Constructive neural network design for the solution of two-state classification problems with application to channel equalization

Describes a deterministic algorithm for designing a MLP for the solution of a two-state classification problem. The Slab Algorithm was motivated by the problem of reconstructing digital signals which have been passed through a real linear dispersive channel of finite impulse response and corrupted with additive noise. The authors´ algorithm is designed to separate two finite disjoint sets of points by constructing a MLP with one hidden layer and a single output node. In the linearly separable case, no hidden layer is constructed. The parameters of the network are identified by standard linear programming techniques. The performance of the channel equalizer constructed by the Slab Algorithm is compared with that of a Bayesian optimal equalizer