Recursive equlaizers using a multistage adaptive algorithm

The performance of adaptive recursive equalizers based on steepest descent with zero starting condition is often limited by local minima on the mean square error surface. This paper describes an algorithm which attempts to avoid shallow minima for adaptive recursive equalizers. The algorithm allows the values of tap coefficients as well as the number of numerator and denominator taps of a recursive equalizer to adapt in stages. Simulation results showed that for several representative channels, this algorithm outperforms the commonly used adaptive transversal equalizers and the zero-start recursive equalizers. A major drawback of this algorithm is that a long training sequence is necessary.