Jammer-Aware Semantic-Adaptive Integrated Sensing and Communication for Electronic-Warfare-Resilient 6G Networks


Kaya O., SHAH A. F. M. S., Noor-A-Rahim M.

IEEE Open Journal of the Communications Society, cilt.7, ss.13565-13582, 2026 (ESCI, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 7
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1109/ojcoms.2026.3746431
  • Dergi Adı: IEEE Open Journal of the Communications Society
  • Derginin Tarandığı İndeksler: Emerging Sources Citation Index (ESCI), Scopus, Compendex, INSPEC, Directory of Open Access Journals
  • Sayfa Sayıları: ss.13565-13582
  • Anahtar Kelimeler: 6G, anti-jamming, deep learning, electronic warfare, integrated sensing and communication, jammer classification, semantic communication
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Sixth-generation networks must remain operational under deliberate jamming. Integrated sensing and communication and semantic communication are two of the technologies expected to define these networks, yet both are usually studied under benign interference, and the studies that do classify jammers report a recognition accuracy without evaluating what the recognized label is worth to the system. This paper introduces JamSAC, a jammer-aware framework that carries a decision all the way from raw in-phase and quadrature samples to the choice of waveform and semantic rate. A compact multi-task neural network reads each received frame, identifies the jamming mechanism among six possibilities, and estimates how strong the jammer is. The mechanism selects a matched countermeasure, and the estimated strength selects the compression level of a learned semantic transceiver whose behavior has been measured rather than assumed. The same waveform and the same countermeasure serve the sensing function, so both functions are protected together. The framework is evaluated with the classifier in the loop, so every mistake it makes is charged to the final performance, and a simple theoretical bound, verified against the measurements, explains why only the mistakes that change the chosen countermeasure matter. On a purpose-built dataset covering four modulations, four channel families, and wide ranges of signal and jammer strength, the proposed policy recovers most of the semantic accuracy and goodput that a genie with perfect knowledge of the jammer could obtain, and it delivers an order-of-magnitude goodput improvement over presence-only detection. These findings hold for two semantic tasks, handwritten-digit and CIFAR-10 image classification, and they persist when the assumed countermeasure gains are perturbed and when the channel knowledge is imperfect. The results establish the concrete value of knowing the type and severity of a jammer rather than only its presence.