Assessment of finite precision limitations in LMS and BLMS adaptive algorithms

This paper presents a comparison of the effects of finite arithmetic on least mean squares (LMS) and block least mean squares (BLMS) adaptive algorithms. It is demonstrated, using both simulation and analytic results, that the BLMS algorithm consistently achieves improved converged noise performance compared to an LMS algorithm using the same word lengths. In the case dealt with here the improvement is equivalent to an excess 2-3 bits in word length.