July 2014
July 2014
net/publication/330902230
CITATIONS READS
9 7,656
3 authors, including:
All content following this page was uploaded by Gampa V P Chandra Sekhar Yadav on 13 February 2019.
16
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
comparison of adaptive algorithms for noise cancellation”. resist both desire signal and noise signal. The filter length of
the adaptive system is inherently tied to many of the other
In this paper the authors have compared the performance of performance measures. The length of the filter specifies how
different adaptive algorithms like LMS, NLMS, and RLS accurately a given system can be modeled by the adaptive
algorithms for sinusoidal input. Performance is compared for filter. In addition, the filter length affects the convergence
different step size and different number of iterations. Three rate, by increasing or decreasing computation time, it can
adaptive filter algorithms have been compared by simulation affect the stability of the system, at certain step sizes, and it
to achieve high convergence rate and minimum mean square affects the minimum MSE.
error with noise and different values of µ. Every algorithm
works on different methods for noise cancellation and 6. FILTERS
improves system performance. Filters are the basic elements in the signal processing system.
The authors G.V.P.Chandra Sekhar Yadav et al [4] proposed Filter is a device used to suppress the unwanted signal i.e.
“Study of different adaptive filter algorithms for noise noise from the desired signal as shown in figure 2and there
cancellation in real time environment”. are several techniques are used for filtering. Usual method of
estimating the signal corrupted by noise is to pass to it
In this paper the authors have proposed the performance of through the filter that tend to suppress the noise and leaving
wiener filter and different adaptive filter algorithms like LMS, the desired signal this is so called direct filtering. In general
NLMS and RLS algorithms for noise cancellation in real time filters are of two types. Fixed filters and adaptive filters
environment like recorded speech as the input and different
noise signals are added to it and then desired signal is In fixed filters the frequency response or filter coefficients of
estimated by using the adaptive algorithms. the filters are fixed and it requires prior knowledge of the
input signals. If the signal and noise characteristics are known
4. PROPOSED WORK beforehand then it is easy to design the filter that passes the
In the existing work, the authors G.V.P.Chandra Sekhar frequencies contained in the signal and rejects the frequency
Yadav and Dr. B. Ananda Krishna [4] proposed the band occupied by noise.
performance of wiener filter and adaptive filter algorithms An adaptive filter on the other hand requires less or no
like LMS, NLMS and RLS algorithms in real time knowledge about the signal and noise. In adaptive filters the
environment. Now, this paper is going to work on the part of filter coefficients are adjustable or modifiable.
the existing work like wiener filter and adaptive filter
algorithm i.e. LMS algorithm in real time environment by
taking input as recorded speech with different background
noises to compare their performance and the performance is
analyzed with different lengths and parameters of the input as
shown in figure 1.
17
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
18
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
The use of conventional filters in the above cases would lead 7. ADAPTIVE ALGORITHMS
to unacceptable distortion of the desired signal. Adaptive algorithms are used to adjust the coefficients of the
digital filter. Such that the error signals is minimized
according to some criterion.
The most commonly used adaptive algorithms are least mean
square (LMS), normalized least mean square (NLMS) because
of their robustness and simplicity and also have recursive least
square (RLS) algorithm for noise cancellation.
Where and are the weight and the true gradient vectors
respectively at the sampling instant. µ controls the size of
incremental correction applied to the adaptive filter from one
iteration to the next.
The steepest descent algorithm in the above equation still
requires knowledge of R and P. since is obtained by
evaluating the equation.
19
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
-0.4
-0.6
-0.8
0 100 200 300 400 500 600 700 800 900 1000
No.Of.Iterations
0.1
0.05
0
Amplitude
-0.05
-0.1
20
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
resultant of the mixing is shown in figure 11. And the mixed Output Of LMS algorithm for Sinusoidal Input
2
noise is eliminated as possible asby using different algorithms
at the end of the final iteration. 1.5
Noisy Signal
1
0.6
0.5
0.4
Amplitude
0
0.2
-0.5
0
Amplitude
-1
-0.2
-1.5
-0.4 -2
0 100 200 300 400 500 600 700 800 900 1000
No.Of.Iterations
-0.6
Fig 13: LMS algorithm output for sinusoidal input
-0.8
0 100 200 300 400 500 600 700 800 900 1000 MSE Of Wiener filter for Sinusoidal Input
-2
No.Of.Iterations 10
9. SIMULATION RESULTS
The result analysis from the outputs of wiener filter and LMS
Squared Error Value
-6
10
algorithm shows that Wiener algorithm provides better
performance but it has high degree of Complexity.
-8
10
From the figure 12 it is clear that wiener filter gives good
performance but it has high computational complexity.
Wiener filter has to compute R and P explicitly but LMS -10
10
algorithm is suited for real time applications where sample by
sample bias is done. In wiener filter the computational
-12
complexity is also high because of auto correlation and cross 10
0 100 200 300 400 500 600 700 800 900 1000
correlation so adaptive filters are designed for moderate No.Of.Iterations
computations. Fig 14: MSE for wiener filter for sinusoidal input
1.5
Output Of Wiener filter for Sinusiodal Input Figure 14 and 15 shows the Mean Square Error for sinusoidal
input for wiener filter and LMS algorithm respectively.
Figures show that LMS algorithm has Minimum Mean Square
1
Error which yields better results. MSE is decreasing function
for LMS algorithm where as in wiener filter MSE is a random
0.5 function it decreases first and then increases hence LMS
algorithm is suitable for better performance and low
Amplitude
0 complexity.
MSE Of LMS algorithm for Sinusoidal Input
2
-0.5 10
0
10
-1
-2
10
-1.5
Squared Error Value
0 100 200 300 400 500 600 700 800 900 1000
-4
No.Of.Iterations 10
21
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
Output Of Wiener filter for realtime Input up to some iteration and after some point MSE is going to
0.6
decrease. This is the good scenario but it is going to decrease
up to some other iteration and will suddenly increase at final
0.4
iteration.
0.2 MSE of LMS algorithm for realtime input
-4
10
0
Amplitude
-6
10
-0.2
-0.6
-10
10
-0.8
0 100 200 300 400 500 600 700 800 900 1000
No.Of.Iterations
-12
10
Fig 16: wiener filter output for real time input
Output Of LMS algorithm for realtime Input -14
0.6 10
0 100 200 300 400 500 600 700 800 900 1000
No.Of.Iterations
0.4
Fig 19: MSE for LMS algorithm in real time environment
0.2 Fig 19 shows that the mean square error for LMS algorithm.
Figure shows that MSE is an increasing function for this
0 reason the output of LMS algorithm is somewhat distorted.
Amplitude
-4
increases.From the simulation results it is clear that LMS
10
algorithm is best suited because of its low cost and complexity
-5
than wiener filter.LMS algorithm is the best algorithm but it
10
has low convergence speed. This project compares only LMS
-6 algorithm and wiener filters for real time input. As a future
10
work the same work is going to extend in real time
10
-7 environment with different background noise and compare
their performance with other adaptive algorithms i.e., NLMS
10
-8 and RLS to achieve high convergence rate. The performance
0 100 200 300 400 500 600 700 800 900 1000
No.Of.Iterations
is analyzed with different parameters and lengths of input
signal.
Fig 18: MSE for wiener filter in real time environment
Figure 18 shows the mean square error for wiener filter. From 11. REFERENCES
the figure we can analyze that the mean square error increases [1] Yen-Hsiang chen, Shanq-Jang Ruan, Tom Qi, ”An
22
International Journal of Computer Applications (0975 – 8887)
Volume 97– No.15, July 2014
Automotive Application of real time adaptive wiener saddle River, NJ: Prentice –Hall, 2002.
filter for noise cancellation in a car environment,”
IEEE,2012,4673-2193. [7] J.Gorriz and J.Ramrez, “A Novel LMS Algorithm
Applied to Adaptive Noise Cancellation,” IEEE Signal
[2] H.Kaur and R.Talwar, “Performance and Convergence Process Letters, vol. 16, no. 1, Jan. 2009.
Analysis of LMS Algorithm,” IEEE ICCIC, Dec.2012.
[8] K. A. Lee, W. S. Gan, and S. M. Kuo, Subband Adaptive
[3] Kaur. H and Talwar.R, “Performance comparison of Filtering: Theory and Implementation. Hoboken, NJ:
adaptive algorithms for noise cancellation”, Engineering Wiley, 2009.
trends in communication, C2SPCA 2013.
[9] C.Gabriela and M.Sarachin, “Echo Cancellation Using
[4] G.V.P.Chandra Sekhar Yadav and Dr. B. Ananda LMS Algorithm,” U.P.B Sci Bull., Series C, vol. 71, no.
Krishna, “Study of different adaptive filter algorithms for 4, 2009.
noise cancellation in real time environment”,
International journal of computer applications (0975- [10] B.Widrow, J.R.Glover “Adaptive Noise Cancelling:
887), vol. 96, no. 10, June. 2014. Principles and Applications,” IEEE Proceedings, Vol-63,
No.12, Dec. 1975.
[5] B.Widrow and S.D.Stearns, Adaptive Signal Processing,
Englewood Cliffs, NJ: Prentice- Hall, 1985. [11] Darcy Tsai, Introduction of Wiener Filter, Graduate
Institute of Electronics Engineering Nation Taiwan
[6] S.Haykin, Adaptive Filter Theory, Fourth edition, Upper University, Taipei, Taiwan, ROC.
IJCATM : www.ijcaonline.org
23