A transversal equalizer with an increased adaptation speed and tracking capability

A modification to the well-known least-mean-square (LMS) algorithm for adaptive equalizers is presented. It allows a flexible tradeoff between throughput rate and adaptation speed by adjusting the arithmetic expense per output sample. If the arithmetic expense is increased to that of recursive least-squares (RLS)-based algorithms, a comparable adaptation speed is achieved. A full-custom implementation of a transversal equalizer using timesharing by a factor of two and Booth-coded coefficients proves the feasibility and efficiency of the modified LMS. It achieves data rates of up to 75 MBaud in a 0.8-μm CMOS technology under worst case conditions. The convergence rate of the gradient lattice algorithm can be achieved by reducing the symbol rate to 15 MBaud. The modification requires about 15% additional area in the basic cell of the filter, a few additional control signals, and a buffer to store 40 input symbols