Recursive Bayesian techniques for blind equalization

Abstract

The paper introduces extended Bayesian filters (EBFs), a new family of blind deconvolution filters for digital communications. The EBF performs nonlinear estimation of the channel and data simultaneously, and achieves suboptimal symbol-by-symbol demodulation in unknown channels. The complexity of the EBF is exponential in a parameter that is typically chosen to be less than the channel length and the filter lag. This key characteristic makes the EBF practical for both long channels and large constellations. Simulations characterizing the performance of EBFs in severe intersymbol interference are performed and demonstrate the fast convergence and robust equalization of the EBFs for linearly modulated signals on unknown channels. Also, a principled adaptive complexity reduction algorithm called the reduced-state EBF (RSEBF) is developed and applied to 16-QAM signals.<>