Multi-Label Recognition of Composite Electromagnetic Interference for ILS Localizer Monitoring: Measurement- Constrained Modeling and Robustness Evaluation
Year: 2026
Type: article
Abstract: Overlapping transient, narrowband, and broadband electromagnetic interference complicates the identification of concurrent sources affecting Instrument Landing System (ILS) Localizer signals. This study develops a measurement-constrained framework for composite interference modeling, multi-label recognition, and evaluation of robustness to source-model formulation. Archived field and controlled-source measurements guide equivalent-baseband models of pantograph–catenary arcs, high-voltage power-line interference, illegal broadcasts, mobile-phone jammers, and aging runway-light interference, but are not used as labeled learning samples. Source-wise protection-ratio control produces clean, single-source, and composite synthetic spectrograms in ACE-ILS-MI. The Spectrogram-Native Residual Network (SNRN) achieves a Macro-F1 of 0.9215 on its held-out synthetic test set. In a separate robustness study, paired source-factorized training achieves mean-world and worst-world Macro-F1 values of 0.9420 and 0.9210 on a sealed 32-world synthetic panel, and 0.8905 on an unseen-generator-family synthetic holdout. Cross-factorial and matched-view analyses associate the robustness gains primarily with generator-randomized training and additional view exposure, rather than attention or one-source-flip pairing. One unlabeled measured in-phase/quadrature (IQ) recording provides a preprocessing and inference compatibility check, not a recognition-accuracy estimate. The framework supports controlled analysis of composite-source recognition and generator-dependent performance; quantitative validation on labeled field data remains necessary.
Source: Electronics
Authors: Shengjie Wang(Conceptualization, Formal analysis, Funding acquisition, Project administration, Supervision, Validation, Writing – review & editing), Yuanxu Wang(Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing), Xiaojia Yang(Investigation, Resources), Yaqing Chen(Investigation, Resources), Jiqian Wang(Methodology, Resources), Liangyong Yang(Investigation, Resources), Anzhe Dai(Data curation, Investigation)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7237211149
