ACQUISITION OF SARSAT INFORMATION USING AN AUTO-CORRELATION BASED ADAPTIVE LINE ENHANCER

Transcription

1 International Journal of Information Acquisition Vol. 2, No. 4 (25) c World Scientific Publishing Company ACQUISITION OF SARSAT INFORMATION USING AN AUTO-CORRELATION BASED ADAPTIVE LINE ENHANCER S. E. EL-KHAMY Department of Electrical Engineering, Faculty of Engineering Alexandria University, Alexandria 21544, Egypt elkhamy@ieee.org M. M. HADHOUD Department of Information Technology, Faculty of Computers and Information Menoufia University, 32511, Shebin Elkom, Egypt mmhadhoud@yahoo.com M. I. DESSOUKY, B. M. SALAM and F. E. ABD EL-SAMIE Department of Electronics and Electerical Communications Faculty of Electronic Engineering, Menoufia University 32952, Menouf, Egypt fathi sayed@yahoo.com Received 28 May 25 Accepted 7 October 25 This paper develops a new approach to the detection of the emergency locator transmitter (ELT) signal using an adaptive line enhancer (ALE). The input signal to the ALE is replaced by its autocorrelation function (ACF) because noise affects the central samples of the autocorrelation function only while it affects all samples of the noisy signal. This gives the ALE the ability to get rid of noise, easily. The output ACF is then used in the spectral estimation and detection of the ELT signal. This approach is related to the signal processing using Higher Order Statistics (HOS) since the ACF is the used input. The paper also compares the results of the new approach to other different previously used methods foe ELT signal detection. The first method uses the signal x k as an input to the ALE, and the second method uses x 2 k. Results illustrate the superiority of the proposed method over the other two methods. A comparison study between the performance of two types of ALE; Fixed pole radius and variable pole radius ALEs in detecting the ELT signal, is introduced. Keywords: SARSAT information; emergency locator transmitter; adaptive line enhancer; autocorrection function. 291

2 292 S. E. El-Khamy et al. 1. Introduction Adaptive filters can be finite impulse response (FIR) filters or infinite impulse response (IIR) filters. The majority of the work on adaptive line enhancers involves the use of FIR filters due to their stability advantage [David, 1985; Wang & Wang, 1998; Tichavsky & Nehorai, 1997; Widraw et al., 1975]. On the other hand, IIR filters have also been proposed to achieve high quality (Q) factors [Ahmed et al., 1984; Friendlander & Port, 1989; Johnson, 1984; Chang & Glover, 1993; Boroujeny, 1997]. In sinusoidal signal detection, or in the case of signals that can be analyzed into a few number of sinusoids, high-q factor resonant filters are required. This property of high-q is an inherent property and an advantage of the IIR filters over the FIR filters. High-Q filters require poles that are very close to each other and lie inside the unit circle. The factor of primary concern associated with the use of IIR filters is to maintain the stability of the filter during the adaptation process. The convergence characteristics on a multimodal performance surface, which results from the multiple line frequencies, is another concern [Ahmed et al., 1984; Friendlander & Port, 1989; Johnson, 1984; Chang & Glover, 1993; Boroujeny, 1997]. The usage of a simple two-pole structure alleviates the problem of maintaining filter stability since the pole radius can easily be constrained to be less than unity. The complex conjugated poles of the ALE structure may be fixed or variable [David, 1985; Ahmed et al., 1984]. In this paper, we consider two types of ALE structures, the fixed pole radius ALE and the variable pole radius ALE. Both structures are presented and explained in Sec. 2. ALE structures whether fixed pole or variable pole are used in the detection of signals immersed in noise and they have the ability to track the signal in the presence of noise [Ghogho et al., 1998; Rickard & Zeidler, 1979; Zeidler, 199; Yoganandam et al., 1988]. Some attempts have been carried out to use ALE structures to detect frequency modulated signals [Chew et al., 1994]. In this paper, we introduce a new approach for the detection of the ELT signal merged with noise using the ALE structures rather than traditional spectral estimation techniques used to detect this signal [Dessouky, 1998, 1999; Dessouky & Carter, 1988]. The ELT signal is a low powered emergency radio transmitter radiated signal of about 1 MW power with amplitude modulated having a carrier frequency of either MHz or optionally 243 MHz. This signal is used in the Save And Rescue Satellite (SARSAT) system in the cases of emergency to alleviate the problem of location determination. The ELT signals are processed using a bandpass processor implementation in which the signal is mixed down to the frequency range of to 25kHz, which normally covers the vast majority of ELT signals [Dessouky, 1998, 1999; Dessouky & Carter, 1988]. We propose the usage of the ACF of this ELT down converted signal rather than the signal itself as an input to the ALE because the effect of noise on the ACF is limited as compared to its effect on the signal itself. In addition, we investigate the use of x k, x 2 k as inputs to both the fixed and variable pole radius ALEs. A comparison study between the performances of all of these cases is introduced. This paper is organized as follows. In Sec. 1, we introduce the input ELT signal. In Sec. 2, we present a description of the IIR-ALE structures with fixed and variable pole radius. Section 3 introduces the use of both x 2 k and the new approach which uses the ACF instead of x k as inputs to the ALE structures. Section 4 introduces the frequency error concept. Section 5 gives the simulation results. Finally, the concluding remarks are given in Sec The ELT signal The ELT signal that is immersed in white Gaussian noise is given by the following expression [Dessouky, 1998, 1999; Dessouky & Carter, 1988]. s(t) =A[1 + m(t)] cos(2πf c t + θ)+n(t), (1) where A is the carrier amplitude, f c is the carrier frequency, θ is the phase angle,

3 Acquisition of SARSAT Information 293 m(t) is the modulating signal, n(t) is the additive white Gaussian noise. The modulating term can be classified as either a sine wave or a pulse shaped function. To formulate a sinusoidal-modulated ELT signal, we define: m(t) =µ sin(φ i (t)), (2) where µ is the modulation factor which can be varied from 85% to 1% and φ i (t) isgivenby: φ i (t) =2π f in (t)dt. (3) The instantaneous frequency, f in (t) is approximated in the least squares sense by a linear function and is given as: ( ) t f in (t) = 14 7, (4) where T r is the repetition period (.25 s) of the signal. Solving these equations, the linear sweep sinusoidally-modulated samples ELT signal S L (n) which is corrupted by noise is given by: S L (k) =A[1 + µ sin(2π(14k 14k 2 T r +.75))] sin(2πf c k + θ)+n(k), (5) where k is the sampling time instant. The received noise level of the ELT signal is sometimes given in terms of the noise density ratio in db-hz rather than in db. The SNR in db is related to the CNDR in db-hz using the following relation [Dessouky, 1998, 1999; Dessouky & Carter, 1988]. SNR(dB) = CNDR(dB Hz) 1 log 1 (B), (6) where B is the bandwidth of the signal (in this paper, we assume B =25kHz). The characteristics of the spectrum of the ELT signal are very important, since the probability of locating the downed aircraft is closely related to the quality of the ELT spectrum itself. Since the spectral analysis of the ELT signal is the central problem, different methods of spectral estimation using ALE can provide advantages in estimating the ELT carrier frequency as compared to traditional spectral estimation techniques. 2. The IIR-ALE Structures 2.1. The fixed pole radius ALE The transfer function of the fixed pole radius ALE is given as follows [David, 1985; Ahmed et al., 1984]: ( ) 1 r 2 wz (1 r 2 ) 1+r H(z) = 2 z 2 wz + r 2, (7) where r is the pole radius, <r<1andisthe filter weight w < 2r. In this ALE structure, the error signal is used in the adaptation of the filter transfer function until convergence is achieved. In this algorithm w k is drawn adaptively to minimize the mean squared error [David, 1985; Ahmed et al., 1984]. The filter output is given by: ( 1 r 2 ) y k = 1+r 2 w k x k 1 (1 r 2 )x k 2 + w k y k 1 r 2 y k 2. (8) The error output is defined as: e k = x k y k. (9) The adaptation algorithm we consider here is defined by the following equations [Ahmed et al., 1984]: w k+1 = w k + µe k α k /ψ k, (1) ( ) 1 r 2 α k = 1+r 2 x k 1 + y k 1 + w k α k 1 r 2 α k 2, (11) ψ k = vψ k 1 +(1 v)α 2 k. (12) µ is a constant which controls the rate of convergence and ψ k represents a smoothed estimate of the instantaneous power in α k where, <v<1, is a smoothing parameter [Ahmed et al., 1984] The variable pole radius ALE It is a second order system that is used as an ALE and allows the complex conjugated poles to adapt around a circle of radius r which is inside and concentric to the unit circle. Selection of a radius near unity results in a filter response with a sharp resonant peak, which is a desirable characteristic for the ALE. Starting with a relatively

4 294 S. E. El-Khamy et al. small radius and allowing it to increase when an incoming signal is detected, the rate of convergence is increased while maintaining the desired SNR improvement. Two error signals are used in the adaptation of the transfer function of the filter and the gain b [David, 1985]. The transfer function of the variable pole radius ALE is given as follows [David, 1985]: (1 r 2 )(a z 1 ) H(z) = 1 (1 + r 2 )az 1 + r 2, (13) z 2 where r specifiesthepoleradiusanda specifies a normalized angular frequency [David, 1985]. The normalized recursive least mean square (NRLMS) adaptive algorithm is used to adjust the coefficient a and is summarized as follows: y k =(1 rk 2 )(a kx k 1 x k 2 ) +(1+rk 2 )a ky k 1 rk 2 y k 2 (14) e k = x k y k (15) α k =(1 rk 2 )x k 1 +(1+rk 2 )y k 1 +(1+rk 2 )a kα k 1 rk 2 α k 2 (16) ψ k = ηψ k 1 +(1 η)α 2 k, <η<1 (17) a k+1 = a k + ρe kα k ψ k (18) The adjustment of the pole radius r is controlled by the detection parameter b which is also adapted according to the normalized recursive least mean square (NRLMS) algorithm as follows [David, 1985]: e k = x k b k y k, (19) β k = ηβ k 1 +(1 η)yk 2, <η<1, (2) b k+1 = b k + µe k y k. (21) β k β k is an estimate of the power at the output, ρ is a constant which controls the rate of convergence, ψ k represents a smoothed estimate of the instantaneous power α k and <η<1, is a smoothing parameter. This gain element b k varies between zero and one depending upon whether a signal component has been detected or not. Thus, a threshold value can be selected to define a lock/unlock status for the detector. If the value of b k exceeds the threshold, the system is considered to be locked onto an incoming signal. If b k is below the threshold value, the ALE is searching for a signal. The lock/unlock status determined by b k is used to control the adjustment of the coefficient r. The method of adjusting the coefficient r is summarized as follows [David, 1985]. Lock : Unlock : { r<rmax r k+1 = r k + δ r r = r max r k+1 = r k, { r>rmin r k+1 = r k δ r r = r min r k+1 = r k, (22) (23) where r max and r min are the maximum and minimum allowable values of the pole radius, respectively and δ r is an infinitesimal change in the pole radius. Choosing r max to be very close to unity results in a higher SNR improvement. In selecting r min, the minimum signal SNR should be considered. For low SNR, a smaller r min may be necessary. 3. The New Approach Using the ACF The objective of using the ALE whose algorithm is given in Sec. 2 is to minimize the following cost function: J = x k y k 2. (24) In order to minimize this function, only the mean and variance of the input signal are required. Higher order statistics (HOS) (or moments) have interesting properties that make them useful for many non-linear signal-processing applications. For instance, the third and higher order (HO) moments of Gaussian random process are zero. Hence, HO moments are used to separate non-gaussian signals from additive, independent Gaussian noise [Friendlander & Port, 1989; Ghogho et al., 1998]. In this paper we consider a new approach where the input signal to the ALE is replaced by its autocorrelation function ACF as a pre-processor. Usually, the input signal to the ALE is the noisy signal x k. Recently, [Ghogho et al., 1998] has considered the input signal to the ALE as x 2 k. In this later case the cost function to be

5 Acquisition of SARSAT Information 295 minimized is given as [Johnson, 1984; Chang & Glover, 1993]: J = x 2 k y k 2, (25) where y k is the output of the ALE. It has also the same units as x 2 k. Minimization of this function requires dealing with HO moments of the input signal x k. The results presented in [Friendlander & Port, 1989; Ghogho et al., 1998] concluded that the ALE performance when using x 2 k as an input is superior in producing better SNR to that when using x k as an input to the ALE if certain conditions are satisfied. The power spectrum of a given time sequence is estimated using the ALE output. In this paper, we use the ACF R k as a preprocessor in estimating the power spectrum for a given time sequence using the ALE algorithm. The ACF is a second order measure of the statistics of the input sequence. We refer to this operation as ACFALE technique. The cost function to be minimized for this technique is given as follows: J = R k y k 2, (26) where y k is the output of the ALE. It has also the units as that of ACF. The sample estimate of the auto-correlation function R k of a sequence x k, k =, 1,...,N 1isgivenas: N 1 k R k = 1 x l x k+l k =, 1,...,N 1. N l= (27) The averaging in this estimate is expected to reduce the noise level before applying the ALE to the signal. Thus, the ACF has an interesting property of low noise effect on all samples except the central samples. This is because the white gaussian noise has an autocorrelation in the form of σnδ(k). 2 For ideal white gaussian noise this ACF has no value except at k = which means that the noise effect on the ACF of the signal contaminated in noise is small. If this ACF is used as the input to the ALE, it can get rid of noise effect easily. 4. Frequency Error Study This section is concerned with investigating the accuracy of the estimated frequency for ALE and ACFALE processors of both fixed and variable pole radii. The frequency error is estimated according to the following relation: f e = f c f es, (28) where f e is the frequency error, f c is the carrier frequency, f es is the estimated frequency after processing. The estimation frequency error is a measure of the accuracy of the used adaptive processor. It is of great importance to minimize this error as much as possible because the estimated frequency is used in the determination of the emergency location. A comparison study between the performances of ALE and ACFALE techniques is presented in this paper from the frequency error point of view. 5. Simulation Results 5.1. ALE results In this paper, we process a single ELT signal. We have chosen a sinusoidal-modulated ELT signal with a frequency of 1 khz. Figures 1, 2 and 3 illustrate a single sinusoidal-modulated ELT signals with CNDRs of 4, 35 and 3 db-hz, respectively. It is clear from these figures that the ELT signal is detectable at the CNDR of 4 db-hz and it is deteriorated as the CNDR is reduced until it is completely masked with noise when its CNDR is 3 db-hz. So, we have chosen a signal at CNDR = 3 db-hz as our threshold for processing. Fig Spectrum of ELT signal, CNDR = 4 db-hz.

6 296 S. E. El-Khamy et al. Fig Spectrum of ELT signal, CNDR = 35 db-hz r =.97, µ =.1, w =.5 andv =.99 for the fixed pole radius algorithm and rmax 2 =.9, rmin 2 =.6, η =.95, δ =.5, ψ = 1, β = 1, b =.4, µ =.1 and ρ =.1 for the variable pole radius algorithm. The ELT signal used in all experiments is of CNDR equal to 3 db-hz. The data length of each data block used in all experiment is 512 points. Figures 5 and 6 give the output of fixed and variable pole radius algorithms, respectively for a single block input. It is clear that the data length is not enough in both cases to allow the signal to be detected. The application of 1 times ensemble averaging on the results given in Figs. 5 and 6 is introduced in Figs. 7 and 8, respectively. It is revealed that the ensemble averaging improves the detectability of the signal in both cases and reduces the noise to a lower level in the fixed pole case than in the variable pole case. The side bands of the ELT signal still Fig Spectrum of ELT signal, CNDR = 3 db-hz Fig. 4. Spectrum of 1 times coherent time averaging of ELT signal, for CNDR = 3 db-hz. The 1 times equal weight coherent time averaging is tested as a method for improving the detectability of the signal as in Fig. 4. It is found that its effect is limited to SNR improvement of about 15 db. We begin to investigate the performance both fixed and variable pole radius algorithms. In all of our simulation work, we have used Fig. 5. input Output of fixed pole algorithm for single block Fig. 6. Output of variable pole algorithm for single block input.

7 Acquisition of SARSAT Information Fig. 7. Ensemble averaging for results of Fig. 5. Fig. 9. Output of fixed pole ALE for 2 block input Fig. 8. Ensemble averaging for results of Fig. 6. Fig. 1. Output of variable pole ALE for 2 block input. exist, which means that the band of the ALE is relatively wide. To allow the filter to be more tuned to the carrier frequency of the ELT signal and reject its side bands, the data length of the input to the ALE is made longer. Figures 9 and 1 are the outputs of the fixed and variable pole radius algorithms for 2 block of data input respectively and without the application of ensemble averaging. This technique succeeds with the fixed pole radius algorithm more than with the variable pole radius case. The application of ensemble averaging on the results of Figs. 9 and 1 is illustrated in Figs. 11 and 12. It is clear that the noise is reduced to a smaller level than that given before. Generally, we conclude that the fixed pole radius algorithm performance is preferred to that of the variable pole radius algorithm using the signal as the input to the ALE. In [Chang & Glover, 1993], x 2 k is examined as an input to the ALE instead of x k in the cases of Fig. 11. Ensemble averaging for result in Fig. 9. additive and multiplicative noise. We also examine in this paper x 2 k as an input to the ALE to compare it with our approach. The spectrum of the input signal calculated using x 2 k is given in Fig. 13. It is clear that no information about the carrier frequency at 1 khz is shown in this figure. The spectrum calculated from averaging of x 2 k is drawn in Fig. 14. No more improvement

8 298 S. E. El-Khamy et al Fig. 12. Ensemble averaging for result in Fig. 1. Fig. 15. Spectrum of input signal using x 2 k (variable) Fig. 13. Spectrum of input signal using x 2 k (fixed). Fig Spectrum of output signal using x 2 k (variable) of x 2 k as an input to the ALE is unfeasible in the case of additive white Gaussian noise contamination. Fig Spectrum of output signal using x 2 k (fixed). in carrier estimation is gained using this averaging technique and the signal is still completely immersed in noise. The application of x 2 k as an input to both fixed and variable pole radius algorithms does not result in the detection of the 1 khz signal component as illustrated in Figs. 15 and 16 respectively. So, it is clear that the application 5.2. ACFALE results The spectrum of the input signal to ALE can be calculated either from the sample input or from the ACF of the sample input. The spectrum estimated using the ACF is given in Fig. 17. Also, the spectrum can be calculated using averaged ACFasinFig.18,whichissmootherthanthat in Fig. 17. Our suggested algorithm in this paper is to use the ACF as a pre-processor before the ALE (ACFALE). The results of ACFALE using a single block of data without ensemble averaging is illustrated in Figs. 19 and 2 for both fixed and variable pole radius, respectively. It is clear that the output for the fixed pole radius case

9 Acquisition of SARSAT Information Fig. 17. Spectrum using ACF. Fig. 2. ACFALE result (variable pole). Single block, no ensemble average Fig. 18. Spectrum using average ACF. Fig. 21. ACFALE result (fixed pole). Single block, ensemble average Fig. 19. ACFALE result (fixed pole). Single block, no ensemble average. is better than that given using the ALE only in Fig. 4. The output in the case of variable pole radius is still unimproved. The process of applying the ensemble averaging on the results given in Figs. 19 and 2 is shown in Figs. 21 and 22. The signal at 1 khz becomes detectable in both cases and the output in the fixed pole case becomes smoother. Increasing the length of the Fig. 22. ACFALE result (variable pole). Single block, ensemble average. input data to 2 blocks gives the spectrum illustrated in Figs. 23 and 24. From these obtained results shown in Figs. 23 and 24, we note that for both fixed and variable pole radius ACFALE algorithms, the spectrum is enhanced and the signal can be detected easily as compared to the spectrum results depicted in Figs. 19 and 2,

10 3 S. E. El-Khamy et al Fig. 23. ACFALE result (fixed pole) 2 blocks, ensemble average. respectively. In the case of variable pole radius, the best possible output is obtained in Fig. 24 with a very smooth curve. Thus, the application of ACFALE algorithm gives the best results with the variable pole radius case for long data sequences and gives Fig. 24. ACFALE result (variable pole) 2 block, ensemble average. the best results for the fixed pole case using the ensemble averaging Frequency error results The frequency error study for all of the discussed algorithms is examined in this paper. In this Freq. Error Freq. Error f(khz) (a) f(khz) (b) Freq. Error Freq. Error f(khz) (c) f(khz) Fig. 25. Frequency error study: (a) Fixed pole radius 2 block length with 1 times ensemble averaging, (b) variable pole radius 2 block length with 1 times ensemble averaging, (c) fixed pole radius 2 block length with ACF applied ACFALE, (d) variable pole radius 2 block length with ACF applied ACFALE. (d)

11 Acquisition of SARSAT Information 31 study, the carrier frequency is varied from to 25 khz in 5 Hz steps and the frequency error f e is calculated for all of the used algorithms. Figure 25 gives the frequency error results for the carrier frequency varied from 9.5 khz to 15.5 khz. From this study, it is clear that the frequency error for both ALE and ACFALE techniques with fixed and variable pole radius varies in the range of about 5 Hz up to 5 Hz which is an acceptable range of frequency error. The smaller the frequency error, the more accurate the place of emergency is determined using the detected signal. It is also evident that ELT signals with frequencies near the band center value suffer less values of frequency error. Both ALE and ACFALE techniques succeed in detecting the signals at different carrier frequencies with minimum frequency error near the data center with the same data and this is a desired characteristic in any adaptive processor. 6. Conclusion Different adaptive algorithms of fixed and variable pole radius ALE have been applied to detect an ELT signal immersed in white Gaussian noise. The fixed pole radius algorithm has succeeded in detecting the signal but the variable pole radius algorithm had adverse results. The application of the squared incoming signal does not succeed in detecting the weak signal with both algorithms. The new approach ACFALE improves the ELT signal detectability, especially for the case of variable pole radius. The efficiency of each of these algorithms is studied from the point of view of frequency error. Both of them have succeeded in signal detection with a small frequency error. References Ahmed, N., Hush, D., Elliot, G. R. and Fogler, R. G. [1984] Detection of multiple sinusoids using an adaptive cascade structure, Proc. ICASSP, March 19 21, San Diego, CA, Vol. 2. Boroujeny, B. F. [1997] An IIR adaptive line enhancer with controlled bandwidth, IEEE Trans. Signal Proc. 45(2), Chang, J. and Glover, J. R. [1993] The feedback adaptive line enhancer: A constrained IIR adaptive filter, IEEE Trans. Signal Proc. 41(11), Chew, K. C. Soni, T. Zeidler, J. R. and Ku, W. H. [1994] Tracking model of an adaptive lattice filter for linear chirp FM signal in noise, IEEE Trans. Signal Proc. 42(8), David, R. A. [1985] An extended algorithm for the second order ALE, Proc. 19th Alsilomar on Circuits and Systems, Computer. Dessouky, M. I. [1998] Processing of SARSAT signals using modified digital Lerner filter, 15th National Radio Science Conference NRSC, Cairo, Egypt. Dessouky, M. I. [1999] Processing of multiple SARSAT signals using modified Lerner filter, 1st Minia International Conference for Advanced Trends in Engineering MICATE 99, MINIA, Egypt. Dessouky, M. I. and Carter, C. R. [1987] Spectral analysis of ELT signals for SARSAT, IEEE Trans. Aerospace and Electronic Systems AES- 23(5), Dessouky, M. I. and Carter, C. R. [1988] A baseband processor for SARST signals, Canadian Electrical Engineering Journal 13(2), Friendlander, B. and Port, B. [1989] Adaptive IIR algorithms based on high-order statistics, IEEE Trans. on Acoustics, Speech and Signal Processing 37(4), Ghogho, M. Ibnkahla, M. and Bershad, N. J. [1998] Analytic behavior of the LMS adaptive line enhancer for sinusoids corrupted by multiplicative and additive noise, IEEE Trans. on Signal Proc. 46(9), Johnson, C. R. [1984] Adaptive IIR filtering: Current results and open issues, IEEE Trans. Information Theory IT-3, Rickard, J. T. and Zeidler, J. R. [1979] Second order output statistics of the adaptive line enhancer, IEEE Trans. Acoustics, Speech and Signal Proc. ASSP-27(1), Tichavsky, P. and Nehorai, A. [1997] Comparative study of four adaptive frequency trackers, IEEE Trans. Signal Proc. 45(6), Wang, T. and Wang, C. L. [1998] A new twodimensional block adaptive FIR filtering algorithm and its application to image restoration, IEEE Trans. Image Proc. 7(2), Widraw, B. Glover, J. R. Mc-Cool, J. M. Kaunitz, J. Dong, E. and Goodlin, R. C. [1975] Adaptive

12 32 S. E. El-Khamy et al. noise canceling, principles and applications Proc. IEEE 63(12). Yoganandam, Y., Reddy, V. U. and Kailath, T. [1988] Performance analysis of the adaptive line enhancer for sinusoidal signals in broad-band noise, IEEE Trans. Acoustics, Speech and Signal Proc. 36(11), Zeidler, J. R. [199] Performance analysis of LMS adaptive prediction filters, Proceedings of the IEEE 78(12) Biography S. E. El-Khamy received his PhD degree from the University of Massachusetts, USA, in He is a professor and past chairman of the Department of Electrical Engineering, Alexandria University, Egypt. His current research areas of interest include modern signal processing and their applications in image processing and communication systems. He has published about two hundred scientific papers and has earned many national and international research awards among which are the IEEE, R.W.P. King best paper award in 198 and the Egypt s State Appreciaton award (highest) in Engineering Sciences for 24. He is a Fellow of the IEEE since M. M. Hadhoud received the BSc and MSc degrees in electrical engineering from Menoufia University, Egypt, in 1976 and 1981 respectively. He received the PhD degree from Southampton University in He is currently a professor in the Department of Information Technology, Faculty of Computers and Information, Menoufia University. His areas of interests are signal processing, image processing and digital communications. M. I. Dessouky received the BSc and MSc degrees in electrical engineering from Menoufia University in Egypt in 1976 and 1981 respectively. He received the PhD degree from McMaster University in He is currently a professor in the Department of Electronics and Electrical Communications, Faculty of Electronic Engineering, Menoufia University. His areas of interests include signal processing, image processing and digital communications. B. M. Salam received by BSc degree in electrical engineering from Menoufia University and his MSc degree from Cairo University. He obtained the PhD degree from Drexel University, USA. He is currently in the Department of Electronics and Electrical Communications, Faculty of Electronic Engineering, Menoufia University. His areas of interests are signal processing, image processing and digital communications. F. E. Abd El-Samie received his BSc and MSc degrees in electrical engineering from Menoufia University, Egypt, in 1998 and 21 respectively. He has obtained the PhD degree in 25, and he is currently with the Department of Electronics and Electrical Communications, Faculty of Electronic Engineering, Menoufia University His areas of interests are signal processing, image enhancement, restoration, super resolution and interpolation.