Escolar Documentos
Profissional Documentos
Cultura Documentos
max
(1)
where
max
is the maximum eigen value of the input data covariance matrix. The implementation of LMS algorithm
involves iterative computations involving,
filter output;
1
0
( )
N
k k k i
i
n w i x
=
=
(2)
error estimate;
k k k
e y n
.
= (3)
update of filter weights;
1
( ) ( ) 2
k k k k i
w i w i e x
+
= + (4)
The simplicity of the LMS algorithm and ease of implementation, evident from (2), (3) and (4), makes it the
algorithm of first choice in many real-time systems. The LMS algorithm requires approximately 2N+1 multiplications and
2N+1 additions for each new set of input and output samples. Most signal processors are suited to the mainly multiply-
accumulate arithmetic operations involved, making a direct implementation of the LMS algorithm attractive.
B. RLS Algorithm
The RLS (recursive least squares) algorithm is another algorithm for determining the coefficients of an adaptive
filter. In contrast to the LMS algorithm, the RLS algorithm uses information from all past input samples (and not only from
the current tap-input samples) to estimate the (inverse of the) autocorrelation matrix of the input vector [9].
The RLS algorithm is based on the well-known least squares method. With recursive least squares algorithm, the
estimates of
k
W can be updated for each new set of data acquired without repeatedly solving the time-consuming matrix
inversion directly. A suitable RLS algorithm can be obtained by exponentially weighting the data to remove gradually the
effects of old data on
k
W and to allow the tracking of slowly varying signal characteristics. Thus
1 k k k k
W W G e
= + (5)
1 1
1
( )
T
k k k k
P P G x k P
(
=
(6)
1
( )
k
k
k
P x k
G
o
= (7)
1
( )
T
k k k
e y x k W
= (8)
1
( ) ( )
T
k k
x k P x k o
= + (9)
Adaptive Noise Cancellation using Multirate Techniques
29
k
P is essentially a recursive way of computing the inverse matrix
1
T
k k
X X
(
. The argument k emphasizes the
fact that the quantities are obtained at each sample point. The typical value of (forgetting factor) is between 0.98 and 1.
Smaller values assign too much weight to the more recent data, which leads to wildly fluctuating estimates.
The RLS algorithm is computationally more complex than the LMS algorithm. However, due the recursive
updating, the inversion of matrix is not necessary (which would be a considerably higher computational load). The RLS
algorithm typically shows a faster convergence compared to the LMS algorithm. Other advantages are that it produces a
weight vector estimate only at the end data sequence
III. PROPOSED SCHEME
The proposed structure of adaptive noise cancellation scheme using multirate technique is shown in Fig. 2. Starting
with the basic framework for Adaptive filters, a structure has been built eliminating the basic faults arising like
computational complexities, aliasing and spectral gaps.
Fig. 2. Proposed Structure of ANC with Multirate Technique
The H
0
, H
1
, H
a
are the analysis filters and G
0
, G
1
are the reconstruction filters. The decimation and interpolation
factors have been taken as 2 as the number of sub-bands are 2. The proposed scheme achieves a lower computational
complexity, and this design ensures no aliasing components in the output of the system. The system consists of two main
sub-bands and an auxiliary sub-band. The auxiliary sub-band contains the complement of the signals in the main sub-band.
In the fig. 2, H
a
(z) is the analysis filter for the auxiliary sub-band and H
0
(z) and H
1
(z) are the analysis filters for the
main bands. G
0
(z) and G
1
(z) are reconstruction filters for the main bands. These filters are related to each other as;
1 0
( ) ( ) H z H z = (10)
0 1
( ) 2 ( ) G z H z = (11)
1 0
( ) 2 ( ) G z H z = (12)
2 2
0 1
( ) ( ) ( )
m
a
H z z H z H z
(
=
(13)
The coefficients of all filters are calculated and the scheme is tested for different input types.
IV. RESULTS AND ANALYSIS
The proposed scheme using 2 bands with decimation factor of 2, is tested for deterministic signal, speech and
musical signals. The results are compared with conventional ANC.
A. Deterministic Signal
For deterministic signal, 128 samples of signal and noise are considered. Fig. 3 to Fig. 6 shows the convergence
results of the proposed scheme and conventional ANC for a deterministic signal of 10sin(1500t) and noise of 10 sin(312t).
Adaptive Noise Cancellation using Multirate Techniques
30
Fig. 3. Deterministic signal Conventional ANC Fig. 4. Input and output spectra Conventional ANC
Fig. 5. Deterministic signal Proposed Scheme Fig. 6. Input and output spectra Proposed Scheme
It is evident that, the output of the noise canceller exactly matches the desired signal. The update algorithm used is
LMS. The periodogram clearly demonstrates the removal of noise.
B. Speech Signal
For speech signal, 32000 samples with sampling frequency 8 KHz are considered. The noise signal is a sine wave
of 312 Hz. The Algorithm used is LMS. Fig. 7 to Fig. 10 shows the results of the proposed scheme and conventional ANC
for a speech signal.
Fig. 7. Speech signal Conventional ANC Fig. 8. Spectra of Speech Conventional ANC
Adaptive Noise Cancellation using Multirate Techniques
31
Fig. 9. Speech Signal Proposed Scheme Fig. 10. Spectra of Speech signal Proposed Scheme
C. Music Signal
The music signal considered is 8000 samples of piano along with 8000 samples of tabla taken as noise. The
Algorithm used is LMS. Fig. 11 to Fig. 14 shows the results of the proposed scheme and conventional ANC for a music
signal (piano + tabla). Fig. 15 to Fig. 16 shows the results of the proposed scheme for a music signal of closely matched
frequencies (guitar + violin).
Fig. 11. Music signal (Piano + Tabla) Fig. 12. Spectra of Music signal (Piano + Tabla)
Conventional ANC Conventional ANC
Adaptive Noise Cancellation using Multirate Techniques
32
Fig. 13. Music signal (Piano + Tabla) Fig. 14. Spectra of Music signal (Piano + Tabla)
Proposed Scheme Proposed Scheme
Fig. 15. Music signal (Guitar + Violin) Fig. 16. Spectra of Music signal (Guitar + Violin)
Proposed Scheme Proposed Scheme
The results are encouraging for the proposed multirate adaptive scheme and are indicative of quite less time
(approximately one-fourth) for computation as compared to the conventional ANC structure. Additionally, signals with
closely matched frequencies (eg. violin and guitar) can be effectively segregated using the proposed scheme, which is not
possible with the conventional ANC configuration.
The Table 1 shows the computation time for conventional ANC and the proposed scheme.
TABLE I: COMPUTATION TIME REQUIREMENT
Input Signal Conventional ANC Proposed Scheme
Deterministic 16 ms 4.2 ms
Speech 4.453 sec 1.485 sec
Musical 0.18441 sec 0.134881 sec
V. CONCLUSIONS
This paper proposes a new adaptive noise cancellation structure based on multirate techniques. Noise Cancellation
is chosen as the application because noise is one of the main hindering factors that affect the information signal in any
system. Noise and signal are random in nature. As such, in order to reduce noise, the filter coefficients should change
according to changes in signal behaviour. The adaptive capability will allow the processing of inputs whose properties are
unknown. Multirate techniques can be used to overcome the problem of large computational complexity and slow
convergence rate. The simulations and experiments demonstrate the efficacy of the proposed structure.
REFERENCES
[1]. Britton C. Rorabaugh, Digital Signal Processing Primer, New Delhi, India: TMH, 2005.
[2]. Bernard Widrow and Samuel D. Stearns, Adaptive Signal Processing, 3
rd
Indian Reprint, New Delhi, India:
Pearson Education, 2004.
[3]. Simon Haykin, Adaptive Filter Theory, 4
th
ed., New Delhi, India: Pearson, 2003.
[4]. Emmanuel C. Ifeachor and Barrie W. Jervis, Digital Signal Processing, 2
nd
ed., New Delhi, India: Pearson, 2002.
[5]. P. P. Vaidyanathan, Multirate Digital Filters, Filter Bank, Polyphase Networks and Applications: A Tutorial,
Proceedings of the IEEE, vol.78, no. 1, pp. 56-91, Jan. 1990.
Adaptive Noise Cancellation using Multirate Techniques
33
[6]. Vinay P. Sathe, and P. P. Vaidyanathan, Effects of Multirate Systems on the Statistical Properties of Random
Signals, IEEE Transactions on Signal Processing, vol. 41, no. 1, pp. 131-146, Jan. 1993.
[7]. Vinay P. Sathe, and P. P. Vaidyanathan, Analysis of the effects of Multirate Filter on Stationary Random Input,
with Application in Adaptive Filtering, in Proc. IEEE International Conference on Acoustics, Speech and Signal
Processing, (ICASSP91), April 1991, vol.3, pp. 1681 -1684,
[8]. Vinay P. Sathe, and P. P. Vaidyanathan, Efficient Adaptive Identification and Equalization of Bandlimited
Channels using Multirate/ Multistage FIR Filters, in Proc. Twenty Fourth Asilomar Conference on Signals,
Systems and Computers, Nov. 1990, vol.2, pp. 740-744.
[9]. John J. Shynk, Frequency-Domain and Multirate Adaptive Filtering, IEEE Signal Processing Magazine, vol.9,
no.1, pp. 14-37, Jan. 1992.
[10]. Cheng-Shing Wu, and An-Yen Wu, A Novel Multirate Adaptive FIR Filtering Algorithm and structure, in Proc.
IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP99), Taiwan, Mar. 1999,
vol.4, pp. 1849-1852.
[11]. Junya, Yoshikaju Miyanaga, and Koji Tochinai, Consideration on Decimation Factors in Multirate Adaptive
Filtering for a Time-Varying AR Model, in Proc. IEEE Asia Pacific Conference on Circuits and Systems,
Sapporo, Dec. 2001, pp. 358-363.
[12]. Geoffrey A.Williamson, Sourya Dasgupta, and Minyue Fu, Multistage Multirate Adaptive Filters, in Proc. IEEE
International Conference on Acoustics, Speech and Signal Processing, (ICASSP96), Chicago, May 1996, vol.3,
pp. 1534 -1537.
[13]. Karrakchou Mohsine, and Murat Kunt, New Structure for Multirate Adaptive Filtering: Application to
Interference cancelling in Biomedical Engineering (Invited Paper), in Proc. 16
th
Annual International Conference
of the IEEE on Engineering in Medicine & Biology Society, Switzerland, Nov. 1994, vol.1, pp. 14a-15a.
[14]. Marc de Courville, and Pierre Duhamel, Adaptive Filtering in Subbands using a Weighted Criterion, in Proc.
IEEE International Conference on Acoustics, Speech and Signal Processing, (ICASSP01), Paris, May 2001, vol.2,
pp. 985-988.
[15]. V. S. Somayazulu, S. K. Mitra, and J. J. Shynk, Adaptive Line Enhancement using Multirate Techniques, in
Proc. IEEE International Conference on Acoustics, Speech and Signal Processing, (ICASSP89), California, May
1989, vol.2, pp. 928 -931.
[16]. Eneman Koen, and Marc Moonen, Iterated Partitioned Block Frequency-Domain Adaptive Filtering for Acoustic
Echo Cancellation, IEEE Transaction on Speech Processing, vol. 11, no. 2, pp. 143-158, Mar. 2003.
[17]. Using MATLAB, ver. 7.0, The Mathworks Inc., Natick
[18]. Signal Processing Toolbox, ver. 6.2, The Mathwork Inc., Natick, May 2004.