Location-aware cooperative spectrum sensing via Gaussian Processes

We present a new paradigm to perform spatial spectrum sensing in cooperative cognitive radio networks. In these networks, nodes cooperate to infer information on spectral and spatial occupancy. Following the recently proposed interference temperature metric, we formulate the problem via semi-parametric Gaussian Process modelling. This allows for a development of a flexible probabilistic framework to perform inference on a function space. Our methodology enables sensing spatial reusable zones, by means of a collaborative scheme to estimate the distribution of power emitted by the primary user in space. Hence, rather than a conventional binary decision process in space (busy/idle), our scheme provides a probabilistic spectrum access scheme for cognitive radio networks. We develop an algorithm, based on the Iterated Conditioning on the Modes (ICM) estimation to predict spatial spectral holes at any arbitrary spatial location. The ICM approach is computationally efficient, and suitable for real world applications. Numerical results demonstrate the benefits obtained by the proposed algorithm.