Sampling spectrum occupancy data over random fields: A matrix completion approach

The performance of cognitive radio networks is fundamentally determined by the availability of spectrum resources. Detailed measurement campaigns are needed to collect the spectrum occupancy data to obtain a deeper understanding of the spectrum usage characteristics in cognitive radio networks. This approach, however, is usually inefficient due to the ignorance of the spatial, temporal and spectral correlations of spectrum occupancies, and unpractical because of the geographical and hardware limitations of the cognitive radio nodes. In this paper, we apply the theory of random fields to model the spatial-temporal correlated spectrum usage data, using the two dimensional Ising model and the Metropolis-Hastings algorithm respectively. To efficiently obtain the spectrum occupancy, we adopt the matrix completion technique that leverages the low-rank structure of the data matrix to recover the original data from limited measurements. Simulation results validate effectiveness of the proposed algorithm.