An Innovative Multimodal Strategy for Classifying Modulation Schemes Using Diverse CNN Models

This study highlights the role of artificial intelligence (AI) and machine learning (ML) in the evolution of sixth-generation (6G) wireless communication systems, which aim to provide unprecedented data rates, ultra-low latency, and ubiquitous connectivity. Unlike previous generations, 6G is expected to integrate AI/ML algorithms across all network layers, from signal processing to resource management. However, challenges such as large data requirements and high computational costs have led to the importance of hybrid systems that combine traditional and deep learning techniques using less data. The study focuses on classifying BPSK and QAM modulation types with limited training data. A multimodal approach is proposed, involving three subnetworks, with experiments conducted in Python using the PyTorch deep learning framework. The results indicate that smaller block sizes improve the training accuracy and loss performance. Among the models tested, ResNet-18 model outperformed GoogLeNet, achieving higher accuracy, particularly in noisy environments, demonstrating its robustness and generalization capabilities.