Bidirectional Decision Feedback Equalization and Mimo Channel Training

A major obstacle in reliable digital communication is inter-symbol interference (ISI), which is encountered in transmission over frequency-selective channels. A decision feedback equalizer (DFE) offers an effective and a low-complexity solution to combat ISI. However, the DFE is suboptimal and has a performance gap from the matched filter bound. In addition, the DFE suffers from error propagation caused by the feedback of incorrect decisions. The increase in popularity of packet based transmission systems like GSM or EDGE offers the possibility of block processing of the received signal. With block processing comes the freedom to process the signal in either a causal or a non-causal fashion. A novel bidirectional decision feedback equalizer (BiDFE) architecture that employs time-reversal of the received block of data is proposed in this dissertation. The BiDFE consists of two parallel DFE structures, one to equalize the received signal and the other to equalize the time-reversed version of the received signal. The BiDFE architecture is shown to provide a significant performance improvement over a conventional DFE with little additional complexity. To gain insight into the performance limitations of the BiDFE, the asymptotic (as the noise variance approaches zero) mean-squared error (MSE) performance of an infinite length design is evaluated. In an attempt to further improve performance, the filter coefficients of the BiDFE are optimized to minimize the overall MSE. However, when the ideal feedback assumption is relaxed, the symbol-error-rate (SER) performance does not show an improvement. To overcome this problem, two approaches that offer an additional improvement in SER performance, albeit marginal, are proposed. The BiDFE architecture is extended to the multiple-input multiple-output (MIMO) channel equalization. The design of the BiDFE assumes knowledge of the channel impulse response, which is typically estimated at the receiver. In training based MIMO channel estimation, the choice of training sequence affects performance. The criteria of optimality of MIMO training sequences is derived and design trade-offs in the choice of training length are discussed. Jaiganesh Balakrishnan was born in the city of Madras (now renamed Chennai), India in the year 1976. The first 17 years of his life were spent in Chromepet, a suburb of Madras. equalization, detection and estimation, adaptive signal processing and wireless communications. He is an avid birder and spends most of his spare time watching birds. In the Summer of 2001, he spent a few weeks in an island, off the coast of Maine, working as a research intern …