A new approach to robust, weighted signal averaging

In this paper, a new approach for robust, weighted averaging of time-aligned signals is proposed. Suppression of noise in such case can be achieved with the use of the averaging technique. The signals are time-aligned and then the average template is determined. To this end, the arithmetic mean operator is often applied to the synchronized signal samples or its various modifications. However, the disadvantage of the mean operator is its sensitivity to outliers. The weighted averaging operation can be regarded as special case of clustering. For that reason in this work the averaging process is formulated as the problem of certain criterion function minimization and a few different cost functions are employed. The maximum likelihood estimator of location based on the generalized Cauchy distribution is used as the cost function. Such approach allows to suppress various types of impulsive noise. The proposed methods performance is experimentally evaluated and compared to the reference methods using electrocardiographic signal in the presence of the impulsive noise and the real muscle noise as well as the case of noise power variations.