Performance Analysis of Adaptive Filtering Algorithms for Denoising of ECG signals Nasreen Sultana 1 , Yedukondalu Kamatham 2 and Bhavani Kinnara 3 Dept of ECE 1, 2 , Bhoj Reddy Engineering College, Vinay Nagar, Saidabad, Hyderabad - 500059, India. Dept of EIE 3 , Vignan Institute of Technology and Science, Deshmukhi, Hyderabad. India. E-mail: 1 nasreens368@gmail.com, 2 kyedukondalu@gmail.com and 3 bhavanikinnara@gmail.com Abstract: Electrocardiogram (ECG) can help to diagnose range of diseases including heart arrhythmias, heart enlargement, heart inflammation (pericarditis or myocarditis) and coronary heart disease. ECG consists of noise which is non stationary that affects the reliability of ECG waveform. In this paper an adaptive filter for denoising ECG signal based on Least Mean Squares (LMS), Normalized Least Mean Square (NLMS), Affine Projection LMS (APA-LMS) and Recursive least Squares algorithm (RLS) is presented with experimental results and the results are found to be encouraging. The performances of these algorithms are compared in terms of various parameters such as SNR, PSNR, MSE and SD. To validate the proposed methods, real time recorded data from the MIT-BIH database is used. RLS algorithm is found to exhibit lower MSE, and higher SNR compared to other algorithms. Therefore the results demonstrate superior performance of adaptive RLS filter for denoising of ECG signal. Index terms: ECG, LMS, NLMS, RLS, APA-LMS, MSE, Interference, SNR and denoising I. I.INTRODUCTION As an adaptive filter do not require any signal statistical characteristics, it is considered as a primary method to filter noise and interference in ECG signals [1]. In recent years a remarkable improvement has been shown by medical research in areas of tumor detection, hemophilia, venereal diseases etc. Cardiac syndrome is considered to be one of the prominent focused areas of medical research. This area remains an open challenge to avoid maximum threats occurring to human life despite of many achievements [2]. ECG is considered as the most effective tool developed for cardiac analysis which provides a detailed profile of the electrical impulses that causes the cardiac fibers to expand and contract. As ECG signal is very sensitive there might be a possibility of interference of different types of noises corrupting the ECG signal thereby changing the real amplitude and duration [3]. To effectively process these ECG signals for proper diagnosis, denoising of ECG signal is essential [4- 5]. ECG is used for the detection of sudden death syndromes and different heart diseases. ECG is a composite signal, where different heart activities result in generation of its components. Factors which affect in the improper diagnosis of ECG are the artifacts and noises introduced during its extraction [6]. One cycle of ECG signal corresponds to different phases of heart activities. Conventionally an ECG signal is labeled with peaks P, Q, R, S and T as in Fig.1. The P and T waves are low frequency components i.e. 5-9 Hz, while QRS is at high frequency of 10-40 Hz [7, 8]. The ECG signal enhancement results in presentation of an ECG signal which provides accurate and easy interpretation by separating invalid signal components from undesired artifacts. ECG signals are subjected to various types of noises, which lie in different frequency ranges. These include persistent noises, which reside in a variety of frequency bands such as low frequency range (Base Line Wander), mid frequency range (Power Line Interference) and high frequency range (Electromyography) signals and burst noises, which appear for very short duration (electro-surgical noise) [9]. Fig 1. Components of normal ECG waveform [10] ECG signals are easily corrupted by unwanted interference and noise. This is considered as one of the most serious problem resulting during the acquisition and recording of ECG signals. Many denoising algorithms are developed such as Discrete Cosine transform (DCT), SG smoothing, DWT, LMS, RLS. Many other techniques such as principle component analysis, independent component analysis and neural networks have also been used to extract noise free ECG signal from a noisy ECG which involves statistical analysis. These denoising algorithms should improve SNR to obtain clean and readily observable ECG recording, and preserve the original shape without introducing distortions in low amplitude ST-segment and P, T waves. Verulkar et. al., (2012) used various filters such as Adaptive Volterra filter, FIR filter, Nyquist filter and IIR Notch filter for PLI (Power Line Interference) reduction [11]. The results show that IIR Notch filter gives a noise free output. In [12] a comparison between LMS, NLMS, and RLS algorithm is done to mitigate the power line interference. The performance of the algorithms is analyzed based on MSE (Mean Square Error) and SNR (Signal to Noise Ratio). The results show that, in comparison with LMS and NLMS algorithm, RLS algorithm exhibits high speed of convergence as well as low steady state error. Ahmad et. al., (2013) described the comparison of adaptive filters based on LMS and RLS algorithm which are implemented for cancellation of motion artifact noise in ECG signals. Results show that adaptive filter using RLS algorithm gives best performance based on performance parameters MSE and SNR. The aim of this paper is to investigate and compare the performance of different adaptive filter algorithms in detail for ECG signal denoising. The methods used here for denoising of ECG signals are presented in the next section. 297 978-1-4799-8792-4/15/$31.00 c 2015 IEEE