Noise Adaptive Optimization Scheme for Robust Radio Tomographic Imaging Based on Sparse Bayesian Learning

This paper addresses the robust signal reconstruction problem caused by different types of noise in radio tomographic imaging (RTI). Most of the existing reconstruction algorithms are built on the assumption of Gaussian noise, which is not the case for practical RTI systems, especially in indoor multipath environments. To weaken the effect of different types of noise on RTI performance, we propose a noise adaptive optimization scheme with sparse Bayesian learning (SBL). Specifically, we model the noise as a mixture of Gaussians (MoG) distribution, which provides the flexibility for describing the unknown and time-varying RTI noise. We further automatically estimate the MoG model parameters as well as the signal under a SBL framework, which makes the signal reconstruction more robust to the complex noise and even outliers. Experimental results in the context of device-free localization show that the proposed scheme can effectively reduce mislocalization and improve localization accuracy in the rich multipath environments, as compared with state-of-the-art reconstruction methods.