Self-Learning Localization with LSSVM Algorithm in Wireless Sensor Networks

Abstract

Received Signal Strength Indicator ( RSSI ) is a low-cost ranging method for localization of wireless sensor network. The accurate mathematic model of RSSI ranging is difficult to depict, since the ranging is easily influenced by environmental factors. This paper puts forward an argument that using signal relationship of beacon nodes, Least Square Support Vector Machine (LS-SVM) locates unknown nodes through training vector machine, which can simplify the establishment process of the ranging model. This paper sets forth the location principle of wireless sensor networks based on LS-SVM, through the analysis of LS-SVM localization with three forms of kernel functions and the analysis of localization error of three forms of kernel functions , the paper presents a kernel function with optimal localization precision and does deep analysis on the parameters of the kernel function. And on this basis, the LS-SVM localization technology was examined with TelosB nodes placed on different directions in an open playground. The experimental results show that the method can effectively locate unknown nodes and get high localization accuracy.