An Efficient Kalman Noise Canceller for Cardiac Signal Analysis in Modern Telecardiology Systems

The monitoring of electrocardiography (ECG) in ambulatory conditions is an important task for achieving success in remote healthcare monitoring. In this paper, Kalman-based adaptive artifact cancellation structures, which are the hybrid versions of least-mean-square (LMS) algorithm variants, are proposed for the high-resolution enhancement of an ECG signal. The main advantage of the Kalman-based adaptive filter structure lies in the extraction of the ECG signal at a low signal-to-noise ratio (SNR). This property helps the Kalman noise canceller (KNC) to achieve greater monitoring accuracy. The hybrid version of this Kalman algorithm makes the noise canceller independent of the step-size parameter, whereas the performance of conventional adaptive filters depends on the step-size parameter. In the proposed KNCs, we use discrete wavelet transform to generate a reference component from the contaminated ECG signal itself. In addition to these constraints, in remote health care monitoring, it is necessary to lower the computational burden and increase the convergence rate of the noise canceller. In a practical remote health care monitoring system if the computational burden of the signal conditioning unit is more, then it takes a much greater amount of time to process samples in the filter. This leads to waiting of incoming samples at the input port of the filter. This causes overlapping of samples at the input port and causes ambiguity in the diagnosis process. To achieve the feature of low computational complexity, we combine Kalman-based LMS (KLMS) with sign algorithms. In addition, data normalization is introduced to improve convergence characteristics. Finally, to test the performance of the proposed implementations, real ECG signals from the MIT-BIH database is used. The measured parameters, namely, SNR, excess mean square error, and mis-adjustment are calculated in the enhancement process to judge the ability of various algorithms. Experimental results confirm that the proposed Kalman-based adaptive algorithms are better than the LMS-based algorithms. Among the implemented techniques sign regressor-based KNC performs better in terms of various considered measures.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.