Recursive filtering approaches to channel equalization and estimation in communications

Abstract

Communication poses challenges in the presence of interference channel environments. In order to attain performance gains, signal processing techniques at the receiver need to detect the most likely transmitted signal based on the knowledge of the received signal, the channel state information and the statistics of noise. This thesis develops such practical schemes based on a Kalman filter framework, assuming that the channel knowledge is unknown to the receiver. The communication contexts addressed in this thesis include the equalization of intersymbol interference (ISI) channels in single-transmit single-receive (STSR) systems and the channel estimation combined with the decoding of multiple-input multiple-output (MIMO) systems in fading channels. Signal processing based on a Bayesian filtering framework, built on a state-space model of the given communication system, plays a critical role in the equalization and the estimation of the channel. We develop adaptive channel equalization techniques utilizing recurrent neural networks and Kalman filters for ISI cancellation in STSR systems, which include nonlinear distortions, additive white Gaussian noise and additive white non-Gaussian impulsive noise. In uncoded systems, the extended Kalman filter (EKF) and the unscented Kalman filter (UKF) are used to train the recurrent neural equalizers and achieve the performance improvement in terms of bit error rate and convergence rate, with a shorter training sequence over the conventional equalization schemes. The integration of semiblind channel estimation into a turbo receiver, the so-called turbo-BLAST1, is developed for narrowband MIMO wireless channels. For quasi-static MIMO channels, we present iterative channel estimation schemes based on adaptive filtering algorithms such as least-mean square, recursive least-squares and the Kalman filter. The iterative strategy with adaptive filtering leads to a computationally efficient solution to iterative channel estimation, compared to the conventional snapshot approaches. For time-varying MIMO channels, we present the use of particle filtering in order to track the time variations of the channels. The improved performance by the particle filtering channel tracking is demonstrated for both Gaussian and non-Gaussian noise environments. 1BLAST: Bell-labs LAyered Space-Time architectures.