A parameter estimation algorithm based on a pseudolinear regression and an orthogonal matrix decomposition is presented. It operates on data values directly and reduces the necessary dynamic range for computation. The algorithm shows rapid convergence even in noisy conditions and is able to track system changes effectively. Some comments on convergence and computational complexity are made. A hierarchical signal flow graph for fast versions of the algorithm derivation is presented. Simulation results are presented. They show the performance of the algorithm in different set-ups, and demonstrate that the proposed algorithm is stable with respect to the feedback path and superior to gradient search based algorithms in terms of adaptation speed, achievable minimum mean squared error and computational requirements. It is more powerful than the known sequential techniques for the singular value decomposition.
Reviewer: T.Semerdjiev (Sofia)