Indoor positioning via nonlinear discriminative feature extraction in wireless local area network

The essential challenge in wireless local area network (WLAN) positioning system is the highly uncertainty and nonlinearity of received signal strength (RSS). These properties degrade the positioning accuracy drastically, as well as increasing the data collection cost. To address this challenge, we propose the nonlinear discriminative feature extraction of RSS using kernel direct discriminant analysis (KDDA). KDDA extracts location features in a kernel space, where the nonlinear RSS patterns are well characterized and captured. By performing KDDA, the discriminative information contained in RSS is reorganized and maximally extracted, while redundant features or noise are discarded adaptively. Furthermore, unlike previous monolithic models, we employ a location clustering step to localize the feature extraction. This step effectively avoids the suboptimality caused by variability of RSS over physical space. After feature extraction in each subregion, the relationship between extracted features and physical locations is established by support vector regression (SVR). Experimental results show that the proposed approach obtains higher accuracy while reducing the data collection cost significantly.