Affine Projection Adaptive Filter is a Better Noise Canceller

M Yasin; Pervez Akhtar & M Junaid Khan

Department of Electronics and Power Engineering National University of Sciences and Technology, Islamabad, Pakistan [email protected], [email protected], [email protected]

Abstract Rapid growth in Wireless Network infrastructures creates a lot of noise/interference within limited available bandwidth. By using adaptive filtering technology to remove noise from a signal, the overall efficiency of the wireless communication system can be enhanced. In this paper, we present the performance analysis of two novel algorithms i.e. Affine Projection (AP) and Affine Projection Recursive Updating (APRU) with Sign-Data Least Mean Square (SDLMS), using Welch and Periodogram estimation techniques. The comparison is made on their noise cancellation performance. The power is measured employing subspace methods via Multiple Signal Classification (MUSIC) algorithm for each algorithm being implemented. Simulation results reveal that AP and APRU obtain best result for noise cancellation performance, estimation of power spectral density (PSD) and for power measurement as compared to SDLMS. Keywords: Adaptive Filtering, Affine Projection (AP) algorithm, Affine Projection Recursive Matrix Updating (APRU) algorithm and Sign-Data Least Mean Square (SDLMS) algorithm.

1. Introduction The future of wireless communication depends on the adaptive signal processing techniques and algorithms developed for noise cancellation and reduction, which allows efficient broadband communication favorable to both service provider and subscriber. Company after company are deploying their wireless networks, investing huge amount of capital in the telecom sector to cater the huge demand of subscribers to a wide variety of services. Using adaptive filtering technology, subscribers demands can be fulfilled in terms of voice quality and service connectivity. In the literature, different techniques/approaches and implementations are carried out for adaptive noise canceling [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]. Adaptive noise canceling involves two signals, a primary signal, comprised of input signal plus noise, and a reference signal, comprised of noise only. The idea based on theory is that if the reference signal is not identical to the noise component of the primary signal, it is at least correlated with it. It means that the correlated parts of the two signals are separated from the non-correlated parts, thereby greatly reducing the noise component of the primary signal. It is to be noted here that noise or interference cancellation and line enhancement or speech enhancement are having the same meaning in terms of system performance enhancement. The noise cancellation process removes the noise, leaving the signal whereas in adaptive line enhancement (ALE), the goal is to remove the noise signal from the measured signal to obtain the signal of interest [1] [3] [4] [13] [14] [15]. The paper is organized as follows: In Section 1, The Proposed adaptive noise cancellation mechanism is introduced. Section 2 deals with the System Model. The
IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

AP algorithm along with simulation results is discussed in section 3. Section 4 presents the APRU algorithm along with simulation results. Section 5 describes the SDLMS algorithm along with simulation results. A detailed discussion about the obtained results is introduced in section 6. Concluding remarks and future work are presented in section 7. 2. System Model In noise canceling systems, the goal is to extract a desired signal corrupted by interferences and noises as shown in Figure 1 [3].

Fig. 1. Noise Canceling Adaptive System

The delayed primary signal x is the combination of s + n0 and reference signal n1 is fed to the system. The adaptive filter produces an output y that is a close

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

replica of noise n1 . The given system produces an output, s + n0 y . The signal of interest d is equivalent to s + n0 . Error signal is obtained when signal of interest is subtracted from the adaptive filter output. In other words, we get the cleaned error signal . Then, system output is:

Therefore, y = n0 and = s , means error signal is true copy of input signal. In this case, minimizing output power causes the output signal to be perfectly free of noise. If the reference input is completely uncorrelated with the primary input, then filter will turn itself off and the output power will be:
[ 2 ] = E[( s + n0 ) 2 ] + 2 E[ y ( s + n0 )] + E[ y 2 ] (5) (6)
2

= s + n0 y

(1)

Squaring and taking expectation on both sides of (1) , and assuming that I. s , n0 , n1 and y are statistically stationary and have zero means. II. s uncorrelated with n0 and n1 . III. n1 is correlated with n0 . Then, we have:

[ 2 ] = E[( s + n0 ) 2 ] + E[ y 2 ]

Minimizing output power requires that E[ y ] be minimized, which is accomplished by making all weights zero, bringing E[ y 2 ] to zero [3].
3. Affine Projection (AP) Adaptive Algorithm Simulation Results

E[ ] = E[ s ] + E[(n0 y ) ] + 2 E[ s (n0 y )]
2 2 2

(2)

The adaptive filter is so adjusted to get mean square error (MSE) i.e. E[ 2 ] therefore the signal power
E[ s ] remains constant.
2

Affine projection algorithm uses direct matrix inversion, implemented with non-parametric spectrum estimation techniques such as Periodogram and Welch [10] [16] [17] [18].
Table 1. Common Parameters for AP, APRU and SDLMS Algorithms ParaAP APRU SDLMS meter Algorithm Algorithm Algorithm Method P 0.001 4 0.005 W 0.001 4 0.005 P 0.001 4 0.005 W 0.001 4 0.005 P 0.001 W 0.001 -

Accordingly, the minimum output power is: Emin [ 2 ] = E[ s 2 ] + Emin [(n0 y ) 2 ]

(3)

When the filter is adjusted so that E[ 2 ] is minimized,

E[(n0 y ) 2 ] is therefore also minimized.

From (1) , we have:

( s ) = (n0 y ) (4)

When E[( n0 y ) 2 ] is minimized then E[( s ) 2 ] is also minimized. This causes the output to be a best leastsquares estimate of the signal s for the given structure, adjustability of the adaptive filter and for the given reference input. It means that minimizing the total output power minimizes the output noise power also. Since the signal power E[ s 2 ] in the output remains constant, therefore, output signal-to-noise ratio (SNR) is increased, as it is the ratio between signal output power and noise output power. We see from (3) that the smallest possible output power is Emin [ 2 ] = E[ s 2 ] , when this is achievable, then E[(n0 y ) 2 ] = 0 .

Various parameters for implemented algorithms are listed in Table 1, for computer simulations using MATLAB version 7.6.0 (R2008a). Number of iterations for each algorithms being implemented is taken as 2000. The desired signal used for simulation purpose, is given by
S (t ) = sin(t ) where = 2 f and f is the frequency in Hertz. (7)

3.1. AP Algorithm with Periodogram (P) method This Algorithm is implemented for adaptive tracking of a sinusoidal input signal employing AP finite impulse response (FIR) adaptive filter that uses direct matrix inversion along with periodogram method for calculating signal value and for estimation of PSD as shown in Figures 2 and 3 respectively. Noise is removed from noisy signal and clean error signal (true copy of input signal) is obtained, chases the input signal by adapting filter weights.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

Adaptive Noise Cancellation 3 Input + Noise Error Signal Input Signal

1 Signal Value

-1

3.2. AP Algorithm with Welch (W) method In this section, an AP FIR adaptive filter is employed, which uses direct matrix inversion for adaptive tracking of a sinusoidal input signal using welch method for finding signal value and for estimation of PSD as shown in Figures 4 and 5 respectively. The error signal is obtained by minimizing the noises and thus obtaining a clean error signal (e). Welch obtains best PSD result of sinusoidal input signal than periodogram method.
20 Input + Noise Error Signal Input Signal

-2

-3 1900

1910

1920

1930

1940 1950 1960 Time Index

1970

1980

1990

2000

10 0 Power Spectral Density -10 -20 -30 -40 -50

Fig. 2. Adaptive tracking of a sinusoidal input, error and noisy signals employing AP FIR adaptive filter along with Periodogram method for calculating signal value
20 10 0 Power Spectral Density -10 Input + Noise Error Signal Input Signal

-60 -20 -30 -40 -50 -60

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

Fig. 5. Adaptive tracking of a sinusoidal input, error and noisy signal employing AP FIR adaptive filter along with Welch method for estimation of PSD

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

Fig. 3. Adaptive tracking of a sinusoidal input, error and noisy signal employing AP FIR adaptive filter along with Periodogram method for estimation of PSD
Adaptive Noise Cancellation 3 Input + Noise Error Signal Input Signal

3.3. Convergence Behavior of AP Algorithm The convergence behavior of actual and estimated coefficients for AP FIR adaptive filter is displayed in Figure 6, for same parameters as devised in Table 1. The convergence conditions imposed on step size is given by
0 1 (8)

max

1 Signal Value

-1

-2

where max is the largest eigen value. If is chosen to be very small, then convergence becomes slow. If is kept large, then convergence becomes fast, but stability becomes a problem. Therefore it is better to select within bounded conditions as defined in equation (8) . Therefore, it is said that is directly influence how and when convergence be achieved.
1910 1920 1930 1940 1950 1960 Time Index 1970 1980 1990 2000

-3 1900

Fig. 4. Adaptive tracking of a sinusoidal input, error and noisy signals employing AP FIR adaptive filter along with Welch method for finding signal value

3.4. Power Measurement via Music Algorithm The input and desired output power is measured for AP via MUSIC algorithm as shown in Figures 7 and 8 respectively.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

Actual and Estimated weights for APA 0.6 Actual Estimated

80 100

Pseudospectrum Estimate via MUSIC

0.5

0.4
60

Coefficient Value

0.3

Power (dB)

40

0.2

20

0.1
0

-0.1

-20

10

15 20 Coefficient #

25

30

35

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

Fig. 6. Convergence of coefficients employing AP FIR adaptive filters

Pseudospectrum Estimate via MUSIC 100

Fig. 8. Desired signal power by AP using MUSIC algorithm

70 Output Signal Error Signal Input Signal

60

80
50

60 Power (dB)
Magnitude 40

40

30

20

20

10

-20

20

40

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

60 80 Frequency (Hz)

100

120

140

Fig. 7. Input signal power using MUSIC algorithm

Fig. 9. Fourier analysis of input signal by AP algorithm

The input power is same for AP, APRU and SDLMS algorithms whereas the desired output power for each algorithm is different taking the same parameters as devised in Table 1.
3.5. Fourier analysis by AP algorithm

70 Output Signal Error Signal desired Signal

60

50

It is difficult to identify the frequency components in time domain. If these are converted to the frequency domain, then discrete Fourier transform of the required signals are computed as shown in Figures 9 and 10 respectively. The error signal is chasing the input signal with equal magnitude for both AP and APRU algorithms but deviating in case of SDLMS in magnitude and the error signal (e) is not chasing the input signal properly so minimization of noises is not satisfactory even though the same parameters have been taken as devised in Table 1. In this case desired signal, error signal and output signal is analyzed by taking the 128-point FFT. The error signal is chasing the desired signal with equal magnitude for both AP and APRU algorithms.

Magnitude

40

30

20

10

20

40

60 80 Frequency (Hz)

100

120

140

Fig. 10. Fourier analysis of desired signal by AP algorithm

4. APRU Algorithm - Simulation Results

APRU algorithm uses recursive matrix updating, implemented with non-parametric spectrum estimation techniques such as Periodogram and Welch.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

4.1. APRU Algorithm with Periodogram method

Adaptive Noise Cancellation 3 Input + Noise Error Signal Input Signal

Signal Value

In this section, APRU adaptive filter is employed, which uses recursive matrix updating instead of direct matrix inversion for adaptive tracking of a sinusoidal input signal using Periodogram method for calculating signal value and for estimation of PSD as shown in Figures 11 and 12 respectively. Noisy signal (input + noise) is filtered and noise is removed to get the clean error signal.
20 10 0 Power Spectral Density Input + Noise Error Signal Input Signal

-1

-2

-3 1900 -10 -20 -30 -40

1910

1920

1930

1940 1950 1960 Time Index

1970

1980

1990

2000

Fig. 13. Adaptive tracking of a sinusoidal input, error and noisy signals employing APRU along with Welch method for estimating signal value

20
-50

10
-60 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample) 0.9 1

Input + Noise Error Signal Input Signal

0 P ower S pec tral Dens ity -10 -20 -30 -40 -50

Fig. 11. Adaptive tracking of a sinusoidal input, error and noisy signals employing APRU along with Periodogram method for calculating signal value
Adaptive Noise Cancellation 3 Input + Noise Error Signal Input Signal

1 Signal Value

-60
0

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

-1

Fig. 14. Adaptive tracking of a sinusoidal input, error and noisy signals employing APRU along with Welch method for estimation of PSD

-2

-3 1900

1910

1920

1930

1940 1950 1960 Time Index

1970

1980

1990

2000

Fig. 12. Adaptive tracking of a sinusoidal input, error and noisy signals employing APRU along with Periodogram method for estimation of PSD

4.2. APRU Algorithm with Welch method

This adaptive filter uses recursive matrix updating for adaptive tracking of a sinusoidal input signal using welch method for estimating signal value and for estimation of PSD are shown in Figures 13 and 14 respectively.
4.3. Convergence Behavior of APRU Algorithm

Cleaned error signal is obtained from noisy signal (input + noise). Best PSD result of sinusoidal input signal is obtained by welch as compared to periodogram method. The same convergence conditions imposed on step size as given in (8) . is also known as adaptation constant. It is observed that AP and APRU converges at a rate faster than that of SDLMS, therefore it is more suitable for wireless communication systems than SDLMS algorithm. It is to be noted that almost same convergence of coefficients is obtained for AP and APRU algorithms.
4.4. Power Measurement via Music Algorithm

APRU FIR adaptive filter coefficients (actual and estimated) are displayed in Figure 15, for same parameters as listed in Table 1.

The desired output power is measured for APRU via MUSIC algorithm as shown in Figure 16. The more desired output power is obtained by APRU algorithm as compared to AP and SDLMS.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

Actual and Estimated weights for APRU 0.6 Actual Estimated 70 Output Signal Error Signal Input Signal

0.5

60

0.4 Coefficient Value

50

0.2

Magnitude 0 5 10 15 20 Coefficient # 25 30 35

0.3

40

30

0.1

20

10

-0.1

20

40

60 80 Frequency (Hz)

100

120

140

Fig. 15. Convergence of coefficients employing APRU FIR adaptive filters

Pseudospectrum Estimate via MUSIC 100

Fig. 17. Fourier analysis of input signal by APRU algorithm

70 Output Signal Error Signal desired Signal

60 80 50 60 Power (dB) Magnitude

40

40

30

20

20

10 0 0 -20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample) 0.9 1

20

40

60 80 Frequency (Hz)

100

120

140

Fig. 18. Fourier analysis of desired signal by APRU algorithm

Fig. 16. Desired signal power by APRU using MUSIC algorithm

4.5. Fourier analysis by APRU Algorithm

80 70 60 50 Magnitude 40 30 20 10 0 Output Signal Desired Signal Input Signal

It is difficult to identify the frequency components in time domain. If these are converted to the frequency domain, then discrete Fourier transform of the required signals are computed as shown in Figures 17 to 19 respectively. In this case desired signal, error signal and output signal is analyzed by taking the 128-point FFT. The error signal is chasing the desired signal with equal magnitude for both AP and APRU algorithms but deviating in case of SDLMS. Similarly, the relationship between original input signal and desired signal is shown in Figure 19 that shows close correlation with each other. Same result has also obtained using AP algorithm whereas SDLMS deviates that is not shown here.
5. SDLMS Algorithm - Simulation Results

20

40

60 80 Frequency (Hz)

100

120

140

Fig. 19. Fourier analysis of desired and input signals by APRU algorithm

SDLMS Algorithm is also implemented with nonparametric spectrum estimation techniques such as

Periodogram and Welch and is based on LMS FIR adaptive filter [19] [20].
5.1. SDLMS Algorithm with Periodogram method

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

SDLMS FIR adaptive filter is implemented along with periodogram method for tracking of a sinusoidal input signal for estimating signal value and for PSD of a signal as shown in Figures 20 and 21 respectively.
Adaptive Noise Cancellation Input + Noise Error Signal Input Signal Signal Value 3

Adaptive Noise Cancellation 3 Input + Noise Error Signal Input Signal

-1 1 Signal Value -2 0 -3 1900

-1

1910

1920

1930

1940 1950 1960 Time Index

1970

1980

1990

2000

-2

Fig. 22. Adaptive tracking of a sinusoidal input, error and noisy signals employing SDLMS algorithm along with Welch method for calculating signal value
1910 1920 1930 1940 1950 1960 Time Index 1970 1980 1990 2000
20 10 0 Power Spectral Density -10 -20 -30 -40 -50 -60 Input + Noise Error Signal Input Signal

-3 1900

Fig. 20. Adaptive tracking of a sinusoidal input, error and noisy signals employing SDLMS algorithm along with Periodogram method for estimating signal value

20 10 0 Power Spectral Density -10 -20 -30 -40 -50 -60 Input + Noise Error Signal Input Signal

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

Fig. 23. Adaptive tracking of a sinusoidal input, error and noisy signals employing SDLMS algorithm along with Welch method for estimation of PSD
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample) 0.9 1

Fig. 21. Adaptive tracking of a sinusoidal input, error and noisy signals employing SDLMS algorithm along with Periodogram method for estimation of PSD

The error signal is not chasing the input signal properly so minimization of noises is not satisfactory.
5.2. SDLMS Algorithm with Welch method

For this method, same formats of data as well as the filter coefficients are chosen as used for the SDLMS algorithm with periodogram method. The calculated signal value and PSD of an input, noisy and error signals are shown in Figures 22 and 23 respectively. In this case also, the error signal is not chasing the input signal properly as in the previous method so minimization of noises is not satisfactory.
5.3. Convergence Behavior of SDLMS Algorithm

coefficients for SDLMS FIR adaptive filter is displayed in Figure 24, for same parameters as listed in Table 1. The same convergence conditions imposed on step size as given in (8) . However considering a stronger criterion i.e. convergence in the mean square error, is inherently linked to the ensemble average which explains the practical importance in the adaptive filters that leads to express mean square error (MSE) produced by LMS based filters like SDLMS. Again it is proved that step size parameter kept to be small for stability. It is worth noting that convergence of actual coefficients obtained for SDLMS is same as for AP and APRU algorithms but different in magnitude for estimated coefficients with AP and APRU algorithms.
5.4 Power Measurement via Music Algorithm

The convergence behavior of actual and estimated

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

The desired output power is measured for SDLMS via MUSIC algorithm as shown in Figure 25. The desired

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

Actual and Estimated weights for SDLMS 0.6 Actual Estimated

50 60 Output Signal Error Signal Input Signal

0.5

0.4
40

Coefficient Value

Magnitude

0.3

30

0.2

20

0.1
10

-0.1

10

15 20 Coefficient #

25

30

35

20

40

60 80 Frequency (Hz)

100

120

140

Fig. 24. Convergence of coefficients employing SDLMS FIR adaptive filters

Pseudospectrum Estimate via MUSIC 100

Fig. 26. Fourier analysis of input signal by SDLMS algorithm

80 70 Output Signal Error Signal desired Signal

80

60 50 Magnitude 40 30

60 Power (dB)

40

20

20 10 0

-20

20

40

0.1

0.2

0.3 0.4 0.5 0.6 0.7 0.8 Normalized Frequency ( rad/sample)

0.9

60 80 Frequency (Hz)

100

120

140

Fig. 27. Fourier analysis of desired signal by SDLMS algorithm

Fig. 25. Desired signal power by SDLMS using MUSIC algorithm

output power obtained by SDLMS algorithm is less as compared to APRU.

5.5 Fourier analysis by SDLMS Algorithm

Again it is difficult to identify the frequency components in time domain. If these are converted to the frequency domain, then discrete Fourier transform of the required signals are computed as shown in Figures 26 and 27 respectively. In FFT, the desired signal, error signal and output signal are analyzed. The error signal is chasing the desired signal with equal magnitude for both AP and APRU algorithms. But deviating in case of SDLMS in magnitude and the error signal is not chasing the input signal properly so minimization of noises is not satisfactory even though the same parameters have been taken as devised in Table 1.
6. Discussion and Results

AP, APRU and SDLMS algorithms are compared on the basis of their noise cancellation performance in terms of

MSE, PSD, weights estimation and power measurements. From Figures 2, 4, 11, 13, 20 and 22, it is seen that MSE for AP, APRU are approximately equal and different for SDLMS. Therefore, we can say that AP, APRU algorithms are best to minimize noises to get cleaned error signal as compared to SDLMS. AP and APRU algorithms employing welch method obtain best result of PSD for sinusoidal input signal and error signal rather than periodogram method using same parameters as shown in Figures 3, 5, 12, 14, 21 and 23. The error signal is the true copy of input signal. Further, the rate of convergence is observed fast in AP and APRU as compared to SDLMS. Therefore, AP and APRU algorithms with welch method performs better than SDLMS in both MSE improvement and PSD enhancement. APRU is, basically AP based algorithm, which uses recursive matrix updating, so AP algorithm is found best in terms of MSE, PSD and is fast in convergence; hence it is more suitable for wireless communication systems than SDLMS algorithm.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

It is to be noted that if value of is varied from 0.001 to 0.0001 for SDLMS algorithm, we can get a better simulation result than the one shown in Figures 20 to 23. In this case, we get best performance in terms of noise cancellation and PSD. It is worth noting that convergence of actual coefficients obtained for SDLMS is same as for AP and APRU algorithms but different in magnitude for estimated coefficients, as shown in Figures 24, 6, and 15. Taking the analysis of power measurement, it is clear from the Figures 8, 16 and 25 that desired output power for APRU is much better than AP and SDLMS, therefore APRU exercises good performance in this regard. It is very difficult to identify the frequency components by looking at the original input signal, error signal, desired signal and output signal in time domain. In FFT, it is found easily, by looking into the Figures 9, 10, 17, 18, 19, 26 and 27, the error signal is chasing the input signal with equal magnitude for both AP and APRU algorithms but shows deviation in case of SDLMS in magnitude as well as the error signal is not tracking the input signal exactly. Similarly in case of desired signal, the error signal follows the desired signal with equal magnitude for both AP and APRU algorithms but is deviating in case of SDLMS in magnitude and the error signal is not chasing the input signal exactly so minimization of noises is not satisfactory. Similarly, the relationship between original input signal and desired signal is shown in Figure 19 that describes close correlation with each other. Same result has also obtained using AP algorithm whereas SDLMS deviates. Therefore, we can say that AP based algorithm performs better than LMS based adaptive filters in terms of Fourier analysis, noise cancellation, PSD, output power and fast convergence.
7. Conclusions

[2] [3] [4] [5]

[6]

[7]

[8]

[9] [10]

[11] [12]

[13]

A detailed description of the main ideas explained in the paper and it is found that AP and APRU algorithms perform better in noise cancellation either plotted in time domain or frequency domain. Similarly, both of these algorithms provide best outcome in PSD and weight estimation. Therefore, AP based algorithms have achieved good performance as compared to LMS based algorithm in terms of noise cancellation, PSD, desired output power and fast convergence. In future, AP based algorithms will be extended to perform real-time adaptive noise cancellation in acoustic environments.
8. References
[1] Michael A. Arbib, The Handbook of Brain Theory and Neural Networks, Part III, Articles, Noise Canceling and Channel

[14]

[15]

[16]

[17]

[18] [19]

Equalization, B Widrow and Michael A Lehr , A Bradford Book, The Mit Press, Cambridge, Massachusetts, London, England. Simon Haykin, Adaptive Filter Theory, Fourth edition (Pearson Eduation, Inc., 2002). B. Widrow, S.D. Stearns, Adaptive Signal Processing (Pearson Eduation, Inc., 1985). Dag Strannneby, Digital Signal Processing: DSP and Applications, Oxford, (Reed Elsevier plc group, 1st Published 2001). M. Sheikh M. Algunaidi, M. A. Mohd Ali, K. B. Gan and E. Zahedi Fetal Heart Rate Monitoring Based on Adaptive Noise Cancellation and Maternal QRS Removal Window, European Journal of Scientific Research, ISSN 1450-216X, Vol.27, No.4 (2009), pp.565-575, EuroJournals Publishing, Inc. 2009. Mark Yelderman, Member, IEEE, Bernard Widrow, Fellow, IEEE, John M. Cioffi, Student Member, IEEE, Edward Hesler, Member, IEEE, and Jeffrey A. Leddy, ECG Enhancement by Adaptive Cancellation of Electrosurgical Interference, IEEE Transactions On Biomedical Engineering, VOL. ME-30, NO. 7, July 1983. Jun-II Sohn and Minho Lee, Selective Noise Cancellation Using Independent Component Analysis, O. Kaynak et al. (Eds.), ICANN/ICONIP 2003, LNCS 2714, pp. 530-537, SpringerVerlag Berlin Heidelberg 2003. Alan V. Oppenheim, Ehud Weinstein, Kambiz C. Zangi, Meir Feder and Dan Gauger, Single-Sensor Active Noise Cancellation, IEEE. Transactions on Speech and Audio Processing, VOL. 2, NO. 2, pp. 285-290, April 1994. Mine Kalkan and Feza Kerestecioglu, Optimal Diversity Combining Under Correlated Noise in Mobile Radio, Telecommunication Systems 8(1997), pp. 303-317,1997. Kyu-Young Hwang, Student Member, IEEE, and Woo-Jin Song, Member, IEEE, An Affine Projection Adaptive Filtering Algorithm with Selective Regressors, IEEE Transactions on Circuits and Systems II: EXPRESSBRIEFS, Vol. 54, NO.1, January 2007. O. P. Sharma, V. Janyani and S. Sancheti, Recursive Least Square Adaptive Filter a Better ISI Compensator, International Journal of Electronics, Circuits and System, 2009. Thato Tsalaile and Saeid Sanei, Separation of Heart Sound Signal from Lung Sound Signal by Adaptive Line Enhancement, 15th European Signal Processing Conference (EUSIPCO 2007), Poznan, Poland, September 3-7, 2007. Yan-wei Gong, Xiao-jun Ji, Feng-yi Huang and Xiao-hong Ruan, Fast Affine Projection Algorithm for Adaptive Noise Canceling and its Application on the Fetal Electrocardiogram Extraction, Journal of Shanghai Jiaotong University (Science), Volume 14, NO. 6, December, 2009, ISSN 1007-1172, pp. 690-694, 2009. Sinisa Pajevic, George H. Weiss, Kenneth W. Fishbein, and Richard G.S. Spencer, Use of the Adaptive Line Enhancement Filter for SNR Improvement in NMR Spectroscopy, Proc. Intl. Sot. Mag. Reson. Med. 8 (2000), pp. 1778-1780, 2000. Ersoy Kelebekler and Melih nal, White and Color Noise Cancellation of Speech Signal by Adaptive Filtering and Soft Computing Algorithms, A. Sattar and B.H. Kang (Eds.): AI 2006, LNAI 4304, pp. 970 975, 2006. AIeksandrs BeIvovs and Juris Smotrovs, A Criterion for Attaining the Welch Bounds with Applications for Mutually Unbiased Bases, J. Calmet, W. Geiselmann, J. Muller-Quade (Eds.), Beth Festschrift, LNCS 5393, pp. 50-69, SpringerVerlag Berlin Heidelberg 2008. Sunder. G. Sankaran, Student Member, IEEE, and A. A. (Louis) Beex, Senior Member, IEEE, Convergence Behavior of Affine Projection Algorithms, IEEE Transactions on Signal Processing, Vol. 48, NO. 4, pp. 10861096, April 2000. Hyun-Chool Shin and Ali H. Sayed, Transient Behavior of Affine Projection Algorithms, ICASSP 2003, 0-7803-7663-3/03, P VI-353-356, 2003 IEEE. Kevin S. Biswas and Jason G. Tong, Design and Implementation of an Adaptive LMS-based Parallel System for Noise

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2)

Cancellation, T. Sobh and K. Elleithy (eds.), Advances in Systems, Computing Sciences and Software Engineering, pp. 403409 2006 Springer. [20] Md. Zia Ur Rahman, Rafi Ahamed Shaik and D V Rama Koti Reddy, Noise Cancellation in ECG Signals Using Computationally Simplified Adaptive Filtering Techniques: Application to Biotelemetry, Signal Processing: An International Journal (SPIJ), ISSN 1985-2339, VOL. NO. 3, Issue 5, pp. 1-12, November 2009. Muhammad Yasin is enrolled for PhD in the field of electrical engineering majoring in telecommunication in Pakistan Navy Engineering College, National University of Science and Technology, Karachi (NUST), Pakistan. He is working in Pakistan Navy as naval officer in the capacity of communication engineer since 1996. His research interests include signal processing, adaptive filtering,

implementation of communication networking and its performance evaluation. He has received a B.Sc. degree in electrical engineering with Honour from NWFP University of Engineering and Technology, Peshawar (1994) and M.Sc. degree in electrical engineering from NED, University of Engineering and Technology, Karachi (2006). He has also done a Master degree in Economics (2002) from University of Karachi. In the past, he is involved in implementation of ISO 9000 on indigenous project of AGOSTA 90B Class Submarines along with French engineers. Currently, he is working on indigenous project of Acoustic System Trainer, being used for imparting Sonar related training.

IST Transactions of Computer Systems - Theory and Applications, Vol. 1, No. 1 (2) ISSN 1913-8369, pp. 1-10, 2010