Multi-Modal Fusion of Spatial and Temporal Features for Radar Signal Sensing

Communication and radar signal sensing are pivotal for modern wireless and intelligent sensing systems. Traditional techniques, including feature-engineered methods and standalone neural networks, often struggle to handle multi-modal data effectively and perform poorly in low signal-to-noise ratio (SNR) environments. These approaches fail to capture spatial and temporal dependencies and lack mechanisms for leveraging inter-modal relationships. A CNN-LSTM-Based Attention-Augmented Multi-Feature Fusion Network has been proposed to overcome these challenges. This architecture utilizes convolutional neural networks (CNNs) for spatial feature extraction, long short-term memory (LSTM) networks for modelling temporal dependencies, and an attention mechanism to refine features across modalities dynamically. The model ensures robust signal sensing under challenging conditions by integrating IQ data, spectrograms, and cyclic spectrum representations. Extensive experiments validate the proposed method's superiority, achieving over 99.7% sensing accuracy over Gaussian and racian channel. The network demonstrates strong resilience in SNRs as low as -5 dB, outperforming traditional methods. Performance metrics and experimental results confirm its effectiveness and emphasize its potential as a robust framework for advancing communication and radar signal sensing.