A generalized ℓp-ℓq norm minimization approach for distributed estimation in sensor networks

摘要

A generalized ℓp-ℓq norm minimization approach for in-network distributed estimation is proposed. Different from the existing techniques which are assuming that all the nodes are affected by the same noise model, either Gaussian or non-Gaussian. We consider a general and practical scenario, the spatially distributed nodes are affected by different noise models. To achieve robust estimation performance in different noise environments, each node solves a specific ℓp-norm minimization problem corresponding to the noise model. Meanwhile, the ℓq-norm penalty is imposed on the cost function to exploit prior information of the system, such as sparsity.