A Study of Singular Value Decomposition for Wireless LAN Location Fingerprinting

A challenging problem in the wireless LAN localization is the severe fluctuation of the receive signal strength (RSS) even for the stationary client. The time-varying RSS which is resulted from the noise deteriorates the system accuracy drastically. Previous works presented various approaches to improve the positioning accuracy. Singular-Value-Decomposition (SVD) based-noise reduction technique is one good example of these methods. In the SVD-based positioning system, the Hankel-form matrix constructed from RSS of the available access points is decomposed into two subspaces. The important feature is obtained by suppressing the noise-related subspace and reconstructing RSS from the signal subspace only. Experiments in a real-world indoor environment show that the positioning error can be reduced by choosing proper SVD parameters.