A neural network trained with the extended Kalman algorithm used for the equalization of a binary communication channel

This paper describes a feedforward neural network architecture trained with the extended Kalman filter algorithm instead of the standard (LMS) method. It presents a simplified recursive procedure for calculating the necessary derivatives. The resulting algorithm is then used to train a network to adapt to the decision boundary of an optimal receiver for a binary communication channel, resulting in increased convergence speed and better approximation properties