Deep learning and frequency-domain feature fusion for wireless communication signal modulation recognition and interference detection
Abstract
Automatic modulation recognition (AMR) and interference detection are critical tasks in modern wireless communication systems. Existing approaches often rely on either time-domain or frequency-domain features alone, failing to capture the complementary information across both domains. This paper proposes a Frequency-Domain Feature Fusion Network (FDFNet) that jointly exploits time-domain and frequency-domain representations for simultaneous modulation classification and interference detection. The architecture employs a dual-branch design: a one-dimensional convolutional neural network with bidirectional long short-term memory (BiLSTM) processes raw in-phase and quadrature (IQ) samples, while a two-dimensional CNN extracts features from short-time Fourier transform (STFT) spectrograms. An attention-based fusion module adaptively combines both feature streams before feeding them into two task-specific classification heads. Experiments on the RadioML 2016.10a dataset and a synthetic interference dataset demonstrate that FDFNet achieves 93.2% overall modulation recognition accuracy and 92.4% interference detection accuracy, outperforming several baseline methods including standalone CNN, LSTM, and ResNet architectures. The results confirm that frequency-domain feature fusion significantly enhances recognition performance, particularly under low signal-to-noise ratio conditions.