Increasing Convergence of Adaptive Networks with Noisy Links: Variable Length DILMS Algorithm

Among the distributed adaptive estimation algorithms, incremental least mean square algorithm has a high popularity due to its low computational complexity, and requiring a minimum amount of communication and power. However, it suffers from low initial convergence rate. In this paper we propose a variable tap-length algorithm in order to increase the convergence rate of incremental adaptive networks with noisy links. First, we propose a variable tap-length algorithm in terms of incremental adaptive networks with ideal links. So, without considering the effect of noise on links, we can compare the proposed algorithm with the common distributed incremental least mean square algorithm, and show the convergence rate improvement. Then, we enhance our proposed algorithm to apply it to the networks with noisy links. For this, first we provide an algorithm to reduce the effects of noisy links for a fixed length mode. This algorithm is expressed in terms of a fixed length that can compare the proposed algorithm with algorithms reported in literature to show its better performance. Finally, we express our algorithm in the context of variable length. Simulation results show that the proposed algorithm enhances both the convergence rate and steady state performance.