cs.LGOct 7, 2026

UniCSI Towards a Universal Wi-Fi CSI Encoder for Ubiquitous Human Sensing

Authors: Daniel Eckhoff, Hua Kang, Zhitang Chen, Jie Chuai

Organizations: Huawei Noah’s Ark Lab

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Wi-Fi sensing promises to turn the everyday wireless signals that already surround us into ubiquitous sensors for human sensing. However, a fundamental obstacle is that CSI is acquired under diverse device-specific configurations, including different subcarrier counts, bandwidths, and carrier bands. Consequently, the resulting CSI tensors vary in both spectral resolution and tensor shape, making heterogeneous modeling challenging. Standard architectures struggle with such heterogeneity, forcing lossy pre-processing which compromises the underlying signal. To bridge this gap, we present UniCSI, a unified foundation architecture that directly operates on heterogeneous CSI while preserving the integrity of the native waveform. UniCSI hinges on two core innovations: (1) a physics-informed RF tokenizer that encodes each frequency channel based on its fractional position within the physical spectrum rather than rigid array indices. It preserves intrinsic spectral coherence and enables seamless, frequency resolution-agnostic processing across arbitrary sensing configurations. (2) a spectral aggregator that distills variable-length channel sequences into a fixed-size spectral signature, effectively decoupling the feature dimensionality from the physical subcarrier spacing. Extensive evaluations on a large-scale corpus of 25 heterogeneous public datasets, spanning 14 to 2048 subcarriers, 20 to 160 MHz bandwidth, and the 2.4 and 5 GHz bands, demonstrate that native heterogeneous ingestion substantially improves cross-domain transfer under both supervised and self-supervised training schemes, particularly in regimes where fixed-grid architectures fail to generalize.

Figures & tables

Figure 1: Overview of UniCSI. (Top) Per-link encoder: a temporal CNN and time-patch projection tokenize per-subcarrier CSI; learned queries with band-relative positional encoding cross-attend into the subcarrier tokens, compressing the variable-length spectrum into a fixed Q×P×dQ\times P\times d latent for a Transformer encoder. (Bottom) Training regimes: JEPA pretraining, supervised multi-task training, and downstream probing or fine-tuning.
DatasetTask ( nn )RxTxSub.Band (GHz)BW (MHz)
Wi-Fi Sensing Contest ( Han 2024 )Presence ( n=3n{=}3 )22248, 250580, 160
Person-in-Wi-Fi 3D ( Yan et al. 2024 )3D Pose ( n=4n{=}4 )3×33{\times}3130520
XRFV2 ( Lan et al. 2025 )HAR ( n=30n{=}30 )3×33{\times}3130520
Wi-MIR ( Islam et al. 2024 )HAR ( n=17n{=}17 )3330520
Behavior Auth. ( Shi et al. 2017 )User ID ( n=12n{=}12 )3130520
NTU-Fi Human ID ( Yang et al. 2022 )User ID ( n=14n{=}14 )31114540
Table 1: CSI datasets for pretraining and evaluation. Shaded rows form the self-supervised pretraining corpus; unshaded rows are held out for downstream evaluation only. nn is the number of class labels, Rx/Tx the antenna counts, Sub. the subcarrier count, Band the carrier band and BW the channel bandwidth; D×AD{\times}A denotes DD receiver devices of AA antennas each. Footnote marks are explained in the text.
 Multi-taskJEPAProbe / kk -NNFine-tune
EncoderscratchEMAfrozenfine-tuned
HeadlinearpredictorBN + lin.BN + lin.
ObjectiveCEsmooth- L1L_{1}CE / cosineCE
LR3×10−43{\times}10^{-4}1×10−41{\times}10^{-4}1×10−31{\times}10^{-3}3×10−43{\times}10^{-4}
WD5×10−25{\times}10^{-2}5×10−25{\times}10^{-2}1×10−51{\times}10^{-5}5×10−25{\times}10^{-2}
Budget50 epochs50 k steps200 epochs50 epochs
Table 2: Training hyperparameters. All runs use AdamW, cosine decay to 00 , gradient clip 1.01.0 , and bf16.
Figure 2: Fused F1 of a single encoder trained jointly across eight heterogeneous CSI datasets. Bars denote the mean and whiskers the standard deviation over three seeds. All datasets except NTU-Fi HAR (in-distribution control) use a cross-environment or cross-subject protocol.
Figure 3: Linear probe and kk -NN evaluation of Padded ViT and UniCSI on eight representative dataset splits. Bars report fused macro-F1; Avg is the mean over the selected splits.
DatasetSplitScratch (Ours)Full FT (PaddedViT)Full FT (Ours)
WiMANS (5 GHz)Cross-Env55.69%36.96%55.83%
XRF55Cross-Env39.53%15.5%39.78%
SHARPCross-Env58.50%75.71%71.06%
Exposing-CSICross-Env47.56%7.6%45.16%
FallDarRandom96.78%98.61%98.95%
CSI-Bench (2.4 GHz)Random80.19%91.16%90.67%
Table 3: Comparison between supervised training from scratch with UniCSI, full fine-tuning (FT) from the JEPA-pretrained Padded ViT, and full FT from the JEPA-pretrained UniCSI on unseen downstream datasets (Metric: Macro F1 score).
Figure 4: Few-shot downstream fine-tuning from pretrained versus random initialization, averaged over eight dataset–split pairs.

Explore similar work

CardsList

  1. The Universal Language of CSI:Unifying Wireless Sensing Across Devices and Environments

    Jun 30, 2026Jiayi Chen, Weiting Ou, Guangxu ZhuChannel State InformationWifi

    Shenzhen Research Institute of Big Data · The Chinese University of Hong Kong, Shenzhen · Shenzhen Loop Area Institute (SLAI)

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

    Dec 14, 2025Gaofeng Dong, Kang Yang, Mani SrivastavaIntegrated Sensing And CommunicationChannel State Information

    Department of Electrical and Computer Engineering, University of California, Los Angeles, CA, USA

  3. CSI-JEPA: Towards Foundation Representations for Ubiquitous Sensing with Minimal Supervision

    May 13, 2026Xuanhao Luo, Zhizhen Li, Yuchen LiuChannel State InformationSelf-Supervised Representations

    North Carolina State University, USA