Deep Quantum-Transformer Networks for Multi-Modal Beam Prediction in ISAC Systems

In this paper, we propose hybrid deep quantumtransformer networks (QTNs) to predict the optimal beam in integrated sensing and communication (ISAC) systems employing millimeter-wave (mmWave) band. In mobile applications, vehicle-to-infrastructure (V2I) communications at high frequency require large antenna arrays and narrow beams, which is associated with high beam training overhead. In such a scenario, selecting an optimal beam to maximize the signal power at the receiver can be learned from the sensory data collected at the base station and guided by the position-based data provided by the user equipment. Such multi-modal sensory data can be utilized by deep learning frameworks to create situational awareness for intelligently predicting optimal beams. We evaluate the proposed learning models in real-world V2I scenarios provided by the multi-modal DeepSense 6G dataset and compare them with the existing works. The experimental results show a distance-based accuracy (DBA) score of 0.9124 for multi-modal and 0.8832 for position-based data, respectively. Moreover, the hybrid QTNs achieve the best DBA scores and the highest accuracy compared to other models on zero-shot testing. These QTN models exhibit low complexity and high performance, demonstrating their potential to address the challenges of beam management in mmWave ISAC systems.