ON BLIND ADAPTIVE ALGORITHMS FOR IIR EQUALIZERS

Previous work has shown that the optimum MSE causal equalizer is IIR, with a number of poles equal to the channel order. We investigate two different blind criteria for the adaptation of the recursive part of the equalizer: Output Variance Minimization (OVM) and a Pseudolinear Regression (PLR) method. In sufficient order cases both algorithms converge to the desired setting. In undermodeled cases (i.e. the number of poles in the equalizer is less than the channel order) these algorithms do not necessarily converge to a MSE minimum, but they generally provide acceptable performance. It is shown that under mild conditions PLR always admits a stationary point.