Enhanced radio tomographic imaging with heterogeneous Bayesian compressive sensing

This work explores the novel use of Bayesian compressive sensing (BCS) in radio tomographic imaging (RTI), which aims at addressing the performance degradation of shadow fade imaging due to multipath interferences, through the sophisticated efforts on enhancing BCS with the capability of heterogeneous-noise-variance learning. Our contribution is twofold. Firstly, we incorporate a hierarchical model of heterogeneous noise variances into sparse Bayesian learning, which can contribute to the enhancement of BCS in terms of noise-variance awareness. Then, under our enhanced BCS (namely heterogeneous BCS) framework, we develop two learning algorithms for the RTI reconstruction. Theoretical analysis will show the potential advantages of using our heterogeneous BCS in mitigating the effect of multipath interferences, as well as in improving the RTI performance with our learning algorithms. Finally, the experimental results in the context of device-free localization and tracking are reported to demonstrate the effectiveness of the proposed approach.