Cooperative sensing for cognitive radio using decentralized projection algorithms

Abstract

Sensing the radio spectrum is an essential feature of cognitive radio. What is mostly important for a cognitive transmitter is to know the power spectral density at the location of its intended receiver and of the primary receivers. This requires, as a whole, knowledge of the spatial distribution of the power spectral density, an information that could be delivered by a network of sensors distributed over the territory, where each node estimates the local power spectral density and sends this information to a control node. The criticalities of this strategy are the limited sensing capabilities of the single node, shadowing effects and potential congestion around the sink nodes collecting all the estimated wideband spectra. In this paper, we propose a fully decentralized iterative algorithm allowing the network to project the measured spatial spectral density onto the useful signal subspace with the minimum convergence time. The idea is based on the assumption, typically valid in practice, that the spatial distribution of the power spectral density, for each frequency, is mostly concentrated over a vector space of dimension much smaller than the number of sensing nodes. The proposed technique yields a considerable reduction of estimation noise and undesired shadowing effects, without requiring the presence of a centralized control node. Interestingly, we show that in our set-up, the selection of the signal subspace dimension depends not only on the bias error and on the noise variance, as in any MSE algorithm, but also on the transmit power available at each node.