面向无人机集群通信的轻量化自动调制识别OA

Unmanned Aerial Vehicle(UAV)swarms are becoming a key enabling technology for intelligent wireless communications in dynamic and adversarial environments.Automatic Modulation Recognition(AMR)is crucial for spectrum sensing,but Deep Learning(DL)-based solutions are typically too resource-intensive for UAV edge platforms.To overcome this challenge,a Federated Learning Adaptive Pruning Network(FLAP-Net)designed specifically for UAV swarms is proposed,aiming to balance recognition accuracy,efficiency,and communication overhead.It employs a lightweight classifier,Feather-MSA,combining multi-scale convolution,Bidirectional Gated Recurrent Unit(BiGRU),and an additive attention mechanism to effectively extract In-phase and Quadrature(I/Q)signal features.To further reduce transmission costs,a cosine similarity-guided adaptive pruning mechanism and a channel-aware weighted aggregation algorithm are introduced to dynamically adjust the contribution of UAV nodes based on channel conditions.Experiments on the RadioML2016.10b dataset show that FLAP-Net achieves an accuracy of 93％at a Signal to Noise Ratio(SNR)of 4 dB,with a single-sample inference latency of only 2.42 μs and a peak communication bandwidth per UAV of less than 27 Mb/s.The results highlight the practicality of FLAP-Net in real-time collaborative communication for UAV swarms.