Robust Distributed Optimization in Wireless Sensor Network

The main challenges in estimating parameters in a sensor network are link failure and impulsive noise. Under such adverse conditions the conventional square error cost function based learning algorithms offer poor estimation performance. In this paper we propose a robust error saturation nonlinearity as cost function and use it for decentralized incremental estimation of parameters. Simulation results show that the proposed estimation method offers more robust distributed estimation compared to Huber loss function based estimation both in link failure and impulsive noise conditions.