Automatic modulation recognition (AMR) is a key enabler for intelligent spectrum utilization in 5G-and-beyond wireless systems, requiring both high classification accuracy and low computational complexity. This paper proposes a lightweight hybrid deep learning framework, termed CBLGNet, that integrates convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM), and gated recurrent units (GRU) for efficient AMR from raw in-phase and quadrature (I/Q) samples. The CNN extracts compact spatial representations, while the BiLSTM–GRU structure captures bidirectional temporal dependencies with reduced parameter complexity. Unlike existing hybrid models that rely on deep recurrent stacks or heavy dense layers, the proposed architecture achieves effective feature fusion with a compact parameter budget. Evaluations on the RML2016.10a and RML2016.10b datasets demonstrate that CBLGNet achieves 93.39% classification accuracy, outperforming several state-of-the-art AMR methods while maintaining low computational cost.
Rahman et al. (2026) studied this question.