An improved quantization model for the finite precision LMS adaptive algorithm

An improved model is presented for the analysis of the errors due to the finite word length digital implementation of the least-mean-square (LMS) algorithm. The new formulation takes into consideration the correlation between the correction term and the corresponding quantization error in the updating equation. Analytical expressions are derived for the MSE (mean-square error). This facilitates the determination of the gain constant required to obtain a desired level of residual MSE. The results are suitable for design purposes. Simulation results are presented which validate the theoretical model proposed.<>