Standard-Compliant Wi-Fi Sensing with Beamforming Feedback

Practical, domain-adaptive, and multi-user sensing with IEEE 802.11 beamforming feedback

Wi-Fi sensing can enable applications such as remote healthcare, smart-home monitoring, and human-computer interaction. Most existing systems, however, rely on channel state information (CSI) extracted through specialized hardware or modified firmware—capabilities that are not exposed by standard commercial Wi-Fi devices.

This research program develops a practical alternative based on beamforming feedback information (BFI). BFI is a compressed representation of the wireless channel that is routinely transmitted during IEEE 802.11ac/ax MIMO channel sounding. Because these feedback frames can be captured over the air without modifying the sensing devices, BFI enables standard-compliant sensing with commercial Wi-Fi equipment. It also allows a single monitor to observe the channels between an access point and multiple stations simultaneously.

Unlike conventional CSI extraction, beamforming feedback can be captured from standard-compliant Wi-Fi transmissions and can expose multiple user channels to a single passive monitor.

Research contributions

Wi-BFI: an open-source extraction tool

Wi-BFI is the first open-source tool for extracting beamforming feedback angles (BFAs) and reconstructing BFI from captured Wi-Fi frames. It supports IEEE 802.11ac and 802.11ax, 20–160 MHz channels, single-user and multi-user MIMO, and both real-time and offline analysis.

Understanding compressed MIMO feedback

Our feedback quantization and grouping study evaluates how BFA quantization and OFDM subchannel grouping affect communication performance across commercial devices, propagation environments, and network configurations. The accompanying code and datasets provide a benchmark for designing efficient feedback mechanisms.

BFA-Sense: sensing directly from feedback angles

BFA-Sense demonstrates that standard-compliant BFAs can support human activity recognition without firmware modifications. Across three subjects, twenty activities, and three environments, BFA-based sensing achieved approximately 11% higher accuracy than the evaluated CSI-based approach. Code and data are publicly available.

BeamSense: adapting to unseen domains

BeamSense extends BFA sensing with a cross-domain few-shot learning strategy for unseen environments and subjects. It improved accuracy by up to 30% over the evaluated domain-adaptation baselines and achieved over 98% accuracy in a gesture-recognition case study. The implementation and datasets are open source.

Left: Wi-BFI extracts feedback from devices using different Wi-Fi configurations. Right: BeamSense uses feedback collected from MU-MIMO users for domain-adaptive activity recognition.

Si-Fi: simultaneous multi-subject sensing

Si-Fi uses the multi-user nature of BFI to sense several subjects concurrently. In experiments with three subjects performing twenty activities across three environments, Si-Fi achieved up to 99% classification accuracy. Its few-shot adaptation improved accuracy by up to 27%, while the system reduced latency by 50% and channel occupation by 110 KB per sample per sensing device compared with the evaluated simultaneous multi-subject sensing approach. The code is available online.

Open CSI and BFI datasets

To support reproducible research, we released CSI-BFI-HAR, two human-activity-recognition datasets collected with commercial IEEE 802.11ac devices under line-of-sight and non-line-of-sight conditions. Together, they contain approximately 240 GB of CSI and 230 GB of BFI. The data cover twenty activities, six subjects, six experimental setups, and simultaneous multi-subject sensing.

BeamSense evaluation across sensing representations, spatial diversity, unseen environments, and unseen subjects.

Publications

  1. K. F. Haque, F. Meneghello, and F. Restuccia, “Datasets for Human Activity Recognition with Integrated Sensing and Communications in Wi-Fi”, IEEE Communications Magazine, 2026. Code and datasets.
  2. K. F. Haque, M. Zhang, F. Meneghello, and F. Restuccia, “Si-Fi: Learning the Beamforming Feedback for Simultaneous Multi-Subject Sensing”, Computer Networks, 2025. Code.
  3. K. F. Haque, M. Zhang, F. Meneghello, and F. Restuccia, “BeamSense: Rethinking Wireless Sensing with MU-MIMO Wi-Fi Beamforming Feedback”, Computer Networks, 2025. Code and datasets.
  4. F. Meneghello, K. F. Haque, and F. Restuccia, “Evaluating the Impact of Channel Feedback Quantization and Grouping in IEEE 802.11 MIMO Wi-Fi Networks”, IEEE Wireless Communications Letters, 2024. Code and datasets.
  5. K. F. Haque, F. Meneghello, and F. Restuccia, “BFA-Sense: Learning Beamforming Feedback Angles for Wi-Fi Sensing”, IEEE PerCom Workshops, 2024. Code and data.
  6. K. F. Haque, F. Meneghello, and F. Restuccia, “Wi-BFI: Extracting the IEEE 802.11 Beamforming Feedback Information from Commercial Wi-Fi Devices”, ACM WiNTECH, 2023. Code.