Positioning via Spectrum Sensing Using Multidimensional Scaling and Metric Learning with an Application in Millimeter Wave Networks

Position based clustering and routing are expected to be promising strategies in future wireless networks. However, the position information may be unavailable in many cases (e.g., the loss of GPS signals), while most studies assume the position information, or at least some partial information, to be known. In this paper, we first leverage the spectrum sensing results in cognitive radio networks, to reconstruct the node positions. The principle is that nearby nodes should have similar spectrum sensing results. However, the only direct information obtained from spectrum sensing is the similarities of spectrum among nodes. To fully exploit the meager information, the technique multidimensional scaling (MDS), as one of the basic algorithms, which has been widely used in machine learning, sociology and psychology, will first be used to construct the positions. Next, the technique of metric learning will be used to further improve the performance by fine tuning the definition of metric of the similarity. The numerical results, in the context of cognitive millimeter wave (mmWave), show that the proposed scheme can well reconstruct the node positions, which could be of great importance for potential applications, such as clustering and routing, for future networks.