In this brief, we develop a least mean square (LMS) algorithm thatnconverges in a statistical sense to the global minimum of the meannsquare error (MSE) objective function. This is accomplished bynestimating the gradient as a smoothed version of the MSE, The smoothednMSE objective begins as a convex functional in the mean. The amount ofndispersion or smoothing is reduced, such that over time it becomes thentrue MSE as the algorithm converges to the global minimum. We show thatnthis smoothing behavior is approximated by appending a variable noisensource to the infinite impulse response (IIR)-LMS algorithm. We show,nexperimentally, that the proposed method does converge to the globalnminimum in the cases tested. A performance improvement over the IIR-LMSnalgorithm and the Steiglitz-McBride algorithm has been achieved
展开▼
