TLDR: Researchers have developed a new multimodal dataset combining 3D LiDAR scans and Wi-Fi RSSI measurements to improve indoor radio mapping. This dataset, collected in a multi-room environment with and without human presence, helps understand signal propagation dynamics and is crucial for developing advanced AI models to optimize wireless networks for demanding applications like XR, especially with new Wi-Fi standards.
The increasing demand for reliable wireless connectivity in indoor environments, driven by smart devices and bandwidth-intensive applications like real-time video analytics and Extended Reality (XR), highlights the need for accurate Radio Environment Maps (REMs). These maps are crucial for optimizing wireless network planning and Access Point (AP) placement. However, creating realistic REMs is challenging due to the complexity of indoor spaces.
To address this, a new multimodal dataset has been introduced. This dataset combines high-resolution 3D LiDAR scans with Wi-Fi Received Signal Strength Indicator (RSSI) measurements. These measurements were collected under 20 different AP configurations in a multi-room indoor environment. A unique aspect of this dataset is that it captures two distinct measurement scenarios: one without human presence and another with human presence, allowing for the study of how dynamic environmental factors affect wireless signal propagation.
This resource is designed to support research in data-driven wireless modeling, especially for new high-frequency standards like IEEE 802.11be (Wi-Fi 7). The ultimate goal is to advance the development of robust, high-capacity indoor communication systems.
Modern smart devices, from cameras to voice assistants, rely heavily on continuous internet connections. As these devices integrate advanced features like high-definition video streaming and real-time AI processing, their bandwidth requirements are set to soar. Wi-Fi is well-suited for indoor environments due to its high data rates and cost-efficiency. However, optimizing Wi-Fi performance in complex indoor settings is difficult due to signal attenuation from walls, interference, and physical obstructions. This challenge is amplified by XR technologies (Virtual Reality, Mixed Reality, Augmented Reality) which demand ultra-low latency and extremely high throughput, often exceeding current Wi-Fi capabilities, especially with suboptimally placed APs. Newer standards like Wi-Fi 7 and the upcoming Wi-Fi 8, using higher frequencies, face reduced wall penetration and limited range, making the problem even more pressing.
A comprehensive dataset documenting Wi-Fi propagation characteristics in indoor environments, such as homes or large offices, is therefore critical. Such datasets enable the creation of detailed indoor REMs, support predictive channel models, and facilitate AI-driven network optimization tools for strategic AP placement. This is particularly vital for designing XR-ready environments where seamless experiences are paramount.
Traditional REM techniques include direct methods (interpolating signal measurements) and indirect methods (based on transmission parameters), such as ray-tracing models found in tools like Wireless InSite and NVIDIA’s Sionna. These are useful for generating synthetic datasets for training supervised learning models. Machine Learning (ML) approaches, including Support Vector Machines (SVMs), XGBoost, and more recently deep learning techniques like Generative AI (GAI), Large Language Models (LLMs), and Graph Neural Networks (GNNs), have shown promise in learning complex propagation patterns. GAI, in particular, can combine heterogeneous data like geometric information and wireless signal features, which is key for realistic REM generation. Despite these advancements, the lack of publicly available, high-resolution datasets has been a major hurdle.
This new dataset aims to bridge this gap by combining Received Signal Strength Indicator (RSSI) measurements with detailed 3D point cloud data. This multimodal representation allows learning-based models to account for occlusions, materials, and structural features that impact signal behavior, leading to more accurate and robust REM estimation systems.
The dataset was collected during a two-day measurement campaign to capture the effects of human presence. Measurements were taken in an unoccupied laboratory and later with 7-10 individuals engaged in typical activities. The final dataset includes 16 unique AP placement scenarios, capturing signal propagation dynamics with and without human presence.
The data acquisition involved LiDAR sensors (Velodyne VLP-16 and Livox Avia) for 3D point clouds and a commercial Android smartphone with the Wi-Fi Analyzer app for RSSI measurements. The AP used was a TP-Link TL-WR841N. RSSI data was collected at 53 predefined grid locations, with the UE device held at 1.5 meters and the AP at 1.2 meters. The environment included an office space and an adjacent corridor.
Also Read:
- Robust UWB Jammer Tracking in Dynamic Indoor Environments
- Enhancing Wireless Communication: A Unified Approach to Combat Adversarial Attacks and Environmental Shifts in Modulation Classification
Measurement Scenarios
Two main scenarios were explored:
1. Empty Office (12 Setups): Providing a baseline with the office unoccupied, placing the AP at various locations to capture diverse propagation paths.
2. Office with Regular Activity (8 Setups): Measurements taken with 7-10 people present. This scenario included reusing four AP positions from Scenario 1 for direct comparison and introducing four new AP positions to broaden the dataset.
The data is available on Zenodo. The 3D point cloud data is provided in separate files for the office, corridor, and a combined .ply file. The RSSI data is available in .csv and .h5 formats, including raw measurements and structured matrix data with labels for setup number, AP location, and measurement flow. A lightweight toolbox is also provided for basic point cloud operations and Wi-Fi data processing, including visualization and handling missing values.
Technical validation showed that human presence significantly reduces signal strength, as human bodies absorb and scatter radio signals. Interestingly, RSSI values directly at the AP’s location can sometimes be weaker than adjacent positions, attributed to the antenna’s radiation null.
This dataset is intended to facilitate the development of advanced ML models, particularly those based on Generative AI, for accurate REM estimation using 3D environmental data. This is crucial for applications demanding high throughput and ultra-low latency, such as XR devices. Researchers can use subsets of the data (office only, corridor only) or the full dataset for more complex studies.
The code supporting this study, including scripts for loading, visualizing, and processing the data, is publicly available on Zenodo. For more details, you can refer to the research paper: A Multimodal Dataset for Indoor Radio Mapping with 3D Point Clouds and RSSI.
