Adaptive Filtering: Standard LMS vs. Darwinian Algorithm

In this paper we concentrate on the problem of comparing the effectiveness of the Darwinian, and Least Square methods, with the influence of the input signal (statistics) spreadness highlighted. Our ap- proach in this paper is to carry out an extensive suite of simulations for both the Darwinian algorithm and LMS algorithm under varying initial conditions and eigenvalue spread of the signal autocorrelation ma- trix. Considerable progress in understanding the relative merits of LMS and Darwinian results from this comparison, and it is shown that the Darwinian approach yields respectable results. The main prom- ise for improved flexibility lies when supplementary goodness criteria are appended; some preliminary treatment of this "multicriterion" extension to the normal adaptive problem is included.