Analysis of a recursive least squares hyperbolic rotation algorithm for signal processing (Q1100129): Difference between revisions

In some situations occurring in digital signal processing (DSP), one has to solve the real valued matrix equation \(Xw=s\), where s is a \(p\times 1\) vector, X is a \(p\times n\) matrix of full column rank and w is an \(n\times 1\) vector. It is asked to find the least square solution w. To this end one constructs a new positive definite matrix \(C=X\) TX which is factorized as \(C=R\) TR, the so-called Cholesky factor. In recent digital systems one acts recursively on the signal, which is segmented by a moving window. The problem is to find the new Cholesky factor \(\bar R\) in terms of the initial R when a row is added to X, to so-called updating (UD), or when a row is deleted from X, the so-called downdating (DD). In the paper a solution to this last problem is given by use of the so- called hyperbolic rotations, a special linear transformation. It is shown that this approach offers some advantages, fewer multiplications and additions, over other approaches where orthogonal rotations are used. The paper is splendidly written, the underlying ideas are clearly explained and it provides a beautiful example of interesting mathematical problems raised by modern DSP.