Near optimum detection of TCM signals in coloured noise

Abstract

Viterbi Algorithm is the optimum technique for detection of coded signals in white noise. However in many practical situations noise is coloured or correlated. The optimum detection techniques for both uncoded and coded (Trellis coded) signals in additive coloured Gaussian noise (ACGN) have been derived earlier in the literature. For detection of coded signals in coloured noise, the linear equalizer-predictive Viterbi algorithm (LE-PVA) is the optimum scheme. The whitening property of the prediction filter is the basis for the optimal performance of the LE-PVA. However the performance degradation in the practical LE-PVA is mainly due to the fact that the LE trained using the LMS algorithm may not converge to the global minimum and perfect estimates of the autocorrelation of the error signal at the T-spaced sampler output are not available. In this paper, we simulate a near ideal LE-PVA which consists of a filter matched to the received pulse, a T-spaced sampler and a near optimum T-spaced equalizer followed by PVA, whose predictor coefficients are computed from theory. Note that it is not possible to implement the ideal LE-PVA since it requires an infinite length T-spaced equalizer and an infinite length predictor.