Testing equalisation performance in blind adaptation

Blind adaptation is a means to identify a system or system inverse without explicit recourse to its input (apart from statistical information). Blind adaptation algorithms with memoryless cost functions are known to be susceptible to converging to local minima due to the multimodality of their cost functions, potentially leading to a poor performance in terms of mitigating the distortion incurred by the system input. We propose an off-line statistical test for detecting the errors in the equaliser output resulting from this poor equalisation performance. Although the proposed test has more general application in the non-adaptive context, for example, in the detection of error propagation in decision feedback equalisers, the development here will be based on the important blind equalisation problem for space reasons.