The adaptation of complex layered non-linear structures using global search strategies

Communications channel equalisation is known to be a fundamentally non-linear problem. There is, however considerable confusion relating to the appropriate form of non-linear processor which should be applied in this context. Certainly, the most parsimonious architecture is a nested structure such as the multi-layer-perceptron (MLP). However, this poses problems in terms of reliability of adaptation when using traditional algorithms (gradient descent). This paper examines this problem from the viewpoint of using a global search strategy based on evolutionary algorithms.