A computationally efficient MLSD algorithm using fractionally-spaced linear prediction

We address maximum likelihood sequence detection (MLSD) for communications on flat Rayleigh fading channels. Making use of linear prediction we derive the structure of an oversampled detector which is computationally efficient and can easily accommodate a time varying Doppler frequency. This goal is achieved through a specific analytical derivation of the prediction coefficients which appear in the metric of the sequence detector. For fractionally-spaced observations, we derive the constraints which allows one to establish an explicit relationship between the set of observation predictors and the set of fading predictors. Then, the equivalence between different classes of prediction-based detectors is shown. The numerical results give useful indications to balance the complexity of these detectors in terms of prediction order, description of intersymbol interference and oversampling.