Non-parametric spectrum cartography using adaptive radial basis functions

Abstract

This paper presents a framework for spectrum cartography based on the use of adaptive Gaussian radial basis functions (RBF) centered around a specific number of centroid locations, which are determined, jointly with the other RBF parameters, by the available measurement values at given sensor locations in a specific geographical area. The spectrum map is constructed non-parametrically as no prior knowledge about the transmitters is assumed. The received signal power at each location (over a given bandwidth and time period) is estimated as a weighted contribution from different RBF, in such a way that the both RBF parameters and the weights are jointly optimized using an alternating minimization method with a least squares loss function and a quadratic regularization term. Our method is evaluated through simulations, showing a performance (in terms of normalized MSE) that is comparable to semi-parametric methods, and even superior as the number of sensors or RBF increases.