Robust Multitask Diffusion Affine Projection Algorithm for Distributed Estimation

When the disturbance of impulsive noise exists in the multitask network, the convergence behavior of the traditional multitask diffusion affine projection (AP) algorithm (MD-APA) is significantly suppressed. To alleviate this problem, in this brief, a robust MD-APA is proposed based on maximum correntropy criterion (MCC), which is called MD-APMCC algorithm. Due to the shortcomings of the fixed kernel width, this brief adopts a robust adaptive kernel width strategy to increase the estimation behavior of the MD-APMCC algorithm. Besides, the convergence behavior of MD-APMCC algorithm is studied to derive the convergence range of step-size and the theoretical steady-state mean square deviation (MSD) of the whole network. The simulation verification demonstrates that the proposed MD-APMCC algorithm appears better estimation behavior than MD-APA and MD-APSA for multitask distributed estimation under impulsive noise interference, and the theoretical steady-state MSD of MD-APMCC algorithm is obtained through mean square analysis, which has been well verified by several simulations.

Automated summary

This paper proposes a robust MD-APA algorithm based on the maximum correntropy criterion (MCC) to address the convergence issue caused by impulsive noise in a multitask network. The algorithm adopts a robust adaptive kernel width strategy to enhance estimation behavior and analyzes the convergence range of step-size and theoretical steady-state mean square deviation (MSD) of the whole network.