AMC-CNN-BiLSTM: An Advanced Hybrid Model for Automatic Modulation Classification of Radio Signals
Ha-Khanh Le, Van‐Phuc Hoang, Van‐Sang Doan · 2024
This paper introduces a novel deep neural network, AMC-CNN-BiLSTM, designed by combining convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architectures for automatic modulation classification (AMC) of radio signals. The AMC-CNN-BiLSTM model features a multi-branch CNN structure to capture a broad range of spatial features, while BiLSTM layers effectively capture temporal dependencies in the signal data. By integrating both spatial and temporal feature extraction, this hybrid model enables accurate classification of various modulation schemes. The model is evaluated using the HisarMod2019.1 dataset, which contains 26 modulation types subjected to different channel conditions and additive white Gaussian noise (AWGN). Experimental results show that AMC-CNN-BiLSTM achieves an average classification accuracy of 79%, demonstrating its ability to improve AMC performance in challenging and noise communication environments. In comparison, the proposed model outperforms other models in terms of accuracy under the same experimental conditions.