Artificial neural network and conventional approaches to filtering and pattern recognition

Two approaches to the design of artificial neural networks-the associative memory and multilayer perceptron-are discussed. In particular, the use of nonlinear processing is investigated and the application of these structures to pattern recognition and recall is examined in the context of the conventional matched-filter detection operation, which is widely used in communications systems. One specific application, that of channel equalisation, is examined for the multilayer perceptron, which is shown to offer a performance advantage over conventional linear filter-based equalisers