Deep Learning-Based Hybrid Models for Automatic Modulation Classification

Automatic Modulation Classification (AMC) plays a pivotal role in modern wireless communication systems, enabling receivers to autonomously identify modulation schemes without requiring prior knowledge of the transmitter's settings. This capability is essential for enhancing the efficiency, robustness, and adaptability of real-time applications such as 5G networks, dynamic spectrum allocation, and interference mitigation. Emerging AMC approaches increasingly utilize deep learning models, including Convolutional Neural Networks (CNNs), Residual Neural Networks, and Attention-based architectures. In this work, we propose three novel hybrid deep learning architectures that combine the strengths of the Visual Geometry Group Network (VGGNet) and Residual Attention Networks (RanNet), which are considered as the benchmark models. Trained on the RadioML 2016 dataset, which comprises ten modulation schemes, the proposed models outperform VGGNet with an impressive accuracy of 87.6% at an 18dB Signal-to-Noise Ratio (SNR), highlighting the potential of our hybrid architectures to advance AMC by offering robust, scalable solutions for next-generation wireless communication systems.