Abstract

One of the main problems facing accurate location in wireless communication systems is non-line-of-sight (NLOS) propagation. Though learning location methods perform well in NLOS environments, learning location methods may be improved further since these methods do not consider outliers in the training data set. In this paper, we extend weighted least squares support vector machine (WLS-SVM) algorithm to mobile location problem. The proposed method can effectively suppress outliers with different weights. In simulation, we analyze the effects of the number of training points, the percentage of outliers in training data set, the standard deviation and mean of outliers, and the standard deviation of measurement error. Simulation results show that the proposed algorithm clearly outperforms three other algorithms (LS method, kernel method and LS-SVM based method).

Key words: Weighted least-squares support vector machine (WLS-SVM), non-line-of-sight (NLOS).