Semi-supervised Positioning Algorithm in Indoor WLAN Environment

In Wireless Local Area Network (WLAN) positioning system, the most popular solution for RSS-based positioning is the fingerprinting architecture. In this paper, we present a novel algorithm, known as Semi-supervised Discriminant Embedding (SDE), to reconstruct a radio map by using real-time signal-strength values received at random points. Instead of deploying dense reference points, our approach takes advantage of less labeled data and partial unlabeled data to transform into lower-dimensional feature signals. Through solving the objective functions optimization, with strong discriminative features in Receive Signal Strength (RSS) are retained in the low-dimensional space. We conducted experiments in our office area with a realistic WLAN environment. Compared to the traditional methods, the experimental results show that the proposed algorithm has considerable accuracy improvement in the same positioning environment. Furthermore, the results also show the size of training samples can be greatly reduced in the proposed algorithm in order to achieve the similar accuracy of traditional approaches. That is, the cost of collecting fingerprints in the offline stage and calibrating database in the online stage are thus reduced.