Reduced-Rank Filtering-Based Semiblind MIMO-OFDM Sparse Channel Estimation

Abstract

This article derives and analyzes novel, semiblind data/channel estimation algorithms for block-based multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) transmission in sparse channels. A novel reduced equalizer (with few active taps) is introduced and the semiblind sparse channel/data estimation problem is solved via reduced-rank filtering methods. A novel arrangement of the “full” equalizer matrix allows data/channel estimation, with fewer received OFDM blocks (in channels with short coherence time). The “reduced” equalizer then further reduces the number of OFDM blocks for estimation. The novel, reduced-rank filters are analyzed and shown to be equivalent to some nonblind Bayesian and “natural gradient” methods, widely used for sparse estimation. Thus, the novel method can be considered a blind Bayesian method, which is not available in the literature, and whose perturbation analysis is performed. Simulation results illustrate that the novel semiblind estimators perform much better than existing blind/training-based sparse methods (even including the popular compressed sensing and Bayesian methods), when few training subcarriers are available (which may occur in futuristic, pilot-starved massive MIMO-OFDM systems), at much reduced complexity.