Systolic array architecture for LMS algorithm using Hopfield model network

Abstract

Presents a systolic array implementation of a modified LMS (least mean square) algorithm, which is based on the dynamics of the network in the Hopfield model network. The rate of the adaptation of the modified algorithm is n times as fast as the conventional LMS algorithm with the same control gain, where n is the number of iterations for each piece of sampled data in the network. However, the computational complexity of the algorithm increased. In the modified algorithm, the coefficients can be computed independently. Therefore, parallel array processing such as a systolic array is available. The systolic array consists of one kind of processing element, and the processing element consists on one multiplier, one adder, and one memory. The number of the processing element is the same as the order of the adaptive filter. The computation time for updating the coefficients of the adaptive filter is (L+1)n in time steps, where L is the number of coefficients of the adaptive filter and n is the number of iterations in the network.<>