Data Pre-Processing in Compressive Sensing Based Indoor Fingerprinting Positioning

Abstract

Presented is a novel compressive sensing (CS) based indoor positioning approach, which uses the signal strength differentials (SSDs) as location fingerprints (LFs). By using certain kernel-based transformation basis, the 2-D target location is represented as an unknown sparse location vector in the discrete spatial domain. Then it just takes a little number of online noisy SSD measurements for the exact recovery of the sparse location vector by solving an l1-minimization program. In order to effectively apply CS theory for high precision indoor positioning, we further import some data pre-processing algorithms in the LF space. Firstly, to mitigate the influence of large measurements noise on the recovery accuracy, a LF space denosing algorithm is designed to discriminate the unequal localization contribution rate of every SSD measurement in each LF. The basic idea of the denosing algorithm is to transform the original LF space into a robust and decorrelated LF space. Moreover, in order to lower the high computational complexity of the CS recovery algorithm, several LF space filtering algorithms are also exploited to remove certain percentage of useless LFs in the radio map according to the real-time RSS observations. The performance of these denosing and filtering algorithms are investigated and compared in real-world WLAN experiment test. Both experimental results and simulations demonstrate that we achieve remarkable improvements on the positioning performance of the CS based localization by using the proposed algorithms.