Samsung Research Demonstrates ISAC on Commercial 5G vRAN Using Uplink CSI for Human Sensing

Samsung Research implemented Integrated Sensing and Communication (ISAC) on a commercial 5G vRAN network, detecting human presence and entry/exit at building entrances using only existing uplink reference signals and no additional radio hardware.
The system achieves real-time inference with a 200ms update rate, distinguishing four states (enter, exit, presence, clear) with consistent accuracy across two test sites, including one with an automatic revolving door that generates its own Doppler signature.
A practical preprocessing chain for bistatic Channel State Information (CSI) was developed, comprising sidelobe suppression, timing jitter compensation, and clutter removal to convert raw, unusable CSI into clean Range-Doppler Maps (RDMap).
Data volume was reduced by more than one order of magnitude (10x) by selecting a subset of subcarriers and antennas, allowing the sensing load to be offloaded to an external GPU server without competing with RAN communication processing.
The architecture utilizes a Distributed Unit (DU) to pass reduced CSI to an external GPU server for preprocessing, AI inference, and post-processing, enabling software-only updates to add sensing capabilities to existing commercial vRAN deployments.
Future work includes on-site learning for adaptation to new static structures, sensing-aware scheduling for multiple UEs, and privacy measures that report only clear states without identifying information.

Key point