Robust Diffusion Adaptive Networks with Noisy Link and Input

In this paper, we study the problem of adaptive parameter estimation for multi-agent distributed networks, where the input regression vectors of network nodes contain Gaussian noises, while the output values and the communication link are polluted by impulse noises. In this case, the estimation performance of traditional diffusion LMS algorithms and most of the state-of-the-art robust distributed algorithms for output impulse noises will degrade significantly. Aiming at this problem, the Minimal Disturbance Bias-Compensated Diffusion Least Mean Square (MDBC-DLMS) algorithm proposed in this paper can effectively suppress noise interference and achieve an acceptable estimation result of the target parameter vector. MDBC-DLMS uses the principle of minimal disturbance to dynamically update the combination coefficients of the diffusion algorithm to effectively suppress the output and link impulse noise. At the same time, it performs dynamic real-time estimation of the input noise variance information to compensate for the estimation bias caused by the input noise. The simulation results show the excellent estimation performance and effectiveness of the method proposed in this paper, and it can accurately estimate the variance information of input noise at the same time.