Array processing in non-Gaussian noise with the EM algorithm

Richard J. Kozick, Brian M. Sadler, Rick S. Blum

Abstract: A central problem in sensor array processing is the localization of multiple sources and the reception of the signals emitted by those sources. Many approaches have been studied for this problem when the additive noise in the sensor array data is modeled with a Gaussian distribution. However, the schemes designed for Gaussian noise typically perform very poorly when the noise is non-Gaussian. An algorithm is presented for array processing in non-Gaussian noise. The algorithm is based on modeling the noise with a Gaussian mixture distribution. The expectation-maximization (EM) algorithm is then used to derive an iterative processing structure that estimates the source locations, estimates the source waveforms, and adapts the processing to match the characteristics of the noise. Simulation examples are presented to illustrate the performance of the algorithm.