A High-efficiency Collaborative Spectrum Sensing with Gated Recurrent Unit for Multi-UAV Network

Unmanned aerial vehicles (UAV) have been widely used in military and civil fields, and the collaborative multi-UAV network becomes increasingly popular. Cognitive radio (CR) is a feasible way to solve the problem that there is no authorized frequency band for UAV communication. However, the sensing performance of traditional algorithm based on a single artificial design feature is poor in the complex electromagnetic environment including low SNR, high noise-uncertainty, etc., which can not meet the requirement of the cognitive multi-UAV network. Spectrum sensing algorithms based on machine learning can obtain better performance and higher spectral efficiency (SE) by learning the potential feature and rules. Long short-term memory (LSTM) is widely accepted as naturally suitable for processing time series because of its structural characteristics. But the complexity of the algorithm based on LSTM is high due to its redundant structure, which is unsuitable for UAVs as the energy of UAV is always limited. In this paper, we propose a spectrum sensing algorithm based on the gated recurrent unit (GRU). We prove by theoretical derivation and simulation experiments separately that it can not only consume about 1/4 less energy for computation but also achieve slightly better sensing performance compare to the LSTM-based algorithm, which is meaningful. As cooperative spectrum sensing (CSS) can obtain better performance and is suitable for multi-UAV scenarios, we combine the proposed algorithm with a collaborative model which can improve system stability and reach a longer working time than the traditional collaboration model. In general, we propose a high-efficiency CSS scheme based on GRU for a multi-UAV network, which has higher sensing performance, less implementation complexity, and we verify it by theory or simulation.