Acoustic Source Localization Method with Variable Power in WSN Based on LSSVR Regression Learning

Abstract

To realize accurate acoustic source localization with variable power in noisy environments, a novel acoustic source localization method with variable power based on LSSVR regression learning (ALVP-LRL) was proposed. The ratio values of any two adjacent nodes’ theoretical measurements of acoustic energy comprise feature vector, which has stable mapping relationship to source’s coordinates. LSSVR was applied to build regression models approximately reflecting that mapping relationship. By inputting feature vector constructed by real measurements into the regression models, the models’ outputs were then regarded as the estimated coordinates. Experiments were performed in 121 test locations. As SNR level reduced, amount of test locations where location errors were less than 2 meters by ALVP-LRL method changed from 77 to 54, while that amount by MLE method rapidly decreased from 121 to 11. It shows ALVP-LRL method preliminarily achieves certain effects and has more significant advantages on lower SNR occasions.