基于多模态环境感知的车联网数字孪生信道多任务实现OA

The integration of ubiquitous sensing and artificial intelligence provides a novel paradigm for channel prediction.This paper proposes a multi-task realization framework for digital twin channel utilizing multi-modal environmental sensing,designed for dynamic vehicular network communication scenarios.Leveraging real-time image,point cloud,and location data collected at the vehicle side,the proposed framework constructs a multi-modal feature extraction-fusion network based on deep learning and cross-attention mechanisms,enabling fine-grained extraction and complementary fusion of environmental features for high-accuracy prediction of path loss and optimal beam index.Experimental results demonstrate that the proposed model effectively establishes an accurate mapping from environmental features to channel information.Specifically,under the multi-modal fusion framework,the prediction accuracy for path loss and beam index improves by 1.45 dB and 9.23%,respectively,over single-modal approaches,thereby validating the effectiveness of multi-modal fusion in enhancing channel prediction performance.