GenAI_Channel_Modeling_Datasets
收藏资源简介:
GenAI信道建模数据集是一个用于MIMO(多输入多输出)无线信道研究的射线追踪特定场景信道数据集,专为论文《Site-Specific MIMO Channel Generation via Diffusion and Flow Matching: Fidelity, Efficiency, and Downstream Utility》设计。数据集包含3个.npz文件,分别对应不同的通信频率(3.5 GHz和28 GHz)和传播场景(仅视距LoS和仅非视距NLoS)。每个文件包含一个名为combined_array的NumPy数组,其形状为(N, N_rx, 1, N_tx, 1, 1, 4),数据类型为复数(complex64)。数组的最后一个维度包含4个元素:复数值的信道系数H,以及用户设备(UE)的三维坐标(x, y, z,单位:米),因此数据集同时提供了信道矩阵和对应的UE地理位置信息。数据是使用GPU加速的无线信道仿真射线追踪引擎NVIDIA Sionna RT,在单个室外城市场景中生成的。该数据集适用于MIMO信道建模、生成、AI驱动的无线通信研究以及信道仿真等任务。
The GenAI Channel Modeling Dataset is a ray-tracing-based site-specific channel dataset for MIMO (Multiple-Input Multiple-Output) wireless channel research, specifically designed for the paper *Site-Specific MIMO Channel Generation via Diffusion and Flow Matching: Fidelity, Efficiency, and Downstream Utility*. The dataset contains three .npz files, corresponding to different communication frequencies (3.5 GHz and 28 GHz) and propagation scenarios (Line-of-Sight only (LoS) and Non-Line-of-Sight only (NLoS)) respectively. Each file contains a NumPy array named `combined_array` with a shape of (N, N_rx, 1, N_tx, 1, 1, 4) and a data type of complex64. The last dimension of the array consists of 4 elements: the complex-valued channel coefficient H, and the three-dimensional coordinates (x, y, z, unit: meter) of the User Equipment (UE). Thus, the dataset provides both the channel matrix and the corresponding UE geographic location information. The data was generated in a single outdoor urban scenario using the GPU-accelerated ray-tracing engine for wireless channel simulation, NVIDIA Sionna RT. This dataset is applicable to tasks such as MIMO channel modeling, channel generation, AI-driven wireless communication research, and channel simulation.
数据集概述
该数据集用于论文《Site-Specific MIMO Channel Generation via Diffusion and Flow Matching: Fidelity, Efficiency, and Downstream Utility》,提供了通过光线追踪生成的特定场景MIMO信道数据。
文件与频率
- 3.5 GHz
Final_Single_Scene_Channel_Sionna_V1_3_5GHz_LoS.npz(仅视距传播)Final_Single_Scene_Channel_Sionna_V1_3_5GHz_NLoS.npz(仅非视距传播)
- 28 GHz
Final_Single_Scene_Channel_Sionna_V1_28GHz_LoS.npz(仅视距传播)
数据格式
每个 .npz 文件包含一个键为 combined_array 的数组:
- 形状:
(N, N_rx, 1, N_tx, 1, 1, 4) - 数据类型:
complex64 - 最后一维含义:
[0]: 复数信道系数 H[1]: UE x 坐标(米)[2]: UE y 坐标(米)[3]: UE z 坐标(米)
提取示例: python import numpy as np npz = np.load("Final_Single_Scene_Channel_Sionna_V1_3_5GHz_LoS.npz") data = npz["combined_array"][:, :, 0, :, 0, 0, :] # (N, N_rx, N_tx, 4) H = data[:, :, :, 0] # 复数信道矩阵 (N, N_rx, N_tx) coords = data[:, 0, 0, 1:] # UE (x, y, z) 位置 (N, 3)
生成方式
数据使用 NVIDIA Sionna RT(GPU加速光线追踪引擎)在单个室外城市场景中生成,生成脚本及说明位于代码仓库。
下载方式
bash git clone https://huggingface.co/datasets/PaulAlm/GenAI_Channel_Modeling_Datasets
也可通过 Hugging Face 网页界面手动下载单个文件。
相关资源
- 代码仓库: GenAI_Channel_Modeling
- 预训练模型: GenAI_Channel_Modeling_Models
许可
MIT 许可证




