Dictionary Refinement for Compressive Sensing Based Device-Free Localization via the Variational EM Algorithm

Abstract

Device-free localization (DFL) plays an increasingly important role in many security and military applications. It can realize localization without the requirement of equipping targets with any devices for signal transmitting or receiving. To reduce the number of measurements in DFL, compressive sensing (CS) theory has been applied. By exploiting the sparse nature of location finding problem, the target location vector can be estimated from a few measurements. However, in changing environments, measurements may diverge from those in a fixed dictionary (sensing matrix), and the mismatches between the dictionary and runtime measurements can significantly deteriorate the localization performance of CS-based DFL methods. To address this, we propose a novel dictionary refinement-based DFL method. It adopts the saddle surface model to characterize the shadowing effects caused by targets and parameterizes the dictionary with the shadowing rate of each link as the underlying parameters. Then, the variational expectation-maximization algorithm is adopted to realize joint localization and dictionary refinement. Simulation results show that the proposed approach achieves higher accuracy and robustness compared with the state-of-the-art fixed dictionary DFL methods.