White spaces pattern finding and inference based on machine learning for multi-frequency spectrum footprints

Abstract

Spectrum surveys performed worldwide demonstrate that the spectrum utilization efficiency is less than 0.25. Therefore, the traditional long-term licensed spectrum allocation by regulators is not sustainable. Although dynamic spectrum access networks allow increasing the efficiency of spectrum utilization, coexistence is still a major problem. In this paper, we investigate the capability of machine learning for estimating the white space availability based on a dataset from a spectrum survey from 170 MHz to 1 GHz. In addition, we present an algorithm for minimizing the effect of hidden nodes on wrongful spectrum allocation and interference. Our optimization algorithm based on supervised machine learning allows increasing the spectrum utilization efficiency with a factor 5 (from 0.09 to 0.47) in the surveyed region. In addition, our algorithm allows decreasing the interference probability caused by the effect of hidden nodes by a factor 6, compared to the traditional distributed allocation of spectrum in Dynamic Spectrum Access networks.