Non-parametric Modeling with Gaussian Processes for Spatial Radio-Scene Analysis

To increase area spectral efficiency, overlay use of spectrum already assigned to a primary system is considered in the context of cognitive radio research. Minimizing interference with the primary system is a key requirement and equivalent to the problem of transmission opportunity identification. Transmission opportunities are unused slots in the communication space which is spanned by time, frequency, and location. In this paper we describe a method to identify transmission opportunities by modeling the spatial power distribution with Gaussian processes. This approach to distributed sensing is verified with real-world measurements using the GSM downlink band as the primary system and contrasted with traditional wave propagation modeling approaches.