Blind equalization and identification of time-varying communication channels

In digital communications, data signals are transmitted through linearly distortive analog channels. Two major sources of linear channel distortion are multipath propagation and limited bandwidth. Linear channel distortion leads to intersymbol interference (ISI) at the receiver which, if not compensated, may result in high error rates in symbol detection. The device to battle the effect of ISI is called an equalizer. Traditionally, an equalizer is designed with the aid of a training sequence. Equalization based on initial adjustment of the coefficients without the benefit of a training sequence is said to be self-recovering or blind equalization. Time-varying channels arise when ISI is induced by multipath effects from a changing environment. They can cause severe problems to traditional equalizers that were designed to equalize time-invariant or “slowly” changing channels.; In this dissertation, we first concentrate on blind equalization and identification of time-varying single-input multiple-output (SIMO) channels. Such channels arise when antenna arrays are used or when signals are oversampled or when both scenarios are applicable. Only second-order statistics of the observed data are exploited. The time-varying channel is represented by a complex exponential basis expansion model (CE-BEM). Two methods, single-step linear prediction error (SLPE) and multi-step linear prediction error (MLPE), are proposed. Both of them deliver better performance than existing subspace methods, since they are more robust to channel order over-estimation. MLPE is superior to SLPE when the leading channel coefficients are “small”. Sufficient conditions for channel identifiability are investigated and a regularized zero-forcing equalizer is implemented. In addition, a brute-force frequency selection algorithm is proposed to estimate the set of active basis frequencies.; We then consider multiple-input multiple-out (MIMO) channels, which arise when a multi-user system is considered. The MLPE method is extended for blind channel estimation in time-varying short-code DS-CDMA systems. In addition to temporal diversity, spatial diversity is also exploited by using multiple receiving antennas.; Finally, we propose a superimposed training-based method to estimate the channel coefficients. Unlike existing similar methods, our method allows mean-value uncertainty at the receiver and more general training sequences. Only first-order statistics of the data is used. A performance analysis is also provided for the method.