EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

CNN‐DDQN Spatial Learning for Fractional Frequency Reuse Optimization in 5G Heterogeneous Networks

Christian Onu, Abdulwaheed Musa, Olalekan Ogunbiyi, Lambe Mutalub Adesina, Bilkisu Jimada‐Ojuolape, Monsurat Balogun, Abdurrhaman Ademola Isa, Aliyu Ahmed, Ishmael Zibani, Adamu Murtala Zungeru

Engineering Reports · 2026-07 · 8 (7)

DOI: 10.1002/eng2.70970 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering Reports, online erschienen am 22.07.2026. Verfasst von Christian Onu und 9 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Fractional frequency reuse (FFR) is an efficient interference coordination method for enhancing cell‐edge performance of dense fifth‐generation (5G) heterogeneous networks (HetNets). Unfortunately, traditional FFR arrangements based on static or rule‐oriented settings cannot adapt to dynamic traffic distribution and spatial interference patterns. To overcome these limitations, we present a spatially aware resource allocation framework that integrates convolutional neural networks (CNNs) with double deep Q‐learning (DDQN) to optimize FFR. The proposed approach represents the network state as a multi‐channel spatial tensor that captures the signal‐to‐interference‐plus‐noise ratio (SINR), interference levels, user density, and channel quality across the network area. A lightweight CNN architecture is employed to extract hierarchical spatial features from the network state, enabling the DDQN agent to learn interference‐aware FFR policies. Extensive simulations based on 3GPP‐compliant HetNet models are conducted across multiple deployment scenarios, including urban dense, suburban, and rural environments. The results demonstrate that the proposed CNN‐DDQN framework achieves significant performance improvements, including up to a 21% increase in average throughput, a 30% improvement in cell‐edge throughput, and a 15% improvement in fairness compared with traditional static FFR schemes. In addition, the proposed model improves performance by 8%–12% over conventional DRL‐based approaches that lack spatial feature extraction. The lightweight architecture contains only 26,425 trainable parameters, enabling fast training and real‐time inference with latency below 10 ms. These results demonstrate that spatial feature extraction can significantly enhance reinforcement learning for interference management and resource allocation in next‐generation wireless networks.

Zitieren

APA

Onu, C., Musa, A., Ogunbiyi, O., Adesina, L. M., Jimada‐Ojuolape, B., Balogun, M., Isa, A. A., Ahmed, A., Zibani, I., & Zungeru, A. M. (2026). CNN‐DDQN Spatial Learning for Fractional Frequency Reuse Optimization in 5G Heterogeneous Networks. Engineering Reports, 8(7). https://doi.org/10.1002/eng2.70970

Vancouver

Onu C, Musa A, Ogunbiyi O, Adesina LM, Jimada‐Ojuolape B, Balogun M, et al. CNN‐DDQN Spatial Learning for Fractional Frequency Reuse Optimization in 5G Heterogeneous Networks. Engineering Reports. 2026;8(7). doi:10.1002/eng2.70970

BibTeX
@article{Onu2026,
  author = {Onu, Christian and Musa, Abdulwaheed and Ogunbiyi, Olalekan and Adesina, Lambe Mutalub and Jimada‐Ojuolape, Bilkisu and Balogun, Monsurat and Isa, Abdurrhaman Ademola and Ahmed, Aliyu and Zibani, Ishmael and Zungeru, Adamu Murtala},
  title = {CNN‐DDQN Spatial Learning for Fractional Frequency Reuse Optimization in 5G Heterogeneous Networks},
  journal = {Engineering Reports},
  year = {2026},
  volume = {8},
  number = {7},
  doi = {10.1002/eng2.70970}
}
Kostenlos starten →