Beyond Physical Constraints: AI-Powered Cross-Modal Interference Suppression in Wireless Multi-Modal Communication Systems

Cross-modal interference (CMI) presents major challenges in wireless multi-modal communication, from physical-layer conflicts to deep learning-induced modal inertia in Artificial Intelligence of Things(AIoT) devices. Traditional approaches focus on mitigating electromagnetic interference while overlooking multi-modal fusion effects. This paper proposes a deep learning framework integrating multi-modal semantic transformation and Neural Architecture Search (NAS) to suppress CMI at both physical and device levels. By standardizing diverse modal data into token representations, our approach ensures balanced feature extraction and mitigates modal dominance. NAS autonomously optimizes neural architectures for efficient high-dimensional feature learning. Simulations show significant reductions in bit error rate (BER), improved radar sensing robustness, and effective mitigation of modal inertia, paving the way for interference-resilient AI-driven wireless communication.