A Multidimensional Fingerprint Positioning Method using Support Vector Machine

Fingerprint positioning is an effective positioning method for a rich scattering indoor environment. Most fingerprint positioning methods are based on the received signal strength (RSS). However, as a coarse-grained parameter, RSS may fluctuate at a fixed position, making the positioning results unstable. The recent advanced positioning systems use the amplitude information of channel state information (CSI) instead of RSS. Yet the non-line-of-sight (NLOS) environment limits their applicable scene. This paper proposes a dispersion maximization least squares support vector machine (DM-LSSVM)-based multi-dimensional fingerprint positioning model that can work in an NLOS environment. We first use a Multiple Signal Classification (MUSIC) algorithm to estimate the angle-of-arrival (AoA) and multipath number, and then a multi-dimensional fingerprint library is built by combining the RSS value with the estimated AoA and multipath number. Finally, we introduce the LSSVM as the positioning algorithm to realize positioning and calculate the weight of each fingerprint through the dispersion maximization method, implement a method which can enhance the positioning accuracy of AoA in NLOS scenarios through RSS and multipath numbers. The result shows that in a high NLOS deployment environment, the median accuracy is 0.38m.