FreqSpaNet: Frequency and Spatial Learning of SFPF for Open-Set Hardware Anomaly Detection

Forum topic · 小凯 · 2026-09-17

Summary

FreqSpaNet is a representation learning network for open-set hardware anomaly detection in wireless devices, addressing unauthorized hardware replacement that preserves a device's logical identity while changing its physical implementation. The method leverages spatio-frequency polarization fingerprints (SFPFs), which capture device-dependent responses across multiple frequencies and directions. Because the frequency and spatial dimensions of SFPFs show different structural dependencies, FreqSpaNet uses two specialized branches: a frequency branch that captures local variations among neighboring frequencies, and a geometry-aware spatial branch that models directional relationships using angular information. The two representations are combined through adaptive fusion, with complementary pretraining that captures shared information while preserving the distinctive features of each branch. Experiments show FreqSpaNet achieves an average AUROC of 96.31%, outperforming baselines by 9.05 percentage points, and results across seven hardware replacement scenarios validate its effectiveness. Paper: arXiv 2609.17491, by Xiaoxuan Huang, Jinlong Xu, YiZhe Wang, Meng Zhang, Xian Li, and Yuying Bian.

论文概要

研究领域: ML 作者: Xiaoxuan Huang, Jinlong Xu, YiZhe Wang, Meng Zhang, Xian Li, Yuying Bian 发布时间: 2026-09-15 arXiv: 2609.17491

Chinese Abstract (translated)

The authors propose FreqSpaNet, an SFPF representation learning network for open set hardware anomaly detection:

A frequency branch captures local variations among neighboring frequencies.
A geometry-aware spatial branch models directional relationships using angular information.
The two representations are combined through adaptive fusion.
Complementary pretraining further captures shared information while preserving the unique characteristics of the frequency and spatial representations.

Experiments show that FreqSpaNet achieves an average AUROC of 96.31%, which is 9.05 percentage points higher than the baseline. Results across seven hardware replacement scenarios further verify the effectiveness of FreqSpaNet.

Original Abstract (excerpt)

Unauthorized hardware replacement can preserve a wireless device's logical identity while altering its physical implementation, posing a challenge to hardware integrity verification. Spatio-frequency polarization fingerprints (SFPFs) capture device-dependent responses across multiple frequencies and directions, but their frequency and spatial dimensions exhibit different structural dependencies. We propose FreqSpaNet, an SFPF representation learning network for open set hardware anomaly detection. A frequency branch captures local variations among neighboring frequencies, while a geometry-aware spatial branch models directional relationships using angular information. The two representations are combined through adaptive fusion, and complementary pretraining further captures shared informa...

Full paper: https://arxiv.org/abs/2609.17491

--- *Auto-collected on 2026-09-17*