A semi-blind approach to structured channel equalization

This paper describes a direct equalization approach for channels with some underlying structure. A semi-blind approach is taken here where a small amount of training symbols is available. A family of MMSE equalizers is obtained that includes some prior information about the channel structure. The channel structure assumed in this paper is that the channel vector lies approximately in the subspace of a matrix associated with the samples of the transmit pulse shape. Blind identifiability issues of the structured equalizer are also addressed. Numerical results using experimental indoor channel data indicate that these structured equalisers can achieve bit error rates that are significantly lower than traditional non-blind MMSE equalizers