Adaptive Sensor Selection for Nonlinear Tracking via Sparsity-Promoting Approaches

Abstract

In this paper, we propose a few sparsity-promoting approaches for sensor selection in wireless sensor networks (WSNs) to reach the optimal tradeoff between the tracking performance and the sensing-communication cost. The proposed framework is valid for general nonlinear target tracking without explicit linearization and not restricted to any specific estimator. We formulate the sensor selection problem as the design of a sparse selection vector. The cardinality of the selection vector is added as a sparsity-promoting penalty term to the cost function where the conditional posterior Cramer–Rao lower bound is used as the criterion for sensor selection. To cope with large-scale WSNs, by combining iterative reweighted $\ell _1$ -norm minimization with the accelerated proximal gradient method and the alternating direction method of multipliers (ADMM), we develop two efficient sensor selection algorithms, respectively. We further develop a low-complexity distributed version of the ADMM where each sensor makes a local sensor selection decision. We test the proposed algorithms by simulations based on an extended Kalman filter using analog data and a particle filter using quantized data, respectively, which provide valuable insights and demonstrate the proposed algorithms’ performance.