An Ultra Lightweight Neural Network for Automatic Modulation Classification in Drone Communications

Abstract

Automatic Modulation Classification (AMC) is crucial for unmanned aerial vehicle (UAV) systems in non-cooperative communications. It enables UAVs to effectively identify and track signals transmitted by other communication devices. Deep Learning (DL) has been successfully applied to AMC to improve the accuracy of signal classification. Despite this, many DL-based AMC methods, due to their large number of parameters and high computational complexity, cannot be directly applied to UAV platforms with limited computing power and storage space. To address this challenge, we propose an ultra-lightweight neural network (ULNN). This network incorporates a lightweight convolutional structure, attention mechanism, and cross-channel feature fusion technique. Additionally, we introduce data augmentation (DA) based on signal phase offsets during the model training process, aimed at improving the model’s generalization ability and preventing overfitting. Through experimental validation on the public dataset RML2016.10A, our proposed ULNN network achieves an average precision of 62.83% with only 8,815 parameters and reaches a peak classification accuracy of 92.11% at SNR = 10dB. This demonstrates that our proposed ULNN network maintains high recognition accuracy while keeping the model lightweight, making it highly suitable for deployment in resource constrained environments.