Вы находитесь на странице: 1из 5

IS J AA

International Journal of Systems , Algorithms & Applications

Performance Evaluation of Adaptive Filter Algorithms for EOG Signal Processing


Naga Rajesh A, 2Chandralingam S, 3Anjaneyulu T, 4 Satyanarayana K Department of Physics, Jawaharlal Nehru Technological University, Hyderabad-500085, 2 JNTUH College of Engineering, Manthani, Godavarikhani, Karimnagar(Dist.), India 3 Retd. Professor, Department of Electrical Engineering, IIT Bombay, India, 4 Department of Biomedical Engineering, Osmania University, Hyderabad-500007, e-mail: nagarajesh9@yahoo.co.in1, Chandra_lingam@yahoo.com2, tangiralaanjaneyulu@yahoo.co.in3, kjwala@yahoo.com4
1

Abstract Adaptive filter technique has become one of the effective and popular approaches for the processing and analysis of biomedical signals. This paper compares the performance of various adaptive filter algorithms to identify the right candidate for application in EOG prediction which was done by varying the filter order and past sample size. The results obtained are remarkable and re-affirms the RLS filter as the solution for its very low RMS prediction error in comparison to LMS, NLMS and DLMS filter algorithms. Subsequently, performance evaluation was also carried out between these algorithms based on various step sizes which again showed that RLS filter as the solution for its low RMS prediction error and good signal to noise ratio (SNR). The results suggest the potential utility of RLS algorithm in the EOG prediction, which can also be used in biomedical engineering for detection and diagnosis of various diseases. Figure 1. A sample EOG (Right) signal

I. INTRODUCTION The Electrooculogram (EOG) is the electrical signal produced by the potential difference between the retina and the cornea of the eye. This is due to the large presence of electrically active nerves in the retina compared to the front of the eye. The corneal part is a positive pole and the retina part is a negative pole in the eyeball. Eye movement will respectively generates voltage up to 16uV and 14uV per 1 in horizontal and vertical way [1]. Prediction of EOG from past samples has significance to detect possible abnormalities remedial measure for assessing the function of the pigment epithelium. However, the EOG signal is interfered by signals of electroencephalography, electromyography, electrical network, speech, blink, etc and it is important the interference of noise is dimmed and the user is calm and relaxed [2]. Therefore much attention has been placed recently on filtering methods to overcome the noises. Adaptive filtering has become one of the effective and popular methods for the processing and analysis of the biomedical signals as it could track the actual frequency of the noise and potentially eliminates the interferences [3]. In this paper, we present different adaptive filter algorithms to predict EOG signal from past samples and identify the best algorithm based on the performance of the filters and therefore, we herein present our results of the investigative study. For the simulation, the original EOG (Right) sample shown in Fig. 1 was obtained from MIT-BIH polysomnographic database.

II. ADAPTIVE FILTER ALGORITHMS A system is said to be adaptive when it tries to adjust its parameters with the aid of meeting some target that depends upon the state of the system and its surroundings. The most common signal processing of this type of bioelectric signal separates the deterministic signal from the noise. Adaptive filters are selfdesigning filters based on an algorithm which allows the filter to learn the initial input statistics and to track them if they are time varying. These filters estimate the deterministic signal and remove the noise uncorrelated with the deterministic signal. The principle of adaptive filter is as shown in Fig. 2 [4].

Figure 2. Principle of Adaptive Filter

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

14

PERFORMANCE EVALUATION OF ADAPTIVE FILTER ALGORITHMS FOR EOG SIGNAL PROCESSING

IS J AA

International Journal of Systems , Algorithms & Applications

III. LMS ADAPTIVE FILTER ALGORITHM The LMS algorithm is a method to estimate gradient vector with instantaneous value. The algorithm changes the filter tap weights so that e (n) is minimized in the mean-square sense. The conventional LMS algorithm is a stochastic implementation of the steepest descent algorithm. It simply replaces the cost function (n) = E [e2 (n)] by its instantaneous coarse estimate. The error estimation e(n) is e (n) = d(n) w(n) X(n) Coefficient updating equation is w (n+1) = w(n) + x(n) e(n)
Figure 5. The Percentage RMS error and SNR plots for various step sizes of LMS adaptive filter algorithm

(1)

(2)

Where is an appropriate step size to be chosen as 0 < < 0.2 for the convergence of the algorithm. The larger step sizes make the coefficients to fluctuate wildly and eventually become unstable [4]. In this experiment the EOG signal was sampled and submitted the discrete samples as x0, x1..xM-1 to the input of an LMS algorithm. The predicted (estimated) signal for a filter of order 30 was recorded and presented in Fig. 3. It is evident from Fig. 4 that the average error is around -0.0235 units. In the above computer simulation, the (step size) was taken as 0.002. Subsequently, the percentage RMS error and Signal to Noise Ratio is calculated for various step sizes where in, the filter order and number of samples were 5 and 600 respectively (Fig. 5).

IV. NLMS ADAPTIVE FILTER ALGORITHM The Normalized Least Mean Square (NLMS) filter has similar structure with the LMS filter with a modified weight adaptation rule, given by W (k +1) = W (k) + . error. Xk / (mean-square (Xk) + offset) (3) The offset is kept to ensure no division by zero. We used an offset settings of 50, and = 0.09 for the computer simulation [5]. The original EOG and the predicted EOG by NLMS algorithm is given in Fig. 6 whereas, Fig. 7 shows the error in each sampling constant. Here, the average value of error was around -0.0257 units. In the above computer simulation, we took the step size, =0.09. The percentage RMS error and Signal to Noise Ratio was recorded and depicted in Fig. 8 for various values of step sizes where in, the filter order was 5 and the number of samples was 600.

Figure 3. EOG prediction by LMS adaptive filter algorithm Figure 6. EOG prediction by NLMS adaptive filter algorithm

Figure 4. The error in EOG prediction by LMS adaptive filter algorithm

Figure 7. The error in EOG prediction by NLMS adaptive filter algorithm

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

15

PERFORMANCE EVALUATION OF ADAPTIVE FILTER ALGORITHMS FOR EOG SIGNAL PROCESSING

IS J AA

International Journal of Systems , Algorithms & Applications

Figure 8. The Percentage RMS error and SNR plots for various step sizes of NLMS adaptive filter algorithm

Figure 10. The error in EOG prediction by DLMS adaptive filter algorithm

V. DLMS ADAPTIVE FILTER ALGORITHM The DLMS has D (0DN) units of Delay in the error feedback path. If the filters impulse response is denoted by wk(n), we can express the output signal y (n) at the time n as

(4) Where x (n) and N are respectively, the input signal and the filter length. The DLMS is described by W (n+1) =w (n) +e (n-1) x (n-D) e (n-D) =d(n-D)-WT(n-1)x(n-D)

(5) (6)

Figure 11. The Percentage RMS error and SNR plots for various step sizes of DLMS adaptive filter algorithm

Where e(n-D) is the error signal of the DLMS and d (n) is the desired signal;W(n) and x(n) are, respectively, the filter coefficient vector and the input vector, and they are given by W (n) = [w0 (n), w1 (n)..wN-1(n)]T (7) x (n) =[x(n), x(n-1).x(n-N+1)]T (8) Where [x]T indicates the transpose of a vector x [6]. The estimated signal for a filter of order 30 (i.e. M=30) was given in Fig. 9. The results obtained clearly demonstrate that the average error is around -0.0057 units from the error plot (Fig. 10). In the above computer simulation, the step size was taken as 0.008. The percentage RMS error and Signal to Noise Ratio was plotted in Fig. 11 for various values of step sizes where in, the filter order and the number of samples were respectively 5 and 600.

VI. RLS ADAPTIVE FILTER ALGORITHM The Recursive Least Square (RLS) filter employs the following algorithm for prediction. Here X(n) = [x(n-1) for a M order adaptive filter x (n-2) x (n-p) ] while |error| not less than a predefined > 0 do begin evaluate i. error e(n)= d(n) WTn X(n) ii. g(n) = P(n-1)X*(n){ + XT(n)P(n-1)X*(n)}-1 iii. P(n) = -1 P(n-1) g(n) XT(n) -1 P(n-1) iv. Wn = Wn-1 + e(n) g(n) end While; Here P(n) is the inverse of the weighted auto correlation matrix of Xk weighted by the forgetting factor . i.e. P (n) = Rx 1(n) where Rx (n) = (n-i) X*(i) XT(i) d (n) = desired signal, e(n)= error, and Wn = weight vector all at time sample t = n. With W0 = 0, P (0) = -1I, where I is a (M+1) (( M +1) identity matrix, and = forgetting factor = 0.99 [5]. The predicted signal for a filter of order 30 (i.e. M=30)

Figure 9. EOG prediction by DLMS adaptive filter algorithm

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

16

PERFORMANCE EVALUATION OF ADAPTIVE FILTER ALGORITHMS FOR EOG SIGNAL PROCESSING

IS J AA

International Journal of Systems , Algorithms & Applications

was given in Fig. 12. The error plot depicted in Fig. 13 demonstrates the average error is around 0.0003 units. The forgetting factor () was taken as 0.99 in the above computer simulation. The percentage RMS error and Signal to Noise Ratio was presented in Fig. 14 for various values of step sizes where in, the filter order was 5 and the number of samples was 600.

% RMS error in Prediction

No. 1

Algorithm LMS

M=30, S=500 13.095

M=60, S=1000 7.563

M=90, S=1500 6.0255

M=120, S=2000 5.2967

M=150, S=2500 4.7079

NLMS

14.2184

8.318

6.5992

5.8273

5.2611

3 4

DLMS RLS

6.817 0.5109

4.6746 0.3664

3.8825 0.2944

3.3547 0.24

2.8661 0.21

Figure 12. EOG prediction by RLS adaptive filter algorithm

VII. RESULTS The performance of various adaptive filter algorithms to identify the right candidate for application in EOG prediction was done by varying the filter order and past sample size. It is evident from Table-1 that the DLMS filter performs better followed by NLMS and LMS adaptive filter algorithms. It gives a clear picture that as the filter order and sample size are varied to calculate the percentage RMS error, the RLS algorithm performs better than the other three algorithms as shown in Fig. 14. The computer simulation envisages that RLS filter yields less % RMS error in comparison to LMS, NLMS and DLMS filters irrespective of the filter order and sample size. Experiments were carried out to determine the SNR along with % RMS error calculation for those algorithms outputs. The filter order (M) and sample size(S) were kept constant at 5 and 600 respectively. The step size was varied from 0.001 to 0.9 for LMS, NLMS and DLMS algorithms where as forgetting factor for RLS algorithm was also varied within the same range as shown in Fig. 14. This clearly shows that the RLS algorithm yields very less % RMS error than the other three algorithms and also maintains approximately constant value irrespective of forgetting factor ranging between 0.001 and 0.9. The results obtained shows a clear picture on SNR of RLS algorithm output where it maintains high value irrespective of forgetting factor proving its better performance over LMS, NLMS and DLMS algorithms. VIII. CONCLUSIONS The work presented in this study is to analyze the performance characteristics of four adaptive filter algorithms and found the RLS algorithm as better for EOG prediction wherein the ophthalmological diagnosis or assessing the function of pigment epithelium could be done with high accuracy. Further work is underway to optimize these algorithms in the implementation of

Figure 13. The error in EOG prediction by RLS adaptive filter algorithm

Figure 14. The Percentage RMS error and SNR plots for various forgetting factors of RLS adaptive filter algorithm

TABLE I. Performance Comparison of all the 4 adaptive filter algorithms for different values of filter order and number of samples

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

17

PERFORMANCE EVALUATION OF ADAPTIVE FILTER ALGORITHMS FOR EOG SIGNAL PROCESSING

IS J AA

International Journal of Systems , Algorithms & Applications

DSPs that estimates the EOG signal for detection and diagnosis of various diseases. ACKNOWLEDGMENTS One of the authors (Naga Rajesh A) personally expresses his thanks to Dr. Ravi Paturi, Dr. Ravi Kiran, Ms. Saroja Ranganath and Ms. Savitha Ramesh for the support provided in the completion of this work. REFERENCES Zhao Lv1,3 , Xiaopei Wu1 , Mi Li2,3 and Chao Zhang. 2008. Implementation of the EOG based Human Computer Interface System. Bioinformatics and Biomedical Engineering, ICBBE, 2008, 2188-2191. [2] Manuel Merino, Octavio Rivera, Isabel Gmez, Alberto Molina, Enrique Dorronzoro. A Method of EOG Signal Processing to Detect the Direction of Eye Movements.
[1]

[3]

http://matrix.dte.us.es/grupotais/images/articulos/eogdir.pd Sachin singh, K.L. Yadav. 2010. Performance Evaluation of Different Adaptive Filters for ECG Signal Processing. International Journal on Computer Science and Engineering Vol. 02, No. 05, 1880-1883. [4] A.Bhavani Sankar, D.Kumar and K.Seethalakshmi. 2010. Performance Study of Various Adaptive Filter Algorithms for noise cancellation in Respiratory signals. Signal Processing : An International Journal (SPIJ), Vol. 4, Issue. 5, 267-278 [5] Ayan Banerjee, Kanad Basu and Aruna Chakraborty. 2007. Prediction of EEG Signal by Digital Filtering. http://www.cise.ufl.edu/~kbasu/IISNPAPER2.pd. [6] Tadaaki KIMIJIMA, Kiyoshi NISHIKAWA and Hitoshi KIYA. 1999. An Effective Architecture of the pipelined LMS adaptive filters. IEICE Trans. FUNDAMENTALS, VOL.E82-A, No. 8. 1428-1434

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

18

Вам также может понравиться