Robust Diffusion Estimation with Noisy Input based on Pseudo Huber Cost Function for Impulsive Noise Suppression Over Networks

When considering the distributed error in variable (EIV) model containing output impulsive noise interferences, the performance of existing traditional diffusion algorithms dramatically deteriorates, and estimations are biased. To cope with this issue, we develop a robust Diffusion Bias-Compensated Least Pseudo Huber (DBCLPH) algorithm based on the Huber cost function. Firstly, to eliminate input noise-induced bias, the Bias-Compensated Least Pseudo Huber (BCLPH) algorithm is derived according to the unbiasedness criterion and promote it to distributed diffusion corporation network to enhance estimation performance of a single agent. Furthermore, to ensure a better robust estimation performance when confronted with impulsive interferences, an adaptive controlling factor method is designed by incorporating the sigmoid function with disturbance metric, and it can exhibit minimal steady-state estimation error while maintaining a considerable convergence speed. Finally, simulations demonstrate the proposed algorithms effectiveness and superiority robustness when compared with other diffusion algorithms under a robust distributed EIV model with impulsive noises.