Adaptive combination of LMS and logistic-linear equalizers to improve the speed-performance compromise

The Least Mean Squares algorithm (LMS) has been the most usual solution to channel equalization: it is easy to implement and it is relatively simple to obtain the appropriate parameters to assure stability and convergence. The convergence speed of LMS is high and it is robust in nonstationary channels. Furthermore, the computational burden of this algorithm is very low, and it can be implemented in low computational power processors, while nonlinear algorithms cannot. The main drawback of this algorithm is that its final Bit Error Rate (BER) is sensitive to the adaptation step and to the effect of well-classified samples which are far from the classification border. Also, in nonlinear channels its performance is reduced. A sigmoidal nonlinearity applied to the output of the filter can reduce the misadjustment due to well-classified samples, and also reduce the dependence from the adaptation step, although it makes the equalizer much less robust. We introduce a new scheme based on an adaptive combination of an standard LMS equalizer and a sigmoidal output based equalizer, which benefits from the advantages of both equalizers reducing their drawbacks.