UniFi: Combining Irregularly Sampled CSI from Diverse Communication Packets and Frequency Bands for Wi-Fi Sensing

TL;DR

UniFi tackles the problem of Wi‑Fi sensing without compromising communication throughput by exploiting irregularly sampled CSI from regular ambient traffic across multiple frames and bands. It introduces a CSI sanitization pipeline to harmonize heterogeneous, bursty inputs and a time-aware attention network that learns directly from irregular CSI without resampling. The work provides CommCSI-HAR, the first HAR dataset with irregular CSI from real dual-band traffic, and demonstrates state-of-the-art performance across five sensing tasks with a compact model, preserving throughput. The results highlight the practicality and robustness of injection-free ISAC on commodity Wi‑Fi devices, leveraging multi-band fusion and diverse frame types to improve sensing fidelity. The approach offers a scalable pathway toward real-world ISAC deployment without hardware changes.

Abstract

Existing Wi-Fi sensing systems rely on injecting high-rate probing packets to extract channel state information (CSI), leading to communication degradation and poor deployability. Although Integrated Sensing and Communication (ISAC) is a promising direction, existing solutions still rely on auxiliary packet injection because they exploit only CSI from data frames. We present UniFi, the first Wi-Fi-based ISAC framework that fully eliminates intrusive packet injection by directly exploiting irregularly sampled CSI from diverse communication packets across multiple frequency bands. UniFi integrates a CSI sanitization pipeline to harmonize heterogeneous packets and remove burst-induced redundancy, together with a time-aware attention model that learns directly from non-uniform CSI sequences without resampling. We further introduce CommCSI-HAR, the first dataset with irregularly sampled CSI from real-world dual-band communication traffic. Extensive evaluations on this dataset and four public benchmarks show that UniFi achieves state-of-the-art accuracy with a compact model size, while fully preserving communication throughput.

Figures (10)

• Figure 1: Application scenario of UniFi. It utilizes irregularly sampled CSI from diverse communication packets across multiple frequency bands for sensing, operating seamlessly with existing Wi-Fi systems.
• Figure 2: CSI comparison: traditional Wi-Fi sensing frameworks vs. UniFi. (a) Traditional frameworks inject controlled packets for uniform CSI with constant intervals, bandwidths, and amplitudes. (b) Our UniFi framework leverages CSI from multi-band communication packets with irregular intervals, bursty timestamps, dynamic amplitudes, variable bandwidths, and different waveforms.
• Figure 3: Architecture of UniFi. Taking the heterogeneous and irregularly sampled CSI (CommCSI-HAR dataset) as input, UniFi applies a CSI Sanitization pipeline to align waveforms, remove bursts, and select subcarriers. The processed data is then fed into a time-aware attention network to generate an output for the sensing tasks.
• Figure 4: Data collection environment ($\sim$4m$\times$4m).
• Figure 5: Model capacity across diverse tasks: accuracy and size of UniFi-DNN and the baselines. For the first four tasks, the CSI-BERT2 accuracies shown are the best results reported in zhao2024mining. CommCSI-HAR$^{*}$ denotes the baseline-compatible subset of the CommCSI-HAR dataset, consisting of 52 subcarriers extracted from QoS data frames in the 5GHz band to match CSI-BERT2's original design with minimal modification to baselines' architectures. The arrow highlights the additional 9.8% accuracy gain obtained when using the full CommCSI-HAR dataset.