pku-pcni-lab/Multi-modal_dataset_for_FM4SoM

Dataset for "Foundation Model Empowered Synesthesia of Machines (SoM): AI-Native Intelligent Multi-Modal Sensing-Communication Integration"

📌 Overview

The dataset is primarily built upon the SynthSoM dataset and consists of five subsets, designed respectively for validating LLM empowered SoM mechanism exploration (Case Study 1.1), LLM empowered SoM-enhanced transceiver design (Case Study 1.2), wireless foundation model empowered SoM mechanism exploration (Case Study 2.1), wireless foundation model empowered SoM-enhanced transceiver Design (Case Study 2.2), and wireless foundation model empowered SoM-enhanced cooperative perception (Case Study 2.3).

---

📂 Dataset Structure

Case Study 1.1 includes multimodal sensing data and wireless channel data collected in an urban intersection scenario. It is divided into two parts. In the first part, the multimodal sensing data consist of RGB images and depth maps collected by a UAV flying at an altitude of 70 m. The communication channel data are generated with the UAV equipped with a single antenna as the transmitter and a 50×50 ground antenna grid as the receiver, forming a total of 2,500 UAV-to-ground communication links. These data constitute a path loss map at a carrier frequency of 28 GHz. In the second part, the sensing data consist of LiDAR point clouds collected by vehicles. The communication channel data involve vehicle-mounted single antennas acting as both transmitters and receivers, including a total of 27,000 sets of inter-vehicle channel propagation scatterer data at carrier frequencies of 28 GHz and sub-6 GHz.

Case Study 1.2 contains 1500 groups of temporally and spatially aligned base-station RGB images, CSI matrices, and vehicle location information in a MISO-OFDM system. The carrier frequency is 28 GHz. The base station is equipped with a 64 × 1 ULA, with 64 subcarriers and 10 MHz bandwidth, while the user side uses a single antenna.

Case Study 2.1 includes sensing data and wireless channel data in an urban intersection scenario. The sensing data consist of UAV-captured RGB images at an altitude of 70 m. The communication channel data are generated with the UAV equipped with a single antenna as the transmitter and a 50×50 ground antenna grid as the receiver, forming 2,500 UAV-to-ground communication links. These data include a path loss map composed of path loss values for all 2,500 links, as well as the angles of departure (AoD) for the same set of links, all at a carrier frequency of 28 GHz.

Case Study 2.2 contains 1500 groups of temporally and spatially aligned base-station RGB images and CSI matrices in a MISO-OFDM system. The carrier frequency is 28 GHz. The base station is equipped with a 64 × 1 ULA, with 64 subcarriers and 10 MHz bandwidth, while the user side uses a single antenna.

Case Study 2.3 contains 1500 groups of temporally and spatially aligned base-station RGB images and CSI matrices in a MISO-OFDM system. The carrier frequency is 28 GHz. The base station is equipped with a 64 × 1 ULA, with 64 subcarriers and 10 MHz bandwidth, while the user side uses a single antenna

---

📜 Related paper

X. Cheng, B. Liu, X. Liu, E. Liu, and Z. Huang, “Foundation model empowered Synesthesia of Machines (SoM): AI-native intelligent multi-modal sensing-communication integration”, IEEE Transactions on Network Science and Engineering,Jul. 2025. https://arxiv.org/abs/2506.07647