Parallel implementation of a class of algorithms linking NLMS and block RLS

First a brief review is given of a fully pipelined algorithm for recursive least squares (RLS) estimation, based on so-called `inverse updating´. Then a specific class of (block) RLS algorithms is considered, which embraces normalized LMS as a special case (with block size equal to one). It is shown that such algorithms may be cast in the `inverse-updating RLS´ framework. This allows one to achieve any degree of pipelining, by performing algorithmic transformations which eliminate critical feedback loops in the original algorithms