Use of the EM algorithm in impulsive noise channels.

An adaptive detector which uses the EM algorithm ('E' for expectation, 'M' for maximize likelihood) in non-Gaussian noise environments is presented. It is shown that the adaptive detector is optimal for small signals and that it degrades smoothly toward the conventional linear detector as the signal-to-noise ratio increases. Performance of the EM algorithm and a CPFSK adaptive detector for Class A impulsive noise are reported.