UAV Spectrum Sensing of Normalized Spectrum based on GRU network

Abstract

Spectrum sensing is a promising technology to deal with the increasing scarcity of spectrum resources since Internet of Things devices have increased tremendously. Specifically, in the cognitive unmanned aerial vehicle (UAV) network, UAVs can opportunistically access the licensed spectrum of the primary user when it is sensed idle. However, since the environment of UAV is varying all the time, the noise power and signal-to-noise ratio (SNR) of the channel are uncertain in UAV communications. Thus, traditional methods are difficult to resist the influence of noise uncertainty on spectrum sensing. In this paper, we combine data preprocessing and machine learning to improve the performance of UAV spectrum sensing and resist the influence of noise power and SNR uncertainty. Specifically, we propose Gated Recurrent Unit (GRU) network spectrum sensing based on normalized spectrum. The simulation results show that compared with the traditional GRU and Long Short-Term Memory (LSTM) network, the proposed algorithm has better performance when the SNR and noise power are uncertain. Besides, when the number of sampling points is more than 600, both the training time and testing time are less than half of traditional GRU and LSTM network.