Diffusion fractional tap-length algorithm with gradient-based cooperation to enhance the performance of adaptive networks with noisy links

In this paper, we consider the adaptive estimation in wireless sensor networks (WSNs) with diffusion strategy, where the communication links between nodes are noisy, and the parameter vector of interest has unknown or variable tap-length. The diffusion fractional tap-length (FT) algorithm seeks a tap-length estimator in real-time through local interactions to best balance steady-state error and the convergence rate. However, its performance is severely degraded in the presence of noisy links. This incorrect estimation of the length due to the noise of the links is followed by an inaccurate estimate of the coefficients. This fact shows the requirement of a noise-robust method. On this basis, we propose gradient-based cooperation in which the cooperation rate is adjusted based on the noise level using the concept of gradient. The concept of gradient-based cooperation is a way to make use of spatial correlation, and at the same time, to reduce the effects of noisy links. Simulation results show the superiority of the proposed method in the diffusion adaptation over networks with noisy links.