pku-pcni-lab/RF-only_channel_dataset_for_LLM4WM

Wireless Multi-Tasking Dataset

📌 Overview

This dataset is designed for multi-task learning at the wireless physical layer. It is generated using QuaDRiGa in accordance with the 3GPP standard and models a dual-band communication scenario operating at 1.9 GHz and 28 GHz. The dataset is organized around six channel-related tasks, including channel estimation, time-domain channel prediction, frequency-domain channel prediction, sub-6G-assisted mmWave beam management, distance estimation, and path-loss estimation. It can be used to investigate cross-task shared representation learning and transfer generalization schemes for wireless physical-layer applications, and is well suited for evaluating the multi-task joint modeling capability of learning-based methods in complex wireless environments.

Specifically, the dataset contains a total of 20,000 samples, which are split into 15,000 training samples, 1,600 validation samples, and 3,400 test samples. Each sample includes complete CSI matrices at both frequency bands, along with user location and path-loss information. The user initial positions are randomly generated, and the user follows a linear trajectory at a speed of 30 km/h. In the sub-6G band, the base station is equipped with a 16×1 ULA, with 64 subcarriers and a bandwidth of 60 MHz. In the mmWave band, the base station employs a 64×1 ULA, also with 64 subcarriers, while the bandwidth is 500 MHz. On the user side, a single omnidirectional antenna is deployed in both frequency bands.

---

📜 Related paper

X. Liu, S. Gao, B. Liu, X. Cheng, L. Yang. “LLM4WM: Adapting LLM for Wireless Multi-Tasking. ” IEEE Transactions on Machine Learning in Communications and Networking, July 2025, 3(51): 835-847. https://arxiv.org/abs/2501.12983