On convergence of weighted least squares algorithms [adaptive control]

The question of convergence of adaptive algorithms for prediction, estimation and control of discrete time stochastic systems has recently received considerable attention. For systems incorporating a delay of N units between the application of a control signal and the response to that signal, various `N-recursion´ algorithms have been proposed and analyzed. These algorithms involve a bank of N interlaced parameter estimators, using stochastic approximation or modified recursive least squares (RLS) type iterations. The author attempts to give a more complete, but concise, convergence analysis for the general (N>1) case. Some differences between the author´s results and the previously reported results are pointed out