Skip to main navigation Skip to search Skip to main content

Blind Detection of FHSS Signals with Image Segmentation Based Low Complexity Deep Learning Architecture

  • ASELSAN Inc.
  • Istanbul Technical University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Frequency-Hopping Spread Spectrum (FHSS) is a resilient communication technique widely adopted in military and commercial systems due to its resistance to jamming, interception, and interference. Reliable blind detection of FHSS signals is essential for real-time situational awareness in electronic warfare and spectrum monitoring, especially under edge constraints. This paper introduces a low-complexity encoder-decoder convolutional neural network (CNN) for blind FHSS detection from spectrograms generated via the ShortTime Fourier Transform (STFT). Designed for edge deployment, the model ensures minimal latency and low power consumption. With only 15K parameters and 1.3 GMACs for a 1024×1024 spectrogram, it enables real-time inference while maintaining high performance. To ensure operational realism, the dataset incorporates signals down to 5 dB SNR. On this dataset, the network achieves 93.82% intersection-over-union (IoU) and 96.01% hop detection accuracy. For comparison, a lightweight UNet (LV-UNet) attains 91.22% IoU and 94.24% hop detection at significantly higher complexity. Notably, classical Otsu thresholding fails to adapt below 10 dB due to excessive false alarms; hence its performance degrades to 77.25% IoU and 81.55% hop detection accuracy. In contrast, our model remains robust, and is able to resolve overlapping signals with a minimum 3-pixel guard interval. Validation with a real-world over-the-air dataset, augmented with synthetic fading and noise, confirms operational reliability in adversarial and contested radio environments.

Original languageEnglish
Title of host publication2026 IEEE International Black Seas Conference on Communications and Networking, BlackSeaCom 2026
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798319543448
DOIs
Publication statusPublished - 2026
Event2026 IEEE International Black Seas Conference on Communications and Networking, BlackSeaCom 2026 - Bucharest, Romania
Duration: 8 Jun 2026 → 11 Jun 2026

Publication series

Name2026 IEEE International Black Seas Conference on Communications and Networking, BlackSeaCom 2026

Conference

Conference2026 IEEE International Black Seas Conference on Communications and Networking, BlackSeaCom 2026
Country/TerritoryRomania
CityBucharest
Period8/06/26 → 11/06/26

Bibliographical note

Publisher Copyright:
© 2026 IEEE.

Keywords

  • Frequency-Hopping Spread Spectrum (FHSS)
  • blind signal detection
  • convolutional neural network (CNN)
  • deep learning
  • image segmentation
  • real-time processing
  • spectrum sensing
  • tactical communications

Fingerprint

Dive into the research topics of 'Blind Detection of FHSS Signals with Image Segmentation Based Low Complexity Deep Learning Architecture'. Together they form a unique fingerprint.

Cite this