo in the auditorium
makes the listeners feel that they are being interrupted
with the repeated voice, forcing them to stop speaking
until the echo faded away and the process is repeated over
and over again. This acoustic echo and reverberation
degrades the quality of the communication considerably.
3.3 Adaptive filters for acoustic echo cancellers in
frequency domain adaptive filter
The fundamental function of the AEC is to
estimate the acoustic transfer function from the speakers
to the microphone including the reflections way. Filtering
the incoming voice signal through the evaluated acoustic
transmission function delivers an estimate of the echo
signal y(n). Subtracting this evaluated echo from the
microphone signal results in the echo free signal e(n)=
d(n)-y(n) which is send to speaker rather than the
microphone signal d(n). Acoustic echo cancellers typically
utilize adaptive finite impulse response filters to assess the
acoustic echo path. The FIR coefficients are adjusted
utilizing an adaptive algorithm to minimize the error
signal. The figure 3.2 represents the block diagram of the
adaptive filter. The adaptive filter is indicated in the dotted
box of the figure 2, which contains the two portions
specifically filter part and update part. The function of the
filter part is to compute the convolution of the input signal
Sout and the filter coefficients resulting in the filter output y
(n). The set of filter coefficients are constantly adjusted by
the update part. The update part is additionally called as
adaptive algorithm, which is responsible for updating the
filter coefficients so that the filter output y (n) turns out to
be as close as possible to the desired signal d (n). In most
cases update part changes the filter coefficients in small
steps to minimize a certain function of the error signal
e(n). The error signal e(n) represents the difference
between the desired signal d(n) and the filter output i.e.
e(n) = d(n)-y(n).
ISO 9001:2008 Certified Journal
Page 1067
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
Figure3.2: Block Diagram of Average RLS Filter
S
out
Adjustabl
e filter
Adaptive
algorithm
y
(n)
e
(n
)
+ d
(n
)
Frequency-domain adaptive filtering is an attractive
solution to deal with this difficult problem. There are two
principal
advantages
to
frequencydomain
implementations of adaptive filters. First the amount of
computation can be greatly reduced by repla- cing timedomain convolution and/or correlation by fast transform
domain block-convolution and/or block cor- relation
based either on the fast Fourier transform (wr). The
second advantage comes from the decorrelating property
of the discrete Fourier transform and the possibility of
using different step sizes for each transform domain
adaptive weight, which results in a quasi-optimal
convergence rate, even in the presence of large variations
in the input power spectrum (a situation where timedomain LMS-type algorithms perform very poorly).
In frequency domain adapative filter, both
filtering and coefficient update can be performed sampleper-sample or n blocks of sample. A block of L sample are
collected in a buffer and the adaptive filter function is
called to process the whole buffer resluting in L output
samples and updating all the filter coefficient every bufferfull samples. In block processing case,it is possible to
perform the filtering and coefficient update functions
entirely in frequency-domain. This is achieved by first
applying the fourier transformation on the data buffer and
performing the filtering and update by complex element
ise multipplication in the frequency domain. The result is
then converted back to time domain using the inverse
fourier transform. This procedure results in a very efficient
implementation of large adaptive filters, suchas those
commonly used in acoustic echo canccellers.
4. PROPOSED AVERAGE RECURSIVE LEAST SQUARES
METHOD
4.1 Recursive least squares adaptive filter
The Recursive least squares (RLS) is an adaptive filter
which recursively finds the coefficients that minimize a
weighted linear least squares cost function relating to the
© 2015, IRJET
e-ISSN: 2395-0056
p-ISSN: 2395-0072
input signals. This is rather than different calculations, for
example, the least mean squares (LMS) that goal is to
decrease the mean square error. In the derivation of the
RLS, the input signals are considered deterministic, while
for the LMS and similar algorithm they are viewed as
stochastic. Contrasted with most of its competitors, the
RLS exhibits extremely fast convergence. As specified the
previously the memory of the RLS algorithm is restricted
to a limited number of values, relating to the order of the
filter tap weight vector. Firstly, two factors of the RLS
implementation ought to be noted: the first is that in spite
of the fact that matrix inversion is crucial to the derivation
of the RLS algorithm, no matrix inversion calculations are
needed for the execution, hence significantly reducing the
amount of computational complexity of the algorithm.
Secondly, unlike the LMS based algorithms, current
variables are updated within the iteration they are to be
utilized, utilizing values from the previous iteration. To
implement the RLS algorithm, the following steps are
executed in the following order.
1.
The filter output is calculated using the filter tap
weights from the previous iteration and the
current input vector.
y n1 (n) w t (n 1) x(n) …
2.
(1)
The intermediate gain vector is calculated using
eq. (2).
1
u(n) (n 1) X (n)
1
k ( n)
u ( n) …
T
X (n)u (n)
(2)
3.
The estimation error value is calculated using eq.
(3).
en 1 (n) d (n) y n 1 (n) …
(3)
4.
The filter tap weight vector is updated using eq.
(4) and the gain vector is calculated in eq. (2).
w(n) w T (n 1) k (n)en1 (n) …
5.
(4)
The inverse matrix is calculated using eq. (5).
1 (n) 1 ( 1 (n 1) k (n)[ X T (n) 1 (n 1)])
(5)
From the above description of the RLS algorithm,
where the forgetting factor remains constant value which
is laid between zero and one. Selecting the value of the
ISO 9001:2008 Certified Journal
Page 1068
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
forgetting factor is a based on the following condition. The
smaller value of the forgetting factor is, the smaller
contribution of previous samples. This makes the filter
more sensitive to recent samples, which means more
fluctuations in the filter co-efficients. The case is referred
to as the growing window RLS algorithm. In practice, is
usually chosen between 0.98 and 1. The constant value of
the forgetting factor is used for stability of the adaptive
algorithm. However, the constant value of the forgetting
factor will not support for the sensitive system. The value
of the forgetting factor is depends on the echo and
reverberation. In an auditorium speech, the echo and
reverberation signals are not in a stable manner since the
constant value of forgetting factor is not suitable for this
application. This problem motivated us to design a RLS
algorithm with flexible forgetting factor.
4.2 Average RLS estimation
In our proposed methodology, we introduce the
average recursive least square adaptive algorithm in the
frequency domain for effective acoustic echo cancellation.
In a standard RLS algorithm, the value of forgetting factor
placed remains constant. In the case of error of the signal
is larger sensitivity of the adaptive algorithm needs to be
increase. The sensitivity of the RLS algorithm depends on
the forgetting factor. By decreasing the value of the
forgetting factor, the sensitivity of the RLS adaptive
algorithm is increased. In our research, we discuss the
problem acoustic echo cancellation in an auditorium. In
the auditorium, the speech signal is affected by the both
echo and reverberation signal since the error value
become larger. In order to remove the larger error in this
paper, we designed novel average recursive lease square
(ARLS) adaptive algorithm.
From the above figure 1, represents the proposed
method of acoustic echo cancellation, from which the
adaptive filer (i) and adaptive filter (j) are presented
where ‘i’ and ‘j’ has the minimum and maximum value of
forgetting factor values respectively. Here, we used the
average recursive least square as adaptive filter. According
to that, the adaptive filters process the input signal mixed
with the echo and reverberation and produces the
estimated echo. Our proposed algorithm selects the
combined estimated echo of the both adaptive filer (i) and
adaptive filter (j) which, leads to reduce the error value.
e-ISSN: 2395-0056
p-ISSN: 2395-0072
1
1 (n) 2 ( 21 (n 1) k (n)[ X T (n) 21 (n 1)])
(7)
2
Calculate Average filter value for RLS filter,
1 (n) = Avg ( 1 (n) , 1 (n) );
Avg
1
(8)
2
5. SIMULATION AND RESULTS
This paper presents a details sketch of an Acoustic
Echo canceller, (AEC). The software simulation and the
results of simulation of the ARLS-AEC algorithm, which
was performed in MATLAB, are discussed. The proposed
Average recursive least square adaptive algorithm in
frequency domain is implemented in MATLAB Version
8.1.0.604 (R2013a). The system on which the technique
was simulated was having 4 GB RAM with 64 bit operating
systems having i5 Processor. For assessment of the
proposed method, randomly generated signals has been
used.
In order to evaluate the quality of the echo
cancellation algorithm the measure of ERLE was used.
ERLE, measured in dB is defined as the ratio of the
instantaneous power of the signal, d(n), and the
instantaneous power of the residual error signal, e(n),
immediately after cancellation. ERLE measures the amount
of loss introduced by the adaptive filter alone.
Mathematically it can be stated as
ERLE = 10log Pd (n) /Pe (n) = 10log E[d (n)]^2/ E[e (n)]^2
For a good echo canceller circuit, an ERLE in the
range of 30 dB – 40dB is considered to be ideal. The Table
1 shows comparison between existing and proposed
methods ERLE values in the range 30-40 dB
Signals
1
2
ERLE
Existing
Existing
method with method with
forgetting
forgetting
factor 0.98
factor 0.90
0.5638
0.5612
13.0075
12.7446
Proposed
method with
average
forgetting
factor
0.0884
0.0861
Table: 5.1 ERLE comparison
a.Comparison between original signal and
estimated signals in RLS Algorithm in max
forgetting factor
Let’s consider two system in parallel, based on forgetting
factor
1
1 (n) 1 ( 11 (n 1) k (n)[ X T (n) 11 (n 1)])
(6)
1
© 2015, IRJET
ISO 9001:2008 Certified Journal
Page 1069
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
e-ISSN: 2395-0056
p-ISSN: 2395-0072
Figure 5.6: Estimated error
Figure 5.1: Input near end and far end signal for
echo cancelation Fs = 8000
Figure 5.7: Weightage Curve for existing system for
forgetting factor 0.90
Figure 5.2: Actual and Estimated output for input
signal
Figure 5.3: Estimated error
Figure 5.4: Weightage Curve for existing system for
forgetting factor 0.98
Comparisons graphs from figure 5.1-5.4:The
Estimated curves are obtained by varying forgetting factor
lamda. The Echo and reverberation both are maintain in
stable condition. The comparative curves have plotted
between actual and estimated signal. Figure 5.2 gives
estimated curves for stable echo and reverberation in
standard time limit. Figure 5.3 gives estimated error in
employed auditorium for sample N = 8000. From
analysing the results, we can infer that all the cases gave
good results. Among the Weightage curves, the distance
between the curves are high in stable situation it
represents this forgetting factor lamda gives only dilute
weightage values.
a. Comparison between original signal and
estimated signals in RLS Algorithm in min
forgetting factor
Figure 5.5: Actual and Estimated output for input
signal
© 2015, IRJET
Comparisons graphs from figure 5.5- 5.7: The
Estimated curves are obtained by varying forgetting factor
lamda into minimum (0.90). The Echo and reverberation
both are maintain in stable condition. The comparative
curves have plotted between actual and estimated signal.
Figure 5.6 gives estimated curves for stable echo and
reverberation in standard time limit and minimum lamda
= 0.90. Figure 5.7 gives estimated error in employed
auditorium for sample N = 8000 and lamda = 0.90. From
analysing the results, we can infer that all the cases gave
worst results. Among the Weightage curves, the distance
between the curves are high in stable situation it
represents this forgetting factor lamda gives only dilute
weightage values. Moreover the estimated and actual
signals are not matched in any condition.
b. Comparison between RLS Algorithm in minmax forgetting factor
Figure 5.9: Estimated error
Figure 5.10: Weightage Curve for proposed system
forgetting factor as a average
ISO 9001:2008 Certified Journal
Page 1070
International Research Journal of Engineering and Technology (IRJET)
Volume: 02 Issue: 05 | Aug-2015
www.irjet.net
6. CONCLUSION
A new algorithm was proposed for an acoustic
echo canceller with average RLS algorithm. Its
performance was studied in comparison with conventional
algorithms in a simulation. Good performance was
confirmed with the proposed algorithm. Furthermore, a
parallel echo cancelling architecture suitable for hardware
implementation by frequency domain transfer processing.
Near end signal, Far end signal echo and reverberation in
auditorium was gradually optimized using average RLS
filters by changing forgetting factor. The proposed system
is stable, when echo and reverberation is high. Finally, the
relationship between the echo cancellation algorithm the
measure ERLE and signal and Weightage of error signal
characteristics was clarified.
7 .REFERENCES
[1] Moura, Jose MF, James L. Flanagan, and Nikil S. Jayant,
"The Discipline of Signal Processing [Reflections]", IEEE
Signal Processing Magazine, Vol. 30, No. 6, pp. 174-176,
2013.
[2] Breining, C, Dreiscitel, P, Hänsler, E, Mader, A, Nitsch, B,
Puder, H, & Tilp, J, “Acoustic echo control. An application of
very-high-order adaptive filters”, IEEE Signal Processing
Magazine, Vol.16, No.4, pp.42-69, 1999.
[3] F.S. Hallack, M.R. Petraglia, Performance comparison of
adaptive algorithms applied to acoustic echo cancelling, In
the proceedings of 2003 IEEE International Symposium on,
vol. 2, pp. 1147–1150, June 2003.
[4] E. Haensler, “The hands-free telephone problem: an
annotated
bibliography
update”,
Annales
Des
Télécommunications, Vol. 49, No. 7-8, pp. 360-367, 1994.
[5] C. Breining, P. Dreiscitel, E. Hansler, A. Mader, B. Nitsch,
H. Puder, T. Schertler, G. Schmidt, J. Tilp, “Acoustic echo
control. An application of very-highorder adaptive”, IEEE
Signal Process. Mag., Vol. 16, No. 4, pp. 42–69, 1999.
[6] G. Enzner and P. Vary, “A soft-partitioned frequencydomain adaptive filter for acoustic echo cancellation”, In
the proceedings of IEEE International Conference on
Acoustics, Speech, and Signal Processing, Vol. 5, pp. 393–
396, 2003.
[7] K. Eneman, M. Moonen, “Iterated partitioned block
frequency-domain adaptive filtering for acoustic echo
cancellation”, IEEE Transaction on Speech Audio Process,
Vol. 11, No. 2, pp. 143-158, 2003.
[8] Sheng Wu, Xiaojun Qiu and Ming Wu, "Stereo Acoustic
Echo
Cancellation
Employing
Frequency-Domain
Preprocessing and Adaptive Filter", IEEE Transactions on
© 2015, IRJET
e-ISSN: 2395-0056
p-ISSN: 2395-0072
Audio, Speech, and Language Processing, vol. 19, no. 3,
pp.614-623, 2010.
[9] Fukui, M, Shimauchi, S., Hioka, Y., Nakagawa, A., and
Haneda, Y., "Double-talk robust acoustic echo cancellation
for CD-quality hands-free videoconferencing system", IEEE
Transactions on Consumer Electronics, IEEE Transactions
onConsumer Electronics, Vol. 60, no. 3, 2014.
[10] Bekrani, M, Khong, A.W.H. and Lotfizad, M., "A
Clipping-Based Selective-Tap Adaptive Filtering Approach
to Stereophonic Acoustic Echo Cancellation", IEEE
Transactions on Audio, Speech, and Language Processing,
Vol. 19 , No. 6, 2011.
[11] Yüksel Özbay, AhmetRe¸sit Kavsao˘ glu, “An optimum
algorithm for adaptive filtering on acoustic echo
cancellation using TMS320C6713 DSP”, Digital Signal
Processing, Vol. 20, No. 1, pp. 133-148, 2010.
[12] K. G Gunale, S. N. Motade, S.L. Nalbalwar, S. B.
Deosarkar, “Frequency Domain Adaptive Filter using FFT
Algorithm for Acoustic Echo Cancellation”, In the
proceedings of 3rd International Conference on Emerging
Trends in Engineering and Technology (ICETET), pp. 582587, 2010.
[13] Gil-Cacho, J.M, van Waterschoot, T. , Moonen, M. ,
Jensen, S.H, "A Frequency-Domain Adaptive Filter (FDAF)
Prediction Error Method (PEM) Framework for DoubleTalk-Robust Acoustic Echo Cancellation", IEEE/ACM
Transactions on Audio, Speech, and Language Processing,
Vol. 22 , No. 12, 2014.
[14] Sarmad Malik and Gerald Enzner, "State-Space
Frequency-Domain Adaptive Filtering for Nonlinear
Acoustic Echo Cancellation", IEEE Transactions on audio,
speech, and language processing, Vol. 20, No. 7, pp. 20652079, September 2012.
[15] Luis A. Azpicueta-Ruiz, Aníbal R. Figueiras-Vidal,
Aníbal R. Figueiras-Vidal, Jerónimo Arenas-García,
"Enhanced Adaptive Volterra Filtering by Automatic
Attenuation of Memory Regions and Its Application to
Acoustic Echo Cancellation", IEEE transactions on signal
processing, Vol. 61, No. 11, pp. 2745- 2750, June 2013.
[16] Zoran M. Šari´c, IstvanI.Papp, DraganD.Kukolj, Ivan
Veliki´c, Gordana Veliki´c, "Partitioned block frequency
domain acoustic echo canceller with fast multiple
iterations", Digital Signal Processing, Vol. 27, pp. 119–128,
2014.
[17] Luis A. Azpicueta-Ruiz, Anibal R. Figuieras-Vidal and
Jeronimo Arenas-Garcia, "Acoustic echo cancellation in
discrete Fourier transform domain based on adaptive
combination of adaptive filters", In the proceedings of
Meetings on Acoustics, Vol. 19, 2013.
ISO 9001:2008 Certified Journal
Page 1071
