Abstract
The constant modulus algorithm (CMA) applied to a fractionally spaced equaliser, with length and zero constraints on the channel, has been shown to be globally convergent when there is no channel noise. However, in the presence of channel noise, the performance of the CMA can suffer because of the existence of undesired local minima corresponding to different delays. A new technique is introduced, based on the Gram-Schmidt orthogonalisation, to obtain convergence to an optimum delay where the mean square error is minimum.