Adaptive MLSDE Receivers for Wireless Communications

grantor: University of TorontoThis thesis presents the structural design, performance evaluation and complexity reduction of adaptive maximum likelihood sequence detection and estimation (MLSDE) receivers for wireless communications. The receiver structure is developed based on the maximum likelihood (ML) criterion for joint channel estimation and data detection and using the expectation and maximization (EM) algorithm. Generalized MLSDE (GMLSDE) algorithm, which alternates between estimation and detection and at the same time increases the likelihood iteratively, is developed using the on-line EM algorithm. The GMLSDE provides the theoretical ML base from which per-survivor processing (using different estimators for each survivor path) and conventional channel estimation (using only one estimator for all survivor paths) can be deduced for different channel models. Numerous adaptive MLSDE receiver structures are developed for different channel models from the GMLSDE algorithm using causal estimation and detection methods which only guarantee increasing the likelihood. Although some structures are new, the method of deriving all the structures shows the power of the GMLSDE algorithm in unifying different ML-based joint channel estimation and data detection receivers. Data detection part in the adaptive MLSDE receiver is implemented by the Viterbi algorithm while its estimation part is based on a modified Titterington's stochastic approximation method. The estimation part leads to a novel RLS/Kalman-type algorithm when the channel impulse response (CIR) is a deterministic process embedded in an additive colored Gaussian noise. The bit error rate performances of the adaptive MLSDE receiver for a differentially encoded quadrature phase shift keying signal transmitted over the frequency flat and selective Rayleigh fading channels are evaluated by simulations for different fading rates when the CIR is modeled at the receiver by deterministic and Gaussian random vectors/processes. Comparison between the performances shows that in many situations using the deterministic CIR model is sufficient. The thesis proposes the adaptive state allocation (ASA) algorithm which uses adaptive threshold and adaptive state partitioning methods to reduce the receiver complexity at high and low signal-to-noise ratios respectively. In the adaptive threshold method, a threshold value is formulated to choose the more likely correct states at each time and in the adaptive state partitioning method, branch metrics of the trellis diagram are fused/diffused adaptively. Results for multipath Rayleigh fading channels show that the ASA algorithm greatly reduces the receiver complexity with negligible performance degradation.Ph.D