Decision Weighted Adaptive Algorithms with Applications to Wireless Channel Estimation

This thesis proposes and studies novel modifications to the least mean squares (LMS) and weighted recursive least squares (WRLS or weighted RLS) adaptive algorithms to estimate the impulse response of a wireless communications channel blindly without the aid of a training or probe sequence. Specifically, we use knowledge of receiver decision quality to weight the LMS and WRLS estimators to increase their robustness to hard decision errors and channel noise. We propose two classes of these decision weighted algorithms: 1) soft decision weighted, where algorithm weights are a function of receiver soft decisions; and 2) ideal decision weighted, where algorithm weights are a function of decision error knowledge. We compared the performance of these decision weighted estimators to their non-weighted blind and non-blind counterparts through simulations over free-space propagation three path (Rummler) and mobile radio channel models. Our results show that the decision weighted LMS (DWLMS) has significant advantages over ordinary LMS in environments with low signal-to-noise ratios (SNR) and high symbol error rates (SER). The decision weighted recursive least squares, however, had mixed results. The soft decision weighted RLS (SDWRLS) had poorer performance than other WRLS algorithms, but the ideal decision weighted RLS (IDWRLS) had similar performance to other WRLS. To improve the performance of the SDWRLS, we also propose and simulate modifications to its estimator update structure. These modifications allow the SDWRLS to perform similar to the soft decision weighted LMS.