New data-reusing LMS algorithms for improved convergence

A geometric framework is adopted which is used to clearly elucidate the operation of and relationships between the LMS, DR-LMS, and NLMS algorithms. This geometrical framework facilitates the proof of an analytical result which explains the superior convergence rate performance of the NLMS algorithm. A new class of computationally efficient data-reusing algorithms is then introduced which provides significant convergence rate improvement over the DR-LMS algorithm. The improved performance is verified with simulations.<<ETX>>