Accurate and Robust Device-Free Localization Approach via Sparse Representation in Presence of Noise and Outliers D. S. Wang 1 , X. S. Guo 2 , Y. X. Zou* 1 1 ADSPLAB/ELIP, School of ECE, Peking University, Shenzhen 518055, China 2 Department of Electronic Engineering, University of Electronic Science Technology of China, Chengdu 611731, China *{zouyx@pkusz.edu.cn} Abstract—Device-free localization (DFL) aims at locating the positions of targets without carrying any emitting devices by monitoring the received signals of preset wireless devices. Research showed that the localization accuracy of conventional DFL algorithms decreases in presence of noise and outliers. To tackle this problem, this paper firstly proposes to study the DFL via sparse representation and the target localization is formulated as a sparse representation classification (SRC) problem. Specifically, an overcomplete sample dictionary is constructed by received signal strength and the target can be located by SRC method. To suppress the adverse impact of noise and outliers, we formulate the DFL-SRC problem in signal subspace. Two DFL algorithms termed as SDSRC and SSDSRC are derived. Experimental results with real recorded data and simulated interferences demonstrate that SDSRC and SSDSRC outperform the nonlinear optimization approach with outlier link rejection in terms of localization accuracy and robustness to noise and outliers. Keywords — Device-free localization; received signal strength (RSS); sparse representation classification (SRC); signal subspace; noise and outliers I. INTRODUCTION In recent years, device-free localization (DFL) [1, 2] technique has attracted tremendous interests due to its ability to locate targets that do not carry any devices. Wireless devices (termed as anchor points, AP) fixed at known positions are transceivers. Once a target moves into the deployment area of the wireless sensor network (WSN), some wireless links will be shadowed as a result of target occlusion. By utilizing the variation of the received signal strength (RSS) measurements of these links, the position of the target can be estimated. Meanwhile, DFL has many applications, it can not only be used in indoor localization including family security and smart aging, but also outdoor localization such as border protection and wildlife searching, etc [3]. There are some research outcomes for DFL problem. Zhang et al. presented a model of signal dynamics to allow tracking targets based on radio signal strength indicator (RSSI) [4, 5]. Youssef et al. proposed a fingerprinting-based approach which locates the target by comparing the RSS of the links that are measured in the online phase with the fingerprint built in the offline phase [6-8]. Wilson and Patwari referred to DFL problem as radio tomographic imaging (RTI) which is an ill- posed inverse problem and applied regularization method to solve it [9, 10]. Wang et al. formulated the DFL problem as a sparse signal reconstruction question, and proposed a novel Bayesian greedy matching pursuit (BGMP) algorithm to realize sparse signal reconstruction and location estimation [11]. However, in practical applications, there exist noise and outliers. Literature study showed that all the algorithms introduced above do not take outlier links into account. To tackle outlier issue, Xiao et al. proposed a novel nonlinear optimization approach with outlier link rejection (NOOLR) for RSS-based DFL [12]. In their study, a detection strategy is firstly applied to find out affected links. Then, the outlier links are rejected by k-means algorithm. Finally, a nonlinear optimization technique is applied to locate the target. Their experimental results showed that the NOOLR outperforms the RTI approach. Careful evaluation reveals that the performance of NOOLR will degrade when noise and outliers become severe. To tackle this problem, we firstly investigate to formulate the DFL problem under the sparse representation framework and locate the target by sparse representation classification (SRC). Firstly, we utilize the RSS sample matrices, acquired by collecting the RSS measurements of wireless devices at predetermined reference points (RP) whose positions are known, to construct an overcomplete dictionary. Then the target can be located by solving a l1 -minimization problem and choosing the smallest residual. To suppress the adverse impact of noise and outliers on the localization accuracy, we formulate the DFL-SRC problem in signal subspace with eigenvalue decomposition technique. To evaluate the performance of our proposed methods, we have conducted extensive experiments to compare our algorithms with NOOLR [12]. It is demonstrated that SDSRC and SSDSRC are robust to noise and outliers and outperform NOOLR in terms of localization accuracy. II. PROBLEM FORMULATION In this section, the concept of DFL is introduced first. By analyzing the property of DFL data mode, the DFL formulated as SRC is proposed and the algorithm SDSRC is summarized. A. The Description of DFL Problem As depicted in Fig. 1, wireless devices that communicate with each other are fixed at the edge of the DFL area. Assuming that D APs are employed in DFL. When there is no target existing in the DFL area, let R i,j denotes the measurement 199 978-1-5090-4165-7/16/$31.00 ©2016 IEEE 199