Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Differentially Private Feature Release for Wireless Sensing: Adaptive Privacy Budget Allocation on CSI Spectrograms

Ipek Sena Yilmaz, Onur G. Tuncer, Zeynep E. Aksoy, Zeynep Yağmur Baydemir

TL;DR

This work tackles the privacy risks of sharing CSI-derived spectrogram features for wireless sensing by introducing an adaptive differential privacy framework. The method estimates time–frequency importance and allocates a global privacy budget across spectrogram blocks, releasing DP-protected features via a blockwise Gaussian mechanism with Rényi DP accounting. Across tasks including multi-user HAR, 3D pose estimation, and respiration monitoring, adaptive budget allocation consistently achieves better privacy–utility trade-offs than uniform perturbation and substantially reduces identity, location, and membership leakage. The approach is practical for benchmark sharing and cloud analytics, offering formal DP guarantees while preserving key sensing cues across environments and bands. The findings demonstrate that task-aligned budget allocation is a powerful, scalable privacy layer for modern Wi‑Fi sensing pipelines.

Abstract

Wi-Fi/RF-based human sensing has achieved remarkable progress with deep learning, yet practical deployments increasingly require feature sharing for cloud analytics, collaborative training, or benchmark evaluation. Releasing intermediate representations such as CSI spectrograms can inadvertently expose sensitive information, including user identity, location, and membership, motivating formal privacy guarantees. In this paper, we study differentially private (DP) feature release for wireless sensing and propose an adaptive privacy budget allocation mechanism tailored to the highly non-uniform structure of CSI time-frequency representations. Our pipeline converts CSI to bounded spectrogram features, applies sensitivity control via clipping, estimates task-relevant importance over the time-frequency plane, and allocates a global privacy budget across spectrogram blocks before injecting calibrated Gaussian noise. Experiments on multi-user activity sensing (WiMANS), multi-person 3D pose estimation (Person-in-WiFi 3D), and respiration monitoring (Resp-CSI) show that adaptive allocation consistently improves the privacy-utility frontier over uniform perturbation under the same privacy budget. Our method yields higher accuracy and lower error while substantially reducing empirical leakage in identity and membership inference attacks.

Differentially Private Feature Release for Wireless Sensing: Adaptive Privacy Budget Allocation on CSI Spectrograms

TL;DR

This work tackles the privacy risks of sharing CSI-derived spectrogram features for wireless sensing by introducing an adaptive differential privacy framework. The method estimates time–frequency importance and allocates a global privacy budget across spectrogram blocks, releasing DP-protected features via a blockwise Gaussian mechanism with Rényi DP accounting. Across tasks including multi-user HAR, 3D pose estimation, and respiration monitoring, adaptive budget allocation consistently achieves better privacy–utility trade-offs than uniform perturbation and substantially reduces identity, location, and membership leakage. The approach is practical for benchmark sharing and cloud analytics, offering formal DP guarantees while preserving key sensing cues across environments and bands. The findings demonstrate that task-aligned budget allocation is a powerful, scalable privacy layer for modern Wi‑Fi sensing pipelines.

Abstract

Wi-Fi/RF-based human sensing has achieved remarkable progress with deep learning, yet practical deployments increasingly require feature sharing for cloud analytics, collaborative training, or benchmark evaluation. Releasing intermediate representations such as CSI spectrograms can inadvertently expose sensitive information, including user identity, location, and membership, motivating formal privacy guarantees. In this paper, we study differentially private (DP) feature release for wireless sensing and propose an adaptive privacy budget allocation mechanism tailored to the highly non-uniform structure of CSI time-frequency representations. Our pipeline converts CSI to bounded spectrogram features, applies sensitivity control via clipping, estimates task-relevant importance over the time-frequency plane, and allocates a global privacy budget across spectrogram blocks before injecting calibrated Gaussian noise. Experiments on multi-user activity sensing (WiMANS), multi-person 3D pose estimation (Person-in-WiFi 3D), and respiration monitoring (Resp-CSI) show that adaptive allocation consistently improves the privacy-utility frontier over uniform perturbation under the same privacy budget. Our method yields higher accuracy and lower error while substantially reducing empirical leakage in identity and membership inference attacks.
Paper Structure (96 sections, 11 equations, 4 figures, 9 tables)

This paper contains 96 sections, 11 equations, 4 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Wireless sensing at scale: capabilities, representations, and hardware trends
  3. Application breadth: from localization and profiling to vital signs and identification
  4. Robustness, domain shift, and physical factors: why "feature release" is hard
  5. Security and privacy: from perturbation attacks to the need for formal guarantees
  6. Our contributions
  7. Related Work
  8. Wi-Fi sensing measurements, protocol evolution, and enabling hardware
  9. Representations and learning models for Wi-Fi human sensing
  10. Activity sensing, multi-user settings, and fine-grained human understanding
  11. Indoor localization and positioning: datasets, learning, and RTT systems
  12. Healthcare and vital signs sensing, and the role of backscatter
  13. Human and device identification, RF fingerprinting, and authentication
  14. Security and robustness: perturbation attacks, physical factors, and generalization under threat
  15. Positioning the gap: toward formal privacy for feature release in Wi-Fi sensing
  16. ...and 81 more sections

Figures (4)

  • Figure 1: System overview
  • Figure 2: WiMANS (5 GHz) privacy--utility trade-off. Adaptive budget allocation achieves higher Macro-F1 under the same global privacy budget $\epsilon$.
  • Figure 3: WiMANS (5 GHz) identity inference leakage (attacker Top-1). DP significantly reduces re-identification risk; adaptive allocation maintains strong leakage suppression.
  • Figure 4: Resp-CSI respiration rate estimation error under DP feature release.