发表机构
Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg; School of Information and Communication Engineering, University of Electronic Science and Technology of China（卢森堡大学安全、可靠与信任跨学科中心（SnT）; 电子科技大学信息与通信工程学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对动态阻塞下车辆通感一体化性能下降问题，提出基于世界模型的主动式框架WM4ISAC，利用雷达热图学习潜在动态并预测事件，实现视距与非视距波束主动切换，显著提升平均速率和低分位速率。
AI 中文摘要
动态阻塞严重降低了车载通感一体化（ISAC）的性能，因为窄波束毫米波必须在视距（LoS）链路被阻塞或失准之前进行调整。本文提出WM4ISAC，一种基于世界模型的主动式ISAC框架，该模型从下行雷达热图中学习交通环境的可复用潜在动态模型，而非直接将观测映射到波束索引。所提框架使用视觉（V）模块压缩感知观测，动作条件记忆（M）模块预测潜在环境演化，以及轻量级事件头来评估候选波束的视距对齐、阻塞和散射辅助的非视距（NLoS）对齐。然后，码本级控制器利用这些事件预测，在不知道未来用户或阻塞状态的情况下，主动在视距和非视距波束之间切换。在动态阻塞下的仿真表明，WM4ISAC将平均速率较当前视距跟踪提高了28.9%，将5%分位速率从0.9 Mbps提升至30.5 Mbps，并达到完美信息上界的88.7%。
英文摘要
Dynamic blockage severely degrades vehicular integrated sensing and communication (ISAC), since narrow millimeter-wave beams must be adjusted before the line-of-sight (LoS) link is blocked or misaligned. This paper proposes WM4ISAC, a proactive ISAC framework built on a world model that learns a reusable latent dynamics model of the traffic environment from downlink radar heatmaps, rather than directly mapping observations to beam indices. The proposed framework uses a vision (V) module to compress sensing observations, an action-conditioned memory (M) module to predict latent environment evolution, and lightweight event heads to evaluate LoS alignment, blockage, and scatter-assisted non-line-of-sight (NLoS) alignment for candidate beams. A codebook-level controller then uses these event predictions to proactively switch between LoS and NLoS beams without knowing the future user or blocker states. Simulations under dynamic blockage show that WM4ISAC improves the mean rate by $28.9\%$ over current LoS tracking, raises the 5\%-tile rate from $0.9$ Mbps to $30.5$ Mbps, and reaches $88.7\%$ of the perfect-information upper bound.
CommentsAccepted to IEEE ISAC 2026
