Classification of Digital Amplitude-Phase Modulated Signals in Time-Correlated Non-Gaussian Channels

In this paper, a new algorithm is proposed for the classification of digital amplitude-phase modulated signals in flat fading channels with time-correlated non-Gaussian noise. The first-order statistics of the additive noise is modeled by a Gaussian mixture distribution and an autoregressive (AR) process is used to model the time-correlation. The proposed classifier involves the use of a whitening filter, necessary to reduce the complexity of the classification process, and maximum-likelihood classification. For the estimation of the whitening filter coefficients, a new blind technique that is based on the use of a robust H_∞ filter is developed. After whitening the received signal, following a composite hypothesis testing approach, the unknown fading and noise distribution parameters are estimated. Results are presented which show that when the noise process is time-correlated non-Gaussian, the proposed classifier outperforms maximum-likelihood classifiers developed under the assumption that the noise process is either white non-Gaussian or white Gaussian. It is also shown that when the noise process is white Gaussian, the proposed classifier´s performance closely approaches that of the maximum-likelihood classifier developed for white Gaussian noise channels.