A decentralized Gauss-Seidel approach for in-network sparse signal recovery

This paper addresses the problem of monitoring and discovering abnormalities in sensing fields with large-scale wireless sensor networks. By exploiting the sparsity of abnormalities, the signal recovery problem is expressed as an l-l regularized least squares formulation with nonnegative constraints. Furthermore, a decentralized Gauss-Seidel approach is proposed for in-network signal processing. Comparing with its centralized counterpart, the decentralized algorithm improves the robustness and scalability of a large-scale network. Parameter settings of the l-l regularized least squares formulation are studied via theoretical analysis and extensive simulations. An illustrative example of structural health monitoring demonstrates the effectiveness of the proposed decentralized sparse signal recovery algorithm in practical applications.