EUVIMEDEUVIMED entdecken
Crossref

Graph Over the Grid: A Spatial K -Nearest Neighbor Approach to Modulation Recognition in OTFS

Tonmoy Rajkhowa, Sanjeev Sharma, Kuntal Deka, Tharmalingam Ratnarajah

IEEE Communications Letters · 2026 · 30 · 3678-3682

DOI: 10.1109/lcomm.2026.3744796 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Tonmoy Rajkhowa und 3 weiteren Autor:innen. Die Arbeit nennt 13 Literaturangaben. Gefördert von: SPARC-UKIERI.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Automatic modulation classification (AMC) for orthogonal time–frequency space (OTFS) modulated signals is critical for next-generation wireless systems operating in high-mobility environments. Existing convolutional and transformer-based approaches generally process the OTFS delay-Doppler (DD) frame as a regular Euclidean grid or a sequence of patches. This may be restrictive for representing dispersed modulation-dependent structures that often exhibit irregular, non-local patterns that are better captured by graph representations. This letter presents ViGNN-OTFS, an OTFS-specific adaptation of the vision-graph neural network (Vision-GNN) paradigm in which the DD frame is partitioned into patches represented as graph nodes. In contrast to the dynamic feature-space $K$ -nearest-neighbor ( $K$ -NN) graph in Vision-GNN, ViGNN-OTFS constructs a fixed directed $K$ -NN graph solely from the spatial coordinates of DD patches, computed once for a given frame configuration and reused during training and inference. This provides a stable spatial inductive bias that is independent of noise-induced variations in the node features. Experimental validations on the extended vehicular-A (EVA) channel demonstrate that ViGNN-OTFS significantly outperforms convolutional and transformer-based approaches, achieving superior accuracy across a wide range of signal-to-noise ratios.

Förderung

  • SPARC-UKIERI

Zitieren

APA

Rajkhowa, T., Sharma, S., Deka, K., & Ratnarajah, T. (2026). Graph Over the Grid: A Spatial K -Nearest Neighbor Approach to Modulation Recognition in OTFS. IEEE Communications Letters, 30, 3678-3682. https://doi.org/10.1109/lcomm.2026.3744796

Vancouver

Rajkhowa T, Sharma S, Deka K, Ratnarajah T. Graph Over the Grid: A Spatial K -Nearest Neighbor Approach to Modulation Recognition in OTFS. IEEE Communications Letters. 2026;30:3678-3682. doi:10.1109/lcomm.2026.3744796

BibTeX
@article{Rajkhowa2026,
  author = {Rajkhowa, Tonmoy and Sharma, Sanjeev and Deka, Kuntal and Ratnarajah, Tharmalingam},
  title = {Graph Over the Grid: A Spatial K -Nearest Neighbor Approach to Modulation Recognition in OTFS},
  journal = {IEEE Communications Letters},
  year = {2026},
  volume = {30},
  pages = {3678--3682},
  doi = {10.1109/lcomm.2026.3744796}
}
Kostenlos starten →