Next-Gen Cooperative Relaying: Leveraging Deep Learning for Automatic Modulation Classification with Antenna Diversity

This is the authors' abstract. We don't add key points for this paper.

Automatic Modulation Classification (AMC) is critical in civilian and military wireless communication networks, especially in cognitive radio and electronic warfare applications, where pre-knowledge of signal parameters is unavailable. Most of the current AMC techniques are challenged by fading, low-SNR channels, and non-cooperative network environments, especially in relay-based communication systems. This work offers improved modulation classification in cooperative relaying networks (CRN) using antenna diversity and hybrid deep learning methods while reducing system complexity and preprocessing cost. This paper presents a novel Relay-AMC system for operation in Amplify-and-Forward (AF) mode with Instantaneous Power Scaling (IPS) constraints, with a direct link to the destination. Antenna diversity at the destination with maximum ratio combining (MRC) for AMC is explored. A novel end-to-end hybrid deep learning architecture, DenseLDNN, is designed that integrates DenseNet for spatial feature extraction, LSTM for temporal modeling, and DNN for non-linear classification. The model takes raw I/Q signal samples with no handcrafted features and preprocessing. Simulation outcomes reveal that DenseLDNN surpasses a number of state-of-the-art DL models, such as ResNet, CLDNN, MCNet, and PET-CGDNN by achieving classification accuracy of up to 99.90% and offers a higher average accuracy improvement by 0.11–9.05%. The model exhibits enhanced robustness, particularly for higher-order modulations, such as 64-QAM, and scales well with antenna diversity. This technique provides an exciting pathway for AMC in secure and adaptive military wireless communication networks.