A new method for multiple source detection and identification from array data using cumulants and its application to shock waves propagation

The problem of multiple component signal estimation is addressed in both frequency and time domains using higher order statistics. A multiple component signal is defined as a superposition of independent non-Gaussian linear processes. Two algorithms are proposed to estimate the transfer function characteristics of the individual component filters: the first approach is based on an eigen-decomposition of the trispectrum matrix whereas the second on an adaptive inverse filter estimation procedure. It is shown that both techniques have the capability to resolve more input signal components than the number of sensors.<>