International Journal of Current Engineering and Technology

E-ISSN 2277 – 4106, P-ISSN 2347 - 5161

®
©2014 INPRESSCO , All Rights Reserved
Available at http://inpressco.com/category/ijcet

Review Article
Noise Cancellation using Adaptive Filtering: A Review
Pratik GhotkarȦ*, Mrinal BachuteȦ and R D KharadkarkarȦ
Ȧ
Department of Electronics & Telecommunication, GHRIET, University of Pune, India

Accepted 20 Sept 2014, Available online 01 Oct 2014, Vol.4, No.5 (Oct 2014)

Abstract

At the end of communication system information signal may be introduced with some noise along with the information
signal. Noise may be additive or convoluted .Adaptive filter finds better solution to deal this problem. Use of Adaptive
filter may filter out or suppressed noise from noisy signal getting the clean speech .This process is known as the adaptive
noise cancellation. Subband adaptive filtering (SAF) finds better performance over a fullband adaptive filtering which
employs multirate filter bank for signal decomposition and reconstruction. This technique allows for fast convergence
and reduced computational complexity however insertion of filter bank introduced artifacts such as: aliasing, amplitude
and phase distortion. This article deals with the review of various advancement in the adaptive algorithms made.
Recently many methods have been proposed to reduce the effect of one or more artifacts. This article deals with the
review of various advancement in this particular area of digital signal processing. Comparison tables of the main
subband adaptive filtering are shown .The more effective comparison parameter of t these methods is Mean square Error
(MSE) Plot.

Keywords: Adaptive noise cancellation, Subband adaptive filtering (SAF), Mean square Error (MSE).

1. Introduction disturbance to yield a quieter environment. The basic

principle of ANC is to introduce a cancelling “antinoise”
1
In practice the received speech signal contains some signal that has the same amplitude but the exact opposite
amount of noise component along with the information. phase, thus resulting in an attenuated residual noise signal.
The noise may be occurs due to coding of transmitted ANC has been used in a number of applications such as
waveform (quantization noise) or an additive noise from hearing protectors, headsets, etc. The traditional wideband
background. Therefore for proper listing of this sound the ANC algorithms work best in the lower frequency bands
noise need to be removed or suppressed form received and their performance deteriorates rapidly as the
noisy signal .The designing of such filter which removes bandwidth and the center frequency of the noise increases.
or suppress noise required the signal and the noise be Most noise sources tend to be broadband in nature and
stationary that the statistics of both signals be known a while a large portion of the energy is concentrated in the
priori. In practice, these conditions are rarely met. lower frequencies, they also tend to have significant high
Various signal processing techniques have been frequency components. Further, as the ANC system is
proposed over the years for noise reduction in the signals. combined with other communication and sound systems, it
There are two different approaches for electrical noise is necessary to have a frequency dependent noise
reduction. The first approach is passive electrical noise cancellation system to avoid adversely affecting the
reduction techniques, such as those applied in hearing desired signal.
aids, cochlear implants, etc. where the signal and ambient The performance of any speech signal processing
noise are recorded using a microphone, noise reduction system is degraded in the presence of noise (either additive
techniques such as spectral subtraction, the LMS or convolution). This is due to the acoustic mismatch
algorithm, etc. are applied and the listener hears only the between the speech features used to train and test this
clean signal. One of the important assumptions of this system and the ability of the acoustic models to describe
technique is that the listener is acoustically isolated from the corrupted speech (B. Widrow et al, 1975).
the environment. This assumption is however not valid in When processing the speech signal, the quality of
a large particularly those number of situations where the speech may be at risk from various sources of interference
ambient noise has very large amplitude. In such situations, or distortions. Typical sources of interference are:
the second approach of Active Noise Cancellation (ANC) • Background noise added to the speech signal: for
is applicable. ANC refers to an electromechanical or example – environmental noise or engine noise when
electroacoustic technique of cancelling acoustic talking on a mobile phone,
• Acoustic or audio feedback: it occurs in two-way
*Corresponding author: Pratik Ghotkar communication when the microphone in the telephone
3353 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)
Pratik Ghotkar et al Noise Cancellation using Adaptive Filtering: A Review

Captures the actual speech of another person and the have the capability of adaptively tracking the signal under
speech of the first person reproduced from loudspeakers, non-stationary conditions.
and sends them both back to the first person,
• Amplifier noise: an amplifier can produce additional
thermal noise, which becomes noticeable during
significant signal amplifications,
• Quantization noise created in the transformation of the
analogue signal to digital: the interference occurs during
Sampling due to rounding up real values of the analogue
signal,
• Loss of signal quality, caused by coding and speech
compression. Fig.2 Adaptive Noise Canceller
2. Noise cancellation The error signal to be used depends on the application.
The criteria to be used may be the minimization of the
Noise Cancellation is a variation of optimal filtering that mean square error, the temporal average of the least
involves producing an estimate of the noise by filtering the squares error etc. (Riitta Niemist and Ioan Tabu et al,
reference input and then subtracting this noise estimate 2001), Different algorithms are used for each of the
from the primary input containing both signal and noise. minimization criteria e.g. the Least Mean Squares (LMS)
algorithm, the Recursive Least Squares (RLS) algorithm
etc. In the case of adaptive noise cancellation, we use the
minimum mean-square error criterion. Examples of the
adaptive filters are the Wiener filter, Recursive-Least-
Square (RLS) algorithm, and the Kalman filters were
proposed to achieve the best performance. Among them
Least Mean Square (LMS) algorithm is most commonly
Fig.1 Noise cancelation used because of its robustness and simplicity, However the
LMS algorithm suffers from significantly degraded
It makes use of an auxiliary or reference input which performance for colored interfering signals due to the
contains a correlated estimate of the noise to be cancelled. eigenvalue spread of the autocorrelation matrix of the
The reference can be obtained by placing one or more input signal. In addition, as the length of the adaptive filter
Sensors in the noise field where the signal is absent or its increases, the computational complexity increases. This
strength is weak enough. Subtracting noise from a can be a serious problem in acoustic applications such as
received signal involves the risk of distorting the signal echo and noise cancellation, where long adaptive filters
and if done improperly, it may lead to an increase in the are required to model the response of the noise path. This
noise level. This requires that the noise estimate nˆ should issue is of great importance in the hands-free application
be an exact replica of n. If it were possible to know the where processing power is kept low.
relationship between n and nˆ, or the characteristics of the An alternative approach to reduce the computational
channels transmitting noise from the noise source to the complexity of long adaptive FIR filters is to incorporate
primary and reference inputs are known, it would be block updating strategies and frequency domain adaptive
possible to make nˆ a close estimate of n by designing a filtering. These techniques reduce the computational
fixed filter. However, since the characteristics of the complexity, because the filter output and the adaptive
transmission paths are not known and are unpredictable, weights are computed only after a large block of data has
filtering and subtraction are controlled by an adaptive been accumulated. However, the application of such
process. Hence an adaptive filter is used that is capable of approaches introduces degradation in the performance,
adjusting its impulse response to minimize an error signal, including a substantial signal path delay corresponding to
which is dependent on the filter output. one block length, as well as a reduction in the stable range
3. Noise Cancellation Using Adaptive Filtering of the algorithm step size. Therefore for nonstationary
signals, the tracking performance of the block algorithms
A basic concept of Adaptive Noise Canceller (ANC) generally becomes worse. (Ali O. Abid Noor 2011)As far
removes or suppresses the noise from a signal using as speed of convergence is concerned, it has been
adaptive filters. Adaptive filters are digital filters with an suggested to use the Recursive Least Square (RLS)
impulse response, or transfer-function that can be adjusted algorithm to speed up the adaptive process. The
or changed over time to match desired system convergence rate of the RLS algorithm is independent of
characteristics (Soni Changlani, et al, 2011); they require the eigenvalue spread. Unfortunately, the drawbacks that
little or no a priori knowledge of the signal and noise is associated with RLS algorithm including its O (N2)
characteristics. (If the signal is narrowband and noise computational requirements, which are still too high for
broadband, which is usually the case, or vice versa, no a many applications, where high speed is required, or when
priori information is needed; otherwise they require a a large number of inexpensive units must be built. The
signal (desired response) that is correlated in some sense Affine Projection Algorithm (APA) shows a better
to the signal to be estimated.) Moreover adaptive filters convergence behavior, but the computational complexity
3354 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)
Pratik Ghotkar et al Noise Cancellation using Adaptive Filtering: A Review

increases with the projection order in relation to LMS, the subband filters wkˆ In contrast to the traditional noise
where Projection denotes the order of the APA. As a cancellation structure, in this setup, P(z) is estimated using
result, adaptive filtering using subband processing a set of parallel, independently updated filters wk Outputs
becomes an attractive option for many adaptive systems to of the subband adaptive filters yk are subtracted from the
reduce these problems subband desired signals v , forming the subband errors e .
These errors are then upsampled and recombined in the
4. Subband Adaptive Filtering in a Noise Cancellation synthesis filter bank G (z), leading to the clean output s.
Scenario The input/output relationship can be expressed as:

Subband adaptive filtering belongs to two fields of digital

signal processing, namely, adaptive filtering and multirate ̂ ( ) ∑ ( ) ( )
signal processing. The basic idea of SAF is to use a set of
parallel filters to divide the wideband signal input of the
adaptive filter into narrower subband signals, each
subband serving as an input to an independent adaptive Where, ( ) ( )
filter. Subband decomposition greatly reduces the adaptive
filter update rate through parallel processing of shorter There is Distortions due to the insertion of analysis and
filters. synthesis filter banks occurs which can be represented as,
Furthermore, subband signals are usually
downsampled in a multirate system. This leads to a
whitening of the input signals and therefore an improved ̂ ( ) ( ) ( ) ∑ () ( )
convergence behavior of the adaptive filter system is
expected. The subband decomposition is aimed to reduce Where,
the update rate, and the length of the adaptive filters,
hopefully, resulting in a much lower computational ( ) ∑ ( ) ( )
complexity (Ali O. Abid Noor et al, 2011).
The conventional noise cancellation model is extended
to a subband configuration by the insertion of sets of ( ) ∑ ( ) ( )
analysis and synthesis filters in signal paths, as depicted
by Figure 2. Both input signals s and n are now fed into
identical M-band analysis filter banks H (z), with nˆ being For i=1, 2…D-1. Here, A0(z) represents the total distortion
a filtered version of n by an unknown system P(z). Here, transfer function of the filter bank for the non-aliased
P(z) represents the acoustic noise path, n being correlated component of the system input S(z), while Ai(z) represents
with and uncorrelated with s. aliasing distortion and determines how well the aliased
The ultimate goal is to suppress nˆ at the output s and components of the system are attenuated.
to retain the non-distorted version of s. The update The filter bank is the key tool in the design of subband
equation of the adaptive filter using the LMS algorithm is adaptive filtering systems. Filter banks can be designed to
given by the following set of equation. be alias-free and perfectly reconstructed when certain
conditions are met by the analysis and synthesis filters.
̂( ) ̂( ) ( ) ( ) However, any filtering operation in the subbands may
( ) ( ) ( ) cause a possible phase and amplitude change, thereby
( ) ̂ (m).n (m) altering the perfect reconstruction property. There are
tradeoffs in controlling the aliasing effect and the
where e(m) represents the error signal, y(m) is the output amplitude distortion level.
of the adaptive filter, w(m) is the filter coefficient vector at Depending on the value of the downsampling factor D,
the mth iteration, μ is the adaptation step-size factor m is a filter banks can be either critically sampled or
time index and (.)T is the matrix transpose operator oversampled. Filter bank can be critically sample or
(Haykin, S et al, 2002). oversample .when downsapling factor is equal to number
This original model is extended to a subband of channels i.e. (M=D) the system is critically sampled and
configuration by the insertion of sets of analysis and when D<M the system is oversampled.
synthesis filters in signal paths, as depicted by Figure 2. Computational savings are maximized when signals
Both input signals s and n are now fed into identical M- are critically downsampled However these systems require
band analysis filter banks H (z), with being a filtered ideal filters in the analysis stage in order to avoid alising
version of n by an unknown system P(z). Here, P(z) distortion on other hand oversampling has reduced
represents the acoustic noise path, n being correlated with aliasing distortion is trade off for extra computational cost.
and uncorrelated with s. The ultimate goal is to suppress at
the output s and to retain the non-distorted version of s. 5. Subband Adaptive Filtering Techniques
After D-fold downsampling, adaptive filtering is
performed in each subband separately. Updating of A. Critically Sampled
adaptive filter coefficients can be done with any kind of
algorithm adaptation. However, for robustness and In The analysis of adaptive filtering algorithms found that
simplicity, the LMS algorithm is commonly used to update critically sampled subband system cannot properly model
3355 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)
Pratik Ghotkar et al Noise Cancellation using Adaptive Filtering: A Review

the unknown system without the use of cross adaptive FFT/IFFT transforms. The proposed scheme offers a
filters .However cross adaptive filter has two major simplified structure that without employing cross-filters or
disadvantages: slow convergence and extra computation gap filter banks reduces the aliasing level in the subbands.
(Gilloir, A. and M. Vetterliet al, 1992). The performance under white and colored environments is
Another researcher (Yamada et al. 2004) proposed the evaluated and compared to the conventional fullband
use of frequency sampling filter (FSF) bank, this idea is method as well as to a critically sampled technique
based on the transformation of the subband signals into the developed by (Kim et al. 2008). This evaluation is offered
frequency domain using discrete Fourier transforms in terms of MSE convergence of the noise cancellation
(DFT), then choosing a number of frequency samples from system.
each subband signal so that non-adjacent frequency bands
have nulls at the center frequency of each subband. Thus Table 1 Comparison of literature proposal for Critical
subband adaptive filtering produced minimum alising Sampling
distortion. Since the output is calculated only after
accumulating a large block of data. In addition, a high Source Convergence rate Computational
computational burden is inevitable due to DFT Complexity
calculations. Gilloire Slower N/M
In an attempt to reduce computational expense (Naylor and
et al.1998) proposed a subband adaptive system using Vitterli
allpass polyphase infinite impulse response (IIR) filter Yamada No Comparison found Approximately 6N
banks. These IIR structures were introduced as an Deleon Faster 2N/M
and Etter
alternative to the standard approach involving finite
Naylor Faster N/M
impulse (FIR) filter banks. Such multirate systems with
Kim Comparable Under 2N/M+2(L’+)+3(Lω
very high interband discrimination and low computational white noise +1)+LFB +3
cost could be built using allpass polyphase structures. It
was verified that adaptive filters in such systems perform B.Oversampled Structures
as well as systems based on FIR filters. Spectral holes and
signal delays are the main drawbacks associated with such (DeLeon and Etter et al. 1995) verified that in spite of the
an approach. These drawbacks are the adverse outcomes increase in computational load, oversampled systems are
of adopting notch filters between subbands for the purpose still more efficient than the equivalent fullband ones for a
of reducing aliasing. The use of IIR filter banks is also certain number of subbands. (sridharan et al.1998) has
discussed by (Noor et al. 2010). also developed an oversampled technique for adaptive
Another research, (Kim et al. 2008) has proposed a filtering purposes. Sridharan’s study targeted issues both
critically sampled structure to reduce aliasing effect. The of complexity and convergence; the computational
inter-band aliasing in each subband is obtained by complexity was reduced to a half of the fullband system.
increasing the bandwidth of a linear-phase FIR analysis This was achieved by restricting non-adjacent filters of a
filter, and then subtracted from the subband signal. This perfect reconstruction filter bank (PRFB) to be non-
aliasing reduction technique introduces spectral dips in the overlapping. In Sridharan’s method, it is not clear whether
subband signals. Therefore, extra filtering operation is the system will give a satisfactory convergence for colored
required to reduce these dips (Gilloir and Vetterli et al. input signals or not.
1992). Oversampled schemes have been suggested as the In a related study, (Chin and Boroujeny et al.2001)
most appropriate solution to avoid aliasing distortion have suggested the use of real-valued subband signals
associated with the use of critically sampled filter instead of the conventional complex-valued types,
banks.This was originally recommended by Gilloire and proposing a subband adaptive filter structure using an
Vitterli .The small eigenvalues are generated by the roll- SSB-modulated filter bank.
off of the subband input power spectrum. It was later
demonstrated by the same authors that this problem could Table 2 Comparison of oversample structures
be mitigated by the use of increased bandwidth analysis
filters (DeLeoAnd Etter 1995).Effectively the amplitude Source Convergence rate Computation
distortion, if the filters constituting the filter bank were DeLeon and Initially faster ,slower 2N/M
Etter on steady state
constrained to be linear phase FIR filters. Using amplitude
Sridharan Faster compared to 2N/M
distortion as a criterion to minimize aliasing by a nonlinear standerd
optimization procedure, it was found that the best Chin an Faster 2N/M+real
prototype filter is the one designed using Kaiser or Dolph- Boroujeny Value
Chebychev windows Optimization techniques have also
been discussed by Noor et al. (2011). An optimized 2-fold The resulting subband signals are real-valued, thus
oversampled M-band noise cancellation technique is used eliminating the need to deal with complex-valued signals
to mitigate the problem of aliasing insertion associated as in the case of conventional subband adaptive filters.
with critically sampled schemes. Variable step size version The resulting subband adaptive filter has performance
of the LMS algorithm is used to control the noise in the comparable to its complex-valued counterpart in terms of
individual branches of the proposed canceller. The system delay, convergence and distortion. However, the technique
is implemented efficiently using polyphase format and poses a potential increase in the computational complexity
3356 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)
Pratik Ghotkar et al Noise Cancellation using Adaptive Filtering: A Review

Table 3 Comparison of Delayless subband adaptive filtering

Algorithm Number of real Multiplication

Proposed UDFT 4l
PFFT-2 [ ]( )+

Table 4 Tabular Comparison on Some Surveyed Literatures

Author Research Description

Boll, S. Suppression of acoustic noise In this paper a novel method for cancellation of broadband/narrowband noise from speech
in speech using two signals is proposed. Independent component analysis (ICA) and wavelet packet approaches
microphone adaptive noise have been combined for blind noise separation from
cancellation Mixtures of speech signals. ICA method is used to estimate matrix A, which defines how the
mixture signals have been mixed. Wavelet packets are used for de-correlation of
approximation of noise and speech.
Darlington, Sub-band, dual-channel An adaptive noise cancellation scheme for speech processing is described. Adaptive filters are
D.J. adaptive noise cancellation implemented in sub-bands, based on a model of the human cochlea. A modification to the
using normalised LMS LMS structure is introduced which guarantees stability and convergence. This modification, a
non-recursive normalisation, is used both in a wideband and in a sub-band implementation of
the scheme. The
effect of this normalisation is to cause the speech to be distorted, indicating that there is little
benefit in using normalised LMS in a sub-band scheme, whether the application uses classical
or intermittent noise cancellation
Ali A. A New Delayless Subband Acoustic paths such as those encountered in ANC application usually have long impulse
Milani Adaptive Filtering Algorithm responses, which require longer adaptive filters for noise cancellation. Subband adaptive
for Active Noise Control filters working with a large number of subbands have been shown to be a good solution to this
Systems problem. The focus of this paper was to design such a high-performance SAF algorithm. The
performance limiting factors of existing SAF structures were found to be due to the inherent
delay and side-lobes of the prototype filter in the analysis filter banks. Hence, the analysis
filter banks were modified to reduce the inherent delay. A new weight stacking transform was
designed to alleviate the interference introduced by the side-lobes.
Ali O. Abid Adaptive Filtering Using Adaptive filter noise cancellation systems using subband processing are developed and tested
Noor, Subband Processing: in this chapter. Convergence and computational advantages are expected from using such a
Application to Background technique. Results obtained showed that; noise cancellation techniques using
Noise Cancellation Critically sampled filter banks have no convergence improvement, except for the case of two-
band QMF decomposition, where the success was only moderate. Only computational
advantages may be obtained in this case. An improved convergence behavior is obtained by
using two-fold oversampled DFT filter bank that is optimized for low amplitude distortion.
Riitta Signal Adaptive Subband A new architecture for adaptive noise cancellation where the signals involved are first
Niemist¨o2 Decomposition for Adaptive decomposed in two subbands and adaptive filtering is performed separately for each subband
Noise Cancellation signal. When the subband decomposition is performed such that the analysis filters compacts
most of noise power in one subband and leaves almost no noise power in the other band, the
adaptive filtering turns out to be more efficient than in the single channel case.

of the system due to the increased number of subbands literature to circumvent this problem. A delayless structure
(Noor et al. 2011). They have optimized a Hamming in a noise cancellation setup is shown in Figure 7. The
window base analysis/synthesis to achieve a good pioneering work involving this type of schemes was
convergence behavior at moderate computational costs. performed by Morgan and Thi et al. (1995). They have
The advantages of subband adaptive filtering systems presented a new class of subband adaptive filter
have been widely acknowledged. Although the gain in architecture in which the adaptive weights adaptive filter
computational complexity is clearly advantageous in long architecture in which the adaptive weights transformed
acoustic environments, the use of SAF may be impractical into an equivalent set of wideband filter coefficients. In
in the presence of high levels of distortion brought about this manner, signal path delay is avoided while retaining
by the insertion of filter banks. Filter banks introduce three the computational and convergence speed advantages of
types of artifacts: aliasing, amplitude and phase subband processing. An additional benefit is accrued
distortions. Another disadvantage of using SAF is the through a significant reduction of aliasing effects. More
extra processing delay, which may rule out the use of these efficient subband filters can be designed by relaxing the
systems for real-time implementations low stopband response necessary to control aliasing. The
delayless structure is very similar to the frequency domain
C. Delayless Subband Schemes structure proposed earlier by Shynk et al.(1992), i.e., the
adaptive weights are computed for each subband’s FFT
The conventional approach to subband adaptive filtering is bins separately and then transmitted to an equivalent
ruled out for many applications because delays are wideband filter.
introduced into the signal path. Delayless subband However, it differs from the frequency domain
adaptive filtering schemes have been proposed in the structure in that the actual processing of the subband
3357 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)
Pratik Ghotkar et al Noise Cancellation using Adaptive Filtering: A Review

signal takes place in the time domain. Subband adaptive .References

filters working with a large number of subbands has been
proposed by The ali milani et al. (2009)in their The B. Widrow, (1975) Adaptive Noise Cancelling: Principles
performance limiting factors of existing SAF structures and Applications; Proc. IEEE, vol. 63, pp.1692-1716.
were found to be due to the inherent delay and side-lobes Soni Changlani & M.K.Gupta, (2011) Simulation of LMS Noise
of the prototype filter in the analysis filter banks. Hence, Canceller Using Simulink; International Journal On Emerging
Technologies,, ISSN : 0975-8364, pp 50-52.
the analysis filter banks were modified to reduce the Riitta Niemist and Ioan Tabu, (2001), Signal Adaptive Subband
inherent delay. A new weight stacking transform was Decomposition for Adaptive Noise Cancellation1, European
designed to alleviate the interference introduced by the Conference on Circuit Theory and Design.
side-lobes. Ali O. Abid Noor(2011) Adaptive Filtering Using Subband
Processing: Application to Background Noise
Cancellation; ISBN 978-953-307-158-9.
Haykin, S(2002)Adaptive Filter Theory, 4th Edn., Prentice Hall,
New Jersey.
Gilloir, A. and M. Vetterli (1992) Adaptive Filtering in Subbands
with Critical Sampling: Analysis, Experiments and
Application to Acoustic Echo Cancellation ; IEEE
Transactions on Signal Processing, and 40(8): 1862-1875.
Yoji Yamada, Hiroshi Ochi and Hitishi Kiya,(2004) A Subband
Adaptive Filter Allowing Maximally Decimation; IEEE
Journal on Selected Areas in Communications, 12(9): 1548-
1552.
Patrick A. Naylor, O¢guz Tanr kulu and Anthony G.
Constantinides, (1998)Subband Adaptive Filtering for
Acoustic Echo Control Using Allpass Polyphase IIR
Fig.3 Delayless subband adaptive filtering configuration Filterbanks; IEEE Transactions on Speech and Audio
Processing, 6(2): 143-155.
Noor, A.O.A., S.A. Samad and A. Hussain,(2011) Adaptive
Conclusion Filtering using Subband Processing: Application to
Background noise Cancellation. In “Adaptive Filtering”; I-
The performance of any speech signal processing system
Tech: Vienna, Austria,pp: 109-134.
is degraded in the presence of noise (either additive or Kim, S.G., Yoo, C.D. & Nguyen, T.Q. (2008)Alias-free subband
Convolution). This is due to the acoustic mismatch adaptive filtering with critical sampling.;IEEE Transactions
between the speech features used to train and test this on Signal Processing. Vol.56, No.5, pp. 1894-1904. ISSN
system and the ability of the acoustic models to describe 1053-587X.
the corrupted speech. Various techniques for filtering the Gilloir, A. and M. Vetterli (1992); Adaptive Filtering in
noise from a speech waveform have been studied. LMS Subbands with Critical Sampling: Analysis, Experiments and
algorithm operating in fullband suffers from significantly Application to Acoustic Echo Cancellation; IEEE
degraded performance for colored interfering signals Transactions on Signal Processing, 40(8): 1862-1875.
DeLeo´N P. and D.M. Etter (1995) Experimental results with
however this effect can be minimize with use of subband
increased bandwidth analysis filters in oversampled, subband
processing. acoustic echo cancellers;IEEE Signal Processing Letters, 2(1):
The Depending upon the value of Downsampling rate 1-3.
and number of subband the SAF (Subband adaptive Akansu, A.N. and Mark J.T. Smith, (1995) Subband and Wavelet
Filtering) is divided as critically sampled ad Oversampled. Transforms Chapter12, Kluwer Academic Publishers Group
Standard critically-sampled systems offer optimum NJ.
computational savings. Unfortunately, these systems Sridharan, M.K.(1998)Subband adaptive Filtering: Oversampling
cannot properly model unknown systems and also require approach ;Elsevier Signal Processing, 71:101 104.
cross-adaptive filters to correctly model the systems. This doi:10.1016/ S01651684(98)00173-X.
Chin, W.H. and B. Farhang-Boroujeny(2001) Subband Adaptive
leads to slow convergence and additional computational
Filtering With Real- Valued Subband Signals for Acoustic
costs. Employing infinite impulse response IIR filter banks Echo Cancellation;IEEE Proceedings. Image Signal
offers the best solution in terms of convergence and Processing, 148(4): 283-288.
computational costs. Oversampled structures have D. R. Morgan and J. C. Thi, (1995) A delayless subband adaptive
mitigated the problem of aliasing in critically sampled filter architecture; IEEE Trans. Signal Process, vol. 43, no. 8,
systems. This comes at the expense of higher pp. 1819–1830, Aug. 1995.
computational costs. As a result, increased bandwidth Shynk, J.J.(1992)Frequency Domain and Multirate Adaptive
analysis filters have been proposed to solve this problem. Filtering,;IEEE Signal Processing Magazine, 9: 14-37.
Delayless proposals have addressed the input-output delay Ali A. Milani, M. S. Panahi, (2009) A New Delayless Subband
Adaptive Filtering, Algorithm for Active Noise Control
while retaining convergence and complexity issues as
Systems;IEEE Transactions on Speech, And language
advantages of subband decomposition. The computational Processing, VOL. 17, No. 5.
requirements of such structures are still too high for many Simon Haykin(1996)Adaptive Filter Theory 3rd edition;
applications, where high speed is required, or when a large Prentice Hall, Englewood Cliffs, New Jersey.
number of inexpensive units must be built. In applications
such as speech and audio, highly selective filter banks are
necessary.
3358 | International Journal of Current Engineering and Technology, Vol.4, No.5 (Oct 2014)