摘要
Abstract
As the sixth-generation mobile communication networks continues to evolve toward an integrated sensing and communication architecture,this leads to a deeper fusion between wireless sensing techniques and communication systems,which has gradually become a key enabler to support smart environments and daily intelligent applications.In real scenarios,the acquisition of channel state information is closely affected by hardware properties,and different devices as well as different collection conditions will produce data with inconsistent structures.This inconsistency has constrained the further development of integrated sensing and communication systems.To deal with this issue,a standardized sensing data protocol is proposed for integrated sensing and communication scenarios.The protocol applies deterministic processing to clean physical-layer signals,performs a normalized projection in the frequency domain,and constructs a unified tensor representation,so that wireless signals from diverse sources and structures into input forms that can be directly used by deep learning models.In this way,the dependence between machine learning tasks and hardware differences can be reduced to a certain extent.This work also builds a unified evaluation setting that covers tasks such as detection,identification,and vital sign estimation,and the experiments have been carried out on four heterogeneous datasets,including Widar3.0,GaitID,XRF55,and ElderAL-CSI.The results demonstrate that the proposed protocol effectively reduces the performance fluctuations caused by different random initialization conditions and decreases the performance variance across random seeds while maintaining a relatively high recognition accuracy.This change indicates that the proposed method can provide a more stable and comparable data processing basis for research in integrated sensing and communication.关键词
通感一体化/无线感知/信道状态信息/数据协议/评测基准平台Key words
integrated sensing and communication/wireless sensing/channel state information/data protocol/evaluation benchmark platform分类
信息技术与安全科学
