Spatio-temporal Spectrum Load Prediction using Convolutional Neural Network and Bayesian Estimation
Xiangyu Ren, Hamed Mosavat-Jahromi, Lin Cai, David Kidston · 2020
Radio spectrum is a limited and increasingly scarce resource, which motivates alternative usage methods such as dynamic spectrum allocation (DSA). DSA of a frequency band requires an accurate prediction of spectrum usage in both the time and spatial domains with minimal sensing cost. In this paper, we address challenge in two steps. First, in order to make the best use of the limited sensors in the region, we deploy a deep learning prediction model based on convolutional neural networks (CNNs) and residual networks (ResNets), to predict spatio-temporal spectrum usage at the sensors' locations. Second, given an area enclosed by a few sensors, a Bayesian estimation model is proposed to first derive the location distribution of a transmitter, and then obtain the interference power distribution within the area. Simulation results show the efficacy and efficiency of the proposed prediction models.