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

Forum topic · 小凯 · 2026-09-17

Summary

Unauthorized hardware replacement can alter a wireless device's physical implementation while preserving its logical identity, making hardware integrity verification difficult. Spatio-frequency polarization fingerprints (SFPFs) capture device-dependent responses across multiple frequencies and directions, but their frequency and spatial dimensions exhibit distinct structural dependencies. FreqSpaNet is 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 captures shared information while preserving each branch's unique features. Experiments show FreqSpaNet achieves a mean AUROC of 96.31%, outperforming baselines by 9.05 percentage points, with results across seven hardware replacement scenarios further confirming its effectiveness.

Overview

Research area: Machine Learning
Authors: Xiaoxuan Huang, Jinlong Xu, YiZhe Wang, Meng Zhang, Xian Li, Yuying Bian
Published: 2026-09-15
arXiv: 2609.17491

Abstract

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 information while retaining the distinctive features of both frequency and spatial representations.

Key Results

FreqSpaNet achieves a mean AUROC of 96.31%, outperforming the baseline by 9.05 percentage points.
Evaluation across seven hardware replacement scenarios further validates the effectiveness of the approach.

---

*Auto-collected on 2026-09-17 from arXiv:2609.17491*