Protecting Human Activity Signatures in Compressed IEEE 802.11 CSI Feedback

Andrea J. Goldsmith; Atsutse Kludze; Doru Calin; H. Vincent Poor; Mohamed Seif; Yasaman Ghasempour

arxiv: 2512.18529 · v2 · submitted 2025-12-20 · 💻 cs.IT · cs.CR· cs.NI· math.IT

Protecting Human Activity Signatures in Compressed IEEE 802.11 CSI Feedback

Mohamed Seif , Atsutse Kludze , Yasaman Ghasempour , H. Vincent Poor , Doru Calin , Andrea J. Goldsmith This is my paper

Pith reviewed 2026-05-16 20:58 UTC · model grok-4.3

classification 💻 cs.IT cs.CRcs.NImath.IT

keywords differentially private quantizationIEEE 802.11 CSI feedbackGivens rotation anglestransmit beamformingprivacy-preserving feedbackactivity signature protectionstochastic quantizer

0 comments

The pith

A standards-compatible differentially private quantizer replaces deterministic angular quantization on the Givens parameters of the 802.11 transmit beamforming matrix to protect activity signatures.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows how to add differential privacy directly to the compressed CSI feedback that WiFi devices already send to report beamforming directions. It replaces the usual fixed quantization of the Givens rotation and phase angles with a randomized version that satisfies ε-differential privacy while keeping the exact message format required by the standard. Because those angles carry fine spatial details of the environment, the change prevents passive eavesdroppers from reliably inferring user movements, identity, or location. Closed-form sensitivity bounds for the angular parameters allow the privacy parameter ε to be set in a principled way. Simulations confirm that beamforming performance stays nearly unchanged even at strong privacy levels.

Core claim

We introduce a standards-compatible differentially private (DP) quantization mechanism that replaces deterministic angular quantization with an ε-DP stochastic quantizer applied directly to the Givens parameters of the transmit beamforming matrix. The mechanism preserves the 802.11 feedback structure, admits closed-form sensitivity bounds for the angular representation, and enables principled privacy calibration. Numerical simulations demonstrate strong privacy guarantees with minimal degradation in beamforming performance.

What carries the argument

The ε-DP stochastic quantizer applied directly to the Givens rotation and phase angles that parametrize the right-singular subspace of the channel matrix.

If this is right

The modified feedback packets remain fully decodable by any standard-compliant 802.11 receiver.
Privacy level ε can be calibrated using the derived sensitivity bounds without ad-hoc tuning.
Beamforming gain loss remains small enough that link performance stays acceptable in typical indoor scenarios.
The same structure blocks inference of activity, identity, and location from plaintext CSI.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same stochastic angle perturbation could be applied to other wireless standards that report quantized rotation angles for precoding.
In dense deployments the mechanism would reduce the effectiveness of passive sensing attacks that rely on CSI side channels.
Adaptive choice of ε based on measured channel coherence time could trade privacy against overhead on a per-packet basis.

Load-bearing premise

The stochastic quantizer can be made to preserve the exact 802.11 feedback structure while still admitting closed-form sensitivity bounds for the angular parameters.

What would settle it

A side-by-side comparison of real 802.11 hardware traces showing whether an eavesdropper can still classify human activity from the modified feedback packets at the same accuracy as with deterministic quantization.

Figures

Figures reproduced from arXiv: 2512.18529 by Andrea J. Goldsmith, Atsutse Kludze, Doru Calin, H. Vincent Poor, Mohamed Seif, Yasaman Ghasempour.

**Figure 2.** Figure 2: Closed-loop CSI feedback architecture with Nt transmit antennas, Nr receive antennas, Ns spatial streams, a shared angles codebook, and the proposed DP–SQ feedback quantization. and the privacy threat model. In Section III, we propose our privacy-preserving stochastic quantization mechanism. Section IV shows numerical results to confirm our findings. Finally, Section V concludes the paper and discuss futur… view at source ↗

**Figure 3.** Figure 3: An example for CSI amplitude spectrogram illustrating activity-dependent channel dynamics for a user moving within an indoor environment. Distinct Doppler patterns correspond to transitions between standing/slow motion, walking, jogging, and running, reflecting how user activity modulates the wireless channel over time. Crucially, this estimation pipeline is agnostic to how the effective CSI h[k, n] is ob… view at source ↗

**Figure 4.** Figure 4: Information leakage from the compressed CSI packets. The adversary is assumed to have a lower SNR then the legitimate AP (20 dB vs 5 dB) but is still able to accurately reconstruct the full CSI and estimate the STA’s speed from the compressed feedback packets. model and its MSE properties, which will serve as the building block for our proposed DP mechanism. A. Stochastic Quantization Consider a uniform sc… view at source ↗

**Figure 5.** Figure 5: Impact of privacy-preserving angle feedback on BER for different modulation orders and angle quantization resolutions where ε = 0.1 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Constellation comparison for a single spatial stream under (a) deterministic midpoint quantization of the Givens angles, and (b) the proposed differentially private stochastic quantization (DP–SQ) mechanism for Nt = 2, Nr = 1, 16-QAM, SNR = 15 dB and ε = 0.1, Bϕ = Bψ = 1 bit. V. CONCLUSION & FUTURE WORK In this paper, we have presented a standards-compatible framework for private CSI feedback based on DP q… view at source ↗

**Figure 7.** Figure 7: Relative beamforming gain for several IEEE 802.11 antenna configurations (2 × 1, 2 × 2, 2 × 3, 2 × 4, 2 × 8). The perfect SVD beamformer defines the reference gain G = 1, where ϵϕ = ϵψ = 0.1, Bϕ = 6 bits and Bψ = 3 bits. For the BeamDancer scheme [11], we adopt the randomization parameters that ensure the adversary’s activity classification accuracy remains strictly below 50% [PITH_FULL_IMAGE:figures/full… view at source ↗

**Figure 9.** Figure 9: Direct beamforming gain comparison under deterministic (non-private) and stochastic (private) quantization. (a) Beamforming gain over time. (b) Distribution of the normalized per-stream beamforming gain. The proposed privacy mechanism disrupts an adversary’s ability to infer STA motion while preserving a nearoptimal median beamforming gain and maintaining communication utility. [4] E. Perahia and R. Stac… view at source ↗

read the original abstract

Explicit channel state information (CSI) feedback in IEEE~802.11 conveys \emph{transmit beamforming directions} by reporting quantized Givens rotation and phase angles that parametrize the right-singular subspace of the channel matrix. Because these angles encode fine-grained spatial signatures of the propagation environment, recent work have shown that plaintext CSI feedback can inadvertently reveal user activity, identity, and location to passive eavesdroppers. In this work, we introduce a standards-compatible \emph{differentially private (DP) quantization mechanism} that replaces deterministic angular quantization with an $\varepsilon$-DP stochastic quantizer applied directly to the Givens parameters of the transmit beamforming matrix. The mechanism preserves the 802.11 feedback structure, admits closed-form sensitivity bounds for the angular representation, and enables principled privacy calibration. Numerical simulations demonstrate strong privacy guarantees with minimal degradation in beamforming performance.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper adds ε-DP via stochastic quantization directly to the Givens angles in 802.11 CSI feedback while keeping the exact standard format, but the closed-form sensitivity bounds may not hold uniformly without extra assumptions on channel singular values.

read the letter

The main contribution is a standards-compatible way to add differential privacy to compressed CSI feedback. Instead of the usual deterministic quantization of the Givens rotation and phase angles that describe the transmit beamforming matrix, they use a stochastic quantizer tuned to satisfy ε-DP. This keeps the feedback packet structure unchanged, so existing 802.11 devices and access points would not need protocol changes to adopt it. The abstract claims closed-form sensitivity bounds for the angular parameters, which would let designers set the noise level from the privacy budget without heavy simulation tuning. Their numerical results suggest beamforming gain stays close to the non-private case across the tested scenarios. That combination of compatibility and principled calibration is the practical hook. The work sits on top of earlier papers that showed plaintext CSI leaks activity and location; this one turns the leakage point into a controlled privacy mechanism. The approach is direct and avoids redesigning the feedback format, which is a plus for real-world uptake. One soft spot is the sensitivity analysis. The angles come from the SVD of the channel matrix, and the mapping is not Lipschitz with a fixed constant when singular values are close. Small channel perturbations can then swing the angles sharply. If the closed-form bounds in the paper rely on a minimum singular-value gap or fixed normalization that is not enforced in actual 802.11 operation, the ε guarantee becomes conditional on the propagation environment rather than general. The abstract does not list those assumptions, so the full derivation needs checking. Simulations are mentioned but without error bars or channel model details in the summary, it is hard to judge how broadly the performance claim holds. This paper is for wireless systems people working on physical-layer privacy, standards contributors thinking about 802.11be or later, and anyone building tools that process CSI on consumer hardware. It is concrete enough that a serious referee should see it. The idea is timely, the compatibility constraint is respected, and the central construction is falsifiable. I would send it to review rather than desk reject.

Referee Report

2 major / 2 minor

Summary. The paper introduces a standards-compatible differentially private (DP) quantization mechanism for IEEE 802.11 CSI feedback. It replaces deterministic angular quantization of Givens rotation and phase parameters (which parametrize the right-singular subspace of the channel matrix) with an ε-DP stochastic quantizer, claiming closed-form sensitivity bounds on the angular representation, preservation of the 802.11 feedback structure, and minimal beamforming degradation via numerical simulations.

Significance. If the closed-form sensitivity bounds hold without unstated restrictions on channel conditioning, the work would supply a practical, standards-compatible method to mitigate leakage of human activity signatures from plaintext CSI feedback while incurring little beamforming loss. This addresses a concrete privacy vulnerability in deployed wireless systems and could inform future 802.11 privacy extensions.

major comments (2)

[§3] §3 (sensitivity analysis): the closed-form sensitivity bounds for the Givens angles rest on the map from channel matrix H to rotation angles obtained via SVD; this map is not uniformly Lipschitz when singular values are close, yet the derivation provides no explicit minimum singular-value gap condition that is enforced in the 802.11 feedback path or in the simulation channel models. Without this, the ε-calibration is conditional rather than principled for arbitrary propagation environments.
[§5] §5 (numerical results): the simulations claim minimal beamforming degradation, but the text supplies neither error bars, explicit channel model parameters (e.g., singular-value distributions), nor details on data exclusion or Monte-Carlo repetitions, preventing verification that the privacy-performance tradeoff is robust across the regimes where the sensitivity bound may fail.

minor comments (2)

[§1] The abstract and introduction cite 'recent work' on CSI leakage without specific references; add the relevant citations in §1 to situate the contribution.
[§2] Notation for the stochastic quantizer (e.g., the exact distribution used for the angular perturbation) should be defined once in §2 and used consistently thereafter.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below with clarifications and commit to revisions that strengthen the explicitness of our assumptions and the reproducibility of our results.

read point-by-point responses

Referee: [§3] §3 (sensitivity analysis): the closed-form sensitivity bounds for the Givens angles rest on the map from channel matrix H to rotation angles obtained via SVD; this map is not uniformly Lipschitz when singular values are close, yet the derivation provides no explicit minimum singular-value gap condition that is enforced in the 802.11 feedback path or in the simulation channel models. Without this, the ε-calibration is conditional rather than principled for arbitrary propagation environments.

Authors: We agree that the mapping from the channel matrix to Givens angles via SVD is not uniformly Lipschitz without a sufficient gap between singular values. Our closed-form sensitivity bounds are derived under the standard assumption of distinct singular values, which holds in typical multipath environments relevant to 802.11. To address the concern, we will revise §3 to explicitly state the minimum singular-value gap condition, discuss its validity in the 802.11 feedback path, and note that the stochastic quantizer still satisfies ε-DP even if the bound is conservative when singular values approach equality. This makes the calibration principled under the channel conditions considered in the work. revision: yes
Referee: [§5] §5 (numerical results): the simulations claim minimal beamforming degradation, but the text supplies neither error bars, explicit channel model parameters (e.g., singular-value distributions), nor details on data exclusion or Monte-Carlo repetitions, preventing verification that the privacy-performance tradeoff is robust across the regimes where the sensitivity bound may fail.

Authors: We will revise §5 to include error bars computed over 1000 Monte-Carlo repetitions, explicit parameters for the channel models (including singular-value distributions drawn from the IEEE 802.11 TGn model), and full details on data exclusion criteria and simulation repetitions. These additions will enable verification of robustness, including in regimes with smaller singular-value gaps, and will confirm that the observed beamforming degradation remains minimal under the stated privacy levels. revision: yes

Circularity Check

0 steps flagged

No circularity: new DP quantizer and sensitivity bounds derived independently

full rationale

The paper presents a standards-compatible ε-DP stochastic quantizer applied to Givens rotation and phase angles extracted from the transmit beamforming matrix. The claimed closed-form sensitivity bounds are derived directly from the angular representation and the quantization mechanism itself, without reducing to a fitted parameter, self-defined quantity, or load-bearing self-citation chain. The mechanism preserves the 802.11 feedback structure by construction, and the privacy calibration follows from the new stochastic quantizer rather than any circular renaming or imported uniqueness theorem. No step equates a prediction to its own input by definition.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the angular representation of the right-singular subspace admits analytically tractable sensitivity bounds under the 802.11 quantization format, plus the standard definition of ε-differential privacy.

free parameters (1)

epsilon
Privacy budget parameter selected for calibration; not fitted to performance data in the abstract description.

axioms (1)

domain assumption The Givens rotation and phase angles admit closed-form sensitivity bounds for the DP mechanism
Invoked to enable principled privacy calibration while preserving the 802.11 feedback structure.

pith-pipeline@v0.9.0 · 5474 in / 1192 out tokens · 23088 ms · 2026-05-16T20:58:42.406825+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

replaces deterministic angular quantization with an ε-DP stochastic quantizer applied directly to the Givens parameters... admits closed-form sensitivity bounds for the angular representation
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 1 (Expected Subspace Distortion under DP–SQ)... E[d²_chord(S⋆,bS)] ≤ d²_q + 2·Ns·Ntot·(σ²_ψ + σ²_ϕ)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

[1]

Goldsmith,Wireless Communications

A. Goldsmith,Wireless Communications. Cambridge University Press, 2005

work page 2005
[2]

Wi- Fi: Twenty-five years and counting,

G. Geraci, F. Meneghello, F. Wilhelmi, D. Lopez-Perez, I. Val, L. G. Giordano, C. Cordeiro, M. Ghosh, E. Knightly, and B. Bellalta, “Wi- Fi: Twenty-five years and counting,”arXiv preprint arXiv:2507.09613, 2025. [3]IEEE Standard for Information Technology–Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks–Spec...

work page arXiv 2025
[3]

Perahia and R

E. Perahia and R. Stacey,Next Generation Wireless LANs: 802.11 n and 802.11 ac. Cambridge University Press, 2013

work page 2013
[4]

WiGest: A ubiquitous WiFi-based gesture recognition system,

H. Abdelnasser, M. Youssef, and K. A. Harras, “WiGest: A ubiquitous WiFi-based gesture recognition system,” inProceedings of the 2015 IEEE Conference on Computer Communications (INFOCOM), 2015, pp. 1472–1480

work page 2015
[5]

Device- free human activity recognition using commercial WiFi devices,

W. Wang, A. X. Liu, M. Shahzad, K. Ling, and S. Lu, “Device- free human activity recognition using commercial WiFi devices,”IEEE Journal on Selected Areas in Communications, vol. 35, no. 5, pp. 1118– 1131, 2017

work page 2017
[6]

Protecting privacy in WiFi sensing: A survey and new insights,

W. Jiang, C. Li, and X. Zhang, “Protecting privacy in WiFi sensing: A survey and new insights,” inProceedings of the 2021 IEEE Conference on Communications and Network Security (CNS), 2021, pp. 1–9

work page 2021
[7]

RF-protect: privacy against device-free human tracking,

J. Shenoy, Z. Liu, B. Tao, Z. Kabelac, and D. Vasisht, “RF-protect: privacy against device-free human tracking,” inProceedings of the ACM SIGCOMM 2022 Conference, 2022, pp. 588–600

work page 2022
[8]

CSI- bench: A large-scale in-the-wild dataset for multi-task WiFi sensing,

G. Zhu, Y . Hu, W. Gao, W.-H. Wang, B. Wang, and K. Liu, “CSI- bench: A large-scale in-the-wild dataset for multi-task WiFi sensing,” arXiv preprint arXiv:2505.21866, 2025

work page arXiv 2025
[9]

Lend me your beam: Privacy implications of plaintext beamforming feedback in WiFi,

Y . Liu, Y . Zeng, A. S. Uluagac, and S. Jana, “Lend me your beam: Privacy implications of plaintext beamforming feedback in WiFi,” in Proceedings of the Network and Distributed System Security Symposium (NDSS), 2024

work page 2024
[10]

Physical-layer privacy via randomized beam- forming against adversarial wi-fi sensing: Analysis, implementation, and evaluation,

M. Cominelli, S. Shahcheraghi, J. Link, M. Hollick, F. Cerutti, F. Gringoli, and A. Asadi, “Physical-layer privacy via randomized beam- forming against adversarial wi-fi sensing: Analysis, implementation, and evaluation,”IEEE Transactions on Wireless Communications, vol. 23, no. 12, pp. 19 603–19 617, 2024

work page 2024
[11]

Dwork and A

C. Dwork and A. Roth,The Algorithmic Foundations of Differential Privacy, ser. Foundations and Trends in Theoretical Computer Science. Now Publishers Inc., 2014, vol. 9, no. 3–4. 8

work page 2014

[1] [1]

Goldsmith,Wireless Communications

A. Goldsmith,Wireless Communications. Cambridge University Press, 2005

work page 2005

[2] [2]

Wi- Fi: Twenty-five years and counting,

G. Geraci, F. Meneghello, F. Wilhelmi, D. Lopez-Perez, I. Val, L. G. Giordano, C. Cordeiro, M. Ghosh, E. Knightly, and B. Bellalta, “Wi- Fi: Twenty-five years and counting,”arXiv preprint arXiv:2507.09613, 2025. [3]IEEE Standard for Information Technology–Telecommunications and Information Exchange between Systems Local and Metropolitan Area Networks–Spec...

work page arXiv 2025

[3] [3]

Perahia and R

E. Perahia and R. Stacey,Next Generation Wireless LANs: 802.11 n and 802.11 ac. Cambridge University Press, 2013

work page 2013

[4] [4]

WiGest: A ubiquitous WiFi-based gesture recognition system,

H. Abdelnasser, M. Youssef, and K. A. Harras, “WiGest: A ubiquitous WiFi-based gesture recognition system,” inProceedings of the 2015 IEEE Conference on Computer Communications (INFOCOM), 2015, pp. 1472–1480

work page 2015

[5] [5]

Device- free human activity recognition using commercial WiFi devices,

W. Wang, A. X. Liu, M. Shahzad, K. Ling, and S. Lu, “Device- free human activity recognition using commercial WiFi devices,”IEEE Journal on Selected Areas in Communications, vol. 35, no. 5, pp. 1118– 1131, 2017

work page 2017

[6] [6]

Protecting privacy in WiFi sensing: A survey and new insights,

W. Jiang, C. Li, and X. Zhang, “Protecting privacy in WiFi sensing: A survey and new insights,” inProceedings of the 2021 IEEE Conference on Communications and Network Security (CNS), 2021, pp. 1–9

work page 2021

[7] [7]

RF-protect: privacy against device-free human tracking,

J. Shenoy, Z. Liu, B. Tao, Z. Kabelac, and D. Vasisht, “RF-protect: privacy against device-free human tracking,” inProceedings of the ACM SIGCOMM 2022 Conference, 2022, pp. 588–600

work page 2022

[8] [8]

CSI- bench: A large-scale in-the-wild dataset for multi-task WiFi sensing,

G. Zhu, Y . Hu, W. Gao, W.-H. Wang, B. Wang, and K. Liu, “CSI- bench: A large-scale in-the-wild dataset for multi-task WiFi sensing,” arXiv preprint arXiv:2505.21866, 2025

work page arXiv 2025

[9] [9]

Lend me your beam: Privacy implications of plaintext beamforming feedback in WiFi,

Y . Liu, Y . Zeng, A. S. Uluagac, and S. Jana, “Lend me your beam: Privacy implications of plaintext beamforming feedback in WiFi,” in Proceedings of the Network and Distributed System Security Symposium (NDSS), 2024

work page 2024

[10] [10]

Physical-layer privacy via randomized beam- forming against adversarial wi-fi sensing: Analysis, implementation, and evaluation,

M. Cominelli, S. Shahcheraghi, J. Link, M. Hollick, F. Cerutti, F. Gringoli, and A. Asadi, “Physical-layer privacy via randomized beam- forming against adversarial wi-fi sensing: Analysis, implementation, and evaluation,”IEEE Transactions on Wireless Communications, vol. 23, no. 12, pp. 19 603–19 617, 2024

work page 2024

[11] [11]

Dwork and A

C. Dwork and A. Roth,The Algorithmic Foundations of Differential Privacy, ser. Foundations and Trends in Theoretical Computer Science. Now Publishers Inc., 2014, vol. 9, no. 3–4. 8

work page 2014