Bi-scale LMS equalization for improved performance

Recent results show that an adaptive transversal least-mean-square (LMS) equalizer in a narrowband-interference dominated environment operates at a mean weight vector that is different from that of the Wiener equalizer of the same structure. In addition, the time-varying component of the LMS weight vector results in a steady-state mean square error (MSE) that can be substantially lower than that for the fixed Wiener equalizer. However, the MSE for this LMS equalizer is higher than the MSE prediction in which LMS is assumed to be operating in a neighborhood of the Wiener weight vector. We find that —although the transversal LMS equalizer itself does not produce the Wiener weight vector as its steady-state mean — the adaptive algorithm can be modified so that its mean weight vector is the fixed Wiener weight vector, while simultaneously facilitating the time-varying weight behavior that is responsible for the reduction in MSE. The resulting bi-scale LMS (BLMS) algorithm achieves further improvement in MSE.