EUVIMEDEUVIMED entdecken
Crossref

A LINK-LAYER MODEL OF A UAV SWARM MESH NETWORK WITH ENCAPSULATION OF INTERMEDIATE DEEP NEURAL NETWORK ACTIVATIONS: ANALYTICAL EVALUATION OF LATENCY AND TRAFFIC VOLUME

Artem Kasianchuk, Halyna Lastivka

Infocommunication technologies and electronic engineering · 2026-11 · 6 (2) · 144-154

DOI: 10.23939/ictee2026.02.144 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Infocommunication technologies and electronic engineering, 2026-10. Verfasst von Artem Kasianchuk und 1 weiteren Autor:innen. Die Arbeit nennt 21 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The transition from single unmanned aerial vehicles (UAVs) to coordinated swarms performing collective perception missions creates a fundamental tension between the volume of sensor data that must be exchanged and the limited, unstable bandwidth of flying ad hoc networks (FANET). In the split computing paradigm, a deep neural network (DNN) is partitioned among swarm nodes, so that instead of raw imagery the nodes exchange intermediate activation tensors; however, existing distributed inference schemes for UAV swarms transmit these features at the application level over the full L2-L7 protocol stack, which adds per-frame header overhead, increases the number of frames, and provides no link-layer mechanisms for controlling the freshness and completeness of tensor delivery. This paper proposes a link-layer model of a UAV swarm mesh network, SwarmNS-MAC, in which IEEE 802.11 frames serve as direct containers for quantized fragments of intermediate DNN activations. The paper introduces a specification of the frame format with a compact 24-byte TLV header that carries the model identifier, split point, tensor geometry, quantization scheme, fragment sequence number, and an Age-of-Information timestamp; a three-phase collective inference algorithm with a quorum-based graceful degradation mechanism; and an analytical model relating quantization bit-width, traffic volume, per-hop transmission time, whole-tensor delivery probability, and the probability of reaching the aggregation quorum. Numerical evaluation for a reference 256×80×80 activation tensor, typical of YOLO-family object detectors, shows that L2 encapsulation reduces traffic volume by ≈3.7 %, transmission time by ≈7.9 %, and the number of frames by ≈36.5 % compared with the L2-L7 stack, while post-training quantization from FP32 to 0.8 bits per element reduces the per-hop transmission time approximately 40-fold, from 4.9 s to 122.8 ms at an effective rate of 12 Mbit/s. Reducing the number of frames also raises the probability of delivering the tensor intact (0.70 versus 0.57 at a per-frame success probability of 0.9995) and, consequently, the probability of reaching the aggregation quorum (0.93 for a quorum of three out of six neighbors). The model deliberately abstracts from collision effects at high swarm densities and from IEEE 802.11 frame aggregation, which are stated as explicit limitations. The results provide a basis for subsequent simulation-based validation of the model in NS-3/OMNeT++ and for selecting the DNN split point under FANET constraints.

Zitieren

APA

Kasianchuk, A., & Lastivka, H. (2026). A LINK-LAYER MODEL OF A UAV SWARM MESH NETWORK WITH ENCAPSULATION OF INTERMEDIATE DEEP NEURAL NETWORK ACTIVATIONS: ANALYTICAL EVALUATION OF LATENCY AND TRAFFIC VOLUME. Infocommunication technologies and electronic engineering, 6(2), 144-154. https://doi.org/10.23939/ictee2026.02.144

Vancouver

Kasianchuk A, Lastivka H. A LINK-LAYER MODEL OF A UAV SWARM MESH NETWORK WITH ENCAPSULATION OF INTERMEDIATE DEEP NEURAL NETWORK ACTIVATIONS: ANALYTICAL EVALUATION OF LATENCY AND TRAFFIC VOLUME. Infocommunication technologies and electronic engineering. 2026;6(2):144-154. doi:10.23939/ictee2026.02.144

BibTeX
@article{Kasianchuk2026,
  author = {Kasianchuk, Artem and Lastivka, Halyna},
  title = {A LINK-LAYER MODEL OF A UAV SWARM MESH NETWORK WITH ENCAPSULATION OF INTERMEDIATE DEEP NEURAL NETWORK ACTIVATIONS: ANALYTICAL EVALUATION OF LATENCY AND TRAFFIC VOLUME},
  journal = {Infocommunication technologies and electronic engineering},
  year = {2026},
  volume = {6},
  number = {2},
  pages = {144--154},
  doi = {10.23939/ictee2026.02.144}
}
Kostenlos starten →