neurips2026citympc/CityMPC

CityMPC 是一个用于美国5个城市（德克萨斯州奥斯汀、纽约州纽约、德克萨斯州达拉斯、德克萨斯州沃斯堡、科罗拉多州丹佛）的大规模合成多径无线信道数据集。该数据集通过在DeepMIMO v4建筑几何体上使用Sionna RT射线追踪技术生成，工作频率为3.5 GHz。每个样本包含两部分：一是完整的多径信道描述——每条链路最多25条路径，包含复增益、延迟以及离开角和到达角；二是传播场景的环境感知POV渲染——包括RGB图像、深度图、表面法线、RF材质图（每通道的介电常数、电导率、粗糙度、散射）以及全局建筑高度图。该数据集旨在训练和评估环境感知的信道预测模型，以在推理时替代射线追踪计算。它附属于一篇NeurIPS 2026评估与数据集赛道的投稿。CityMPC is a large-scale synthetic dataset of multi-path wireless channels for 5 US cities (Austin TX, New York NY, Dallas TX, Fort Worth TX, Denver CO), generated via Sionna RT ray tracing on DeepMIMO v4 building geometries at 3.5 GHz. Each sample pairs: A full multi-path channel description — up to 25 paths per link with complex gains, delays, and angles of departure/arrival. An environment-aware POV rendering of the propagation scene — RGB image, depth map, surface normals, RF material maps (per-channel permittivity, conductivity, roughness, scattering), and global building height map. The dataset is intended for training and evaluating environment-aware channel-prediction models that replace ray tracing at inference time. It accompanies a NeurIPS 2026 Evaluations & Datasets Track submission.