Autoencoder-Based FM Spectrum Anomaly Detection: A Pipeline-Level Study of Preprocessing, Model Capacity, and Reconstruction Scoring
Szilárd László Takács, Konrád Kajdy, Emanuel Puschita · 2026
Read Autoencoder-Based FM Spectrum Anomaly Detection: A Pipeline-Level Study of Preprocessing, Model Capacity, and Reconstruction Scoring by Szilárd László Takács, Konrád Kajdy, Emanuel Puschita (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13611061.txt.
Description
Szilárd László Takács, Konrád Kajdy, and Emanuel Puschita study how well autoencoders spot oddities in FM radio spectrum recordings. Monitoring the crowded airwaves means scanning waterfall images of frequency against time for rare anomalies. Their autoencoders learned only from normal broadcasts in the 87.5 to 108 megahertz band, so anything they failed to reconstruct could be flagged. The team varied the whole pipeline: normalization, image resolution, network depth, latent size, symmetry, and scoring.
More complex networks did not detect better. The best development setup reached an F1 score of 91.4 percent on labeled development data. On a held-out test set, performance dropped and the ranking of setups changed, warning against crowning a winner from development numbers. A controlled ablation showed the scoring formula itself changes how distinct oddities look. The study is pipeline-level guidance: preprocessing and scoring choices deserve the same respect as architecture.
Catalog record
| CREATOR | Szilárd László Takács, Konrád Kajdy, Emanuel Puschita |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |