Sparsity-enabled radio tomographic imaging using quantized received signal strength observations

Abstract

Detection and tracking of objects in wireless sensor networks (WSN) is one of the active research areas in the signal processing community, where device-free localization (DFL) techniques are the potential contributors. Radio tomographic imaging (RTI) is one such DFL technique where the loss field surrounded by any obstacle can be visualized with the help of the received signal strength (RSS) of various transceiver nodes. Due to strict communication constraints and the necessity of low-complexity sensors, we are motivated towards quantized RSS (q-RSS) observations in the RTI system. The RSS data is digitized before sending it to the fusion center (FC). Therefore, uncertainty due to RSS quantization error leads to momentous performance degradation of the RTI system. The proposed support vector regression (SVR) technique reduces this uncertainty. The ϵ-SVR and ν-SVR techniques are affected by noise levels, and their performance is enhanced by the use of fused l2-SVR (F-l2-SVR). However, these estimators have higher computational complexity due to the large number of support vectors. Hence, a robust sparsity-based estimator using the linear programming SVR (LP-SVR) is proposed to reduce the computational complexity. Further, a novel fused lasso-based l1-SVR (FL-l1-SVR) estimator is proposed that provides estimated spatial loss field (SLF) as close as to the true SLF. Finally, the performance metrics and the reduced number of support vectors lead to the efficacy of the proposed FL-l1-SVR estimator over other estimators under q-RSS observations in the RTI system.