A finite wordlength analysis of an LMS-Newton adaptive filtering algorithm

The effects of quantization in an least-mean-square (LMS)-Newton adaptive filtering algorithm are investigated. The algorithm considered uses an optimum convergence factor that forces the output a posteriori error to become zero in each iteration. The propagation of errors due to quantization in the internal variables of the algorithm is investigated, and a closed-form formula for the excess mean square error due to quantization is derived. Fixed-point arithmetic is assumed throughout. Several simulations confirm the accuracy of the formulas presented.<
>