Study of Methods for Decomposition of Superimposed Signals with Application to Electromyograms

The paper presents three deterministic and one statistical procedure for a possible decomposition of superimposed signals modelled as multichannel, linear, time-invariant system outputs. A brief theoretical background is given for short-time cepstral analysis, wavelet transform, iterative parametric search, and higher-order statistics approach. The methods are evaluated according to the results obtained on a synthetic electromyogram (EMG). Their performances are compared and enlightened from the standpoint of an application to the real EMG decomposition. 1 Introduction Many recent signal processing implementations deal with the problem of multichannel signal separation. Diverse approaches have been proposed in the fields of speech processing, data communication, sonar array processing, seismic exploration, and in the analysis of biological systems. All of them model superimposed signals with multichannel systems whose channels characterize the superimposed signal components. Consequen..