DCT-BASED VARIABLE STEP SIZE GRIFFITHS' LMS ALGORITHM FOR RANDOM NOISE CANCELLATION IN ECGs

Abstract

This paper presents a new random noise cancellation technique for cancelling muscle artifact effects from ECG using ALE in the transformed domain. For this a transform domain variable step size griffith least mean square (TVGLMS) algorithm is proposed. The technique is based on the adaptation of the gradient of the error surface. The method frees both the step size and the gradient from observation noise and reduces the gradient mis-adjustment error. The sluggishness introduced due to the averaging of the gradient in the time domain is overcome by the transformed domain approach. The proposed algorithm uses a discrete cosine transform (DCT)-based signal decomposition due to its improved frequency resolution compared to a discrete Fourier transform (DFT). Furthermore, as the data used symmetrical, DCT usage results in low leakage (bias and variance). The performance of the proposed method has been tested on ECG signals combined with WGN, extracted from MIT database, and compared with several existing techniques like LMS, NLMS, and VGLMS.