Weighted averaging with adaptive weight estimation [ECG]

A method to estimate the weights in weighted averaging is presented. The authors first recall what is the optimum weight vector associated to the maximum signal-to-noise ratio (SNR) without bias. Noise power variation is the topic of interest. Significant variations of muscle tone may occur during the acquisition process. In this case, weighted averaging is used instead of classical averaging. It is shown that the optimum weights can be adaptively estimated using a least mean square (LMS) algorithm with a constraint. The constraint ensures that the estimated signal amplitude is not altered by averaging. The theoretical results are checked by simulation and on real electrocardiograms. It is shown that, after convergence of the algorithm, the SNR of the estimated signal is always equal or superior to the SNR of the estimated signal obtained by classical averaging