CSI-Based User Positioning, Channel Charting, and Device Classification with an NVIDIA 5G Testbed

Channel-state information (CSI)-based sensing will play a key role in future cellular systems. However, no CSI dataset has been published from a real-world 5G NR system that facilitates the development and validation of suitable sensing algorithms. To close this gap, we publish three real-world wideband multi-antenna multi-open RAN radio unit (O-RU) CSI datasets from the 5G NR uplink channel: an indoor lab/office room dataset, an outdoor campus courtyard dataset, and a device classification dataset with six commercial-off-the-shelf (COTS) user equipments (UEs). These datasets have been recorded using a software-defined 5G NR testbed based on NVIDIA Aerial RAN CoLab Over-the-Air (ARC-OTA) with COTS hardware, which we have deployed at ETH Zurich. We demonstrate the utility of these datasets for three CSI-based sensing tasks: neural UE positioning, channel charting in real-world coordinates, and closed-set device classification. For all these tasks, our results show high accuracy: neural UE positioning achieves 0.6cm (indoor) and 5.7cm (outdoor) mean absolute error, channel charting in real-world coordinates achieves 73cm mean absolute error (outdoor), and device classification achieves 99% (same day) and 95% (next day) accuracy. The CSI datasets, ground-truth UE position labels, CSI features, and simulation code are publicly available at https://caez.ethz.ch

💡 Research Summary

This paper presents a significant contribution to the field of integrated sensing and communication (ISAC) for future cellular networks by introducing the first publicly available dataset of Channel-State Information (CSI) collected from a real-world, standards-compliant 5G New Radio (NR) system. Recognizing the critical need for realistic data to develop and validate sensing algorithms for 6G, the authors from ETH Zurich deployed a software-defined 5G testbed based on the NVIDIA Aerial RAN CoLab Over-the-Air (ARC-OTA) platform. The testbed operates in the licensed Swiss private 5G band (3.45 GHz center frequency, 100 MHz bandwidth) and utilizes four commercial off-the-shelf (COTS) Open RAN Radio Units (O-RUs) with four antennas each, alongside various COTS User Equipments (UEs). A WorldViz Precision Position Tracking (PPT) system with infrared cameras provides centimeter-accurate ground-truth UE positions synchronized with the CSI measurements.

The collected dataset, termed CSI Acquisition at ETH Zurich (CAEZ), comprises three distinct subsets: CAEZ-5G-INDOOR, recorded in a 3.5m x 3.5m lab/office area using a robot; CAEZ-5G-OUTDOOR, recorded in a 10m x 10m campus courtyard; and CAEZ-5G-DEV-CLASS, which contains CSI from six different COTS smartphones measured during both rotation at a fixed point and random human movement, facilitating device fingerprinting studies.

To demonstrate the utility of these datasets, the paper conducts comprehensive case studies on three key CSI-based sensing tasks. For neural UE positioning, a supervised method using a fully-connected neural network with a probability map output is employed. The network is trained on downsampled absolute values of OFDM-domain CSI. This approach achieves remarkable mean absolute errors (MAE) of 0.6 cm indoors and 5.7 cm outdoors, showcasing the potential for extreme precision.

For channel charting, a self-supervised pseudo-positioning technique, the authors apply a method that maps CSI features to real-world coordinates. This approach, which requires no position labels during training, achieves an MAE of 73 cm in the outdoor environment, proving the feasibility of reconstructing relative UE locations directly from channel measurements.

Finally, for closed-set device classification, the study extracts Radio Frequency Fingerprint Identification (RFFI) features from the CSI and uses a standard classification pipeline. The model achieves 99% accuracy when tested on data from the same day as the training set and maintains a high 95% accuracy when tested on data collected in a slightly modified environment on the following day, highlighting robustness to minor temporal and environmental changes.

In conclusion, this work bridges a major gap in 6G sensing research by providing a crucial real-world 5G NR dataset and benchmark results. The public release of the CSI data, ground-truth labels, extracted features, and simulation code establishes a vital foundation for the development, comparison, and validation of advanced sensing algorithms, thereby accelerating progress towards the realization of integrated sensing in next-generation wireless systems.