A deep learning method for automatic modulation recognition in the time--frequency domain

Automatic modulation recognition (AMR) is essential in modern wireless communication systems to detect interference, analyze signals, and adaptively select modulation. However, conventional recognition systems face challenges in accuracy and complexity under low signal-to-noise ratio (SNR) conditions. To address this, we propose a novel approach that combines convolutional neural networks (CNNs) and long short-term memory (LSTM) models for accurate modulation classification. Our method effectively captures both short-term and long-term patterns in sequential data, utilizing time--frequency representations of collected spectrum data as input. To evaluate the performance of our proposed learning method, we evaluate our approach through extensive training and testing on a diverse dataset comprising eleven modulation types across various SNRs, demonstrating its robustness and high classification accuracy. The results obtained from our experiments validate the effectiveness of our combined CNN and LSTM model for modulation pattern classification. Our model achieves superior performance in capturing the complex temporal and spatial dependencies present in radio signals.