Short word-length LMS filtering

The distinctive advantage of short-word-length systems (single-bit, two-bit, or ternary), namely, the hardware implementation simplicity, has not been put into effect due to its unresolved LMS adaptivity problem. The conventional LMS family of adaptive algorithms fail to converge if translated to the short-word-length domain. In this paper we propose a 2-bit-domain LMS adaptive filtering structure for noise cancelling where all input, output, and filter coefficients are in 2-bit format. The proposed structure is designed and analyzed, and its performance has been evaluated (in comparison to the conventional Widrow-Hoff multi-bit LMS algorithm) in terms of convergence properties, signal-to-noise improvement, and computational complexity. Simulation results showed that the proposed 2-bit adaptive structure exhibits performance that is equivalent to the infinite-precision LMS algorithm.