Joint Resource Optimization for UAV Swarm Communication System Based on DQN

For the line-of-sight (LoS) link space, in order to maximize the data transmission capacity of the unmanned aerial vehicles (UAV) swarm communication system, the joint spectrum, power and flight path resources optimization are carried out. The system contains UAV-to-base station (U2B) and UAV-to-UAV (U2U) data links. In order to improve the spectrum usage efficiency, the U2U links can reuse the orthogonal spectrum of the U2B links. Based on deep Q network (DQN), the UAV are regarded as agents. By interacting with the system, the U2U links intelligently selects the appropriate communication resources. Considering the data characteristics of spatial UAV distribution, three different attention module networks are further introduced including the squeeze-and-excitation module (SE), the convolutional block attention module (CBAM) and the channel attention (CA) to improve the system performance. SE, CBAM and CA are trained and tested respectively. Simulation data shows that compared with benchmark algorithm, the proposed algorithms can reduce the complexity of training and obtain higher data transmission capacity.