Adaptive classification of digital amplitude-phase modulated signals with additive non-Gaussian noise

In this paper we analyze adaptive classification algorithm for digital amplitude-phase modulated signals in flat fading channel with non-Gaussian additive noise, representing possible interference in the channel besides thermal receiver noise. We represent additive noise via normal mixture model, which is able to model various interference scenarios by choosing specific mixture parameters. The adaptive classification algorithm based on application of recursive form of Expectation-Maximization (EM) algorithm for noise parameters estimation is proposed. It allows easy maximum-likelihood classification of the signal of interest using posterior probabilities computed in EM framework. This adaptive approach allows receiver classification algorithm to adapt to time-varying interference scenarios, since recursive form of EM algorithm updates mixture parameters estimates with every new sample obtained. In this paper we analyze the performance of our adaptive classification algorithm for a specific interference scenario expressed with particular interference mixture parameters. We compare its performance to classic optimal maximum-likelihood classification for Gaussian additive noise scenario, and also to algorithm with perfect knowledge of noise parameters.