Abstract:

A new positioning algorithm based on semisupervised discriminant embedding manifold learning is proposed to resolve problems deriving from dense reference point deployment, such as tremendous time on location fingerprints collection, calibration and online computation in wireless local area network. The proposed algorithm utilizes a small amount of labeled data and partial unlabeled data to reduce the dimensionality of received signals. Its strong discriminative features are then retained in the lowdimensional forms through solving the objective function optimization. The reduced signals are taken as inputs to the deterministic positioning algorithm and the mapping between localization features and position coordinates is established. The experimental results show that the new algorithm decreases the labor cost to collect fingerprints in the offline stage and calibrate on time. Compared to the traditional methods, the proposed algorithm shows a considerable accuracy improvement in the same positioning environment.