Performance Analysis of Robust Recursive Least Square and its Lattice Structure for Noise Cancellation in ECG Signals

Abstract

Bio signals are non-stationary signals corrupted by various environmental noises like Power Line Interference (PLI), Baseline Drift, Instrumentation noise generated by electronic devices, Motion artifacts, Electrosurgical Noise, Electrode Motion (EM) noise. Thus removal of the environmental noises becomes a challenging task. The filters with fixed coefficients cannot be applied because the human behavior is not exactly known at any point of time. An adaptive filter has flexible ability to modify the system parameters and adjusts the input signal to bring out optimal performance. This paper introduces the application of Robust Recursive Least Algorithm (RRLS) for adaptive noise cancellation. The Robust RLS algorithm is used to find the optimal value of the regularization parameter (delta) and forgetting factor (lambda). By maintaining a variable forgetting factor the convergence of the algorithm is faster. Architectural modifications such as direct form and transpose form structures are also carried out in filter structure to analyze the delay. To reduce the computational complexity of the RRLS algorithms Lattice RLS Structures are also studied for its performance. Accordingly 4 tap, 8 tap and 16 tap RRLS filters in both direct form and transpose forms are designed and implemented in Virtex5 FPGA kit. Also 4 tap, 8 tap and 16 tap lattice filters and its pipelined version are designed and implemented. The designed filters are tested using various ECG signals taken from MIT BIH database. Result shows an improvement of 10.88% of SNR for PLI Noise and 12.3% of SNR for EM Noise compared to RRLS Filter. From the overall analysis, the number of LUT‘S required for Lattice RLS has been reduced by 12.1% for 8-Tap compared to Robust RLS filter. It is also observed that the combinational path delay of pipelined lattice RLS is reduced by 80.7% for 8-Tap compared to Robust RLS filter.