Diffusion Bias-Compensation RLS Estimation Over Noisy Node-Specific Networks

We study the node-specific parameter estimation problem, where agents in a network collaborate to obtain the different but overlapping vectors of parameters, which can be of local interest, common interest to a subset of agents, and global interest to the whole network. We assume that all the regressors and the measurements are corrupted by additive noise. For these settings, a bias-compensation recursive-least-square algorithm based on a diffusion mode of cooperation is proposed; its stability is obtained via the detailed derivation of convergence in the mean sense. In addition, a closed-form expression for the algorithm's mean-square deviation is also provided to evaluate the steady-state performance of the whole network. Finally, we present simulation results that indicate the efficiency of the proposed method.

Automated summary

The study focuses on the node-specific parameter estimation problem and proposes a bias-compensation recursive-least-square algorithm based on a diffusion mode of cooperation. The stability and mean-square deviation of the algorithm are analyzed for evaluating the network's steady-state performance, and simulation results demonstrate the efficiency of the proposed method.