Elmorsy Sally Mohamed Ali, Osman Samah Mohamed, Gamel Samah Adel
High Institute of Management, Mahalla, Egypt.
Faculty of Computers and Data Science, Alexandria University, Alexandria, Egypt.
Sci Rep. 2025 Jul 10;15(1):24825. doi: 10.1038/s41598-025-07837-2.
This paper introduces a novel approach to enhancing spectrum sensing accuracy for 5G and LTE signals using advanced deep learning models, with a particular focus on the impact of systematic hyperparameter tuning. By leveraging state-of-the-art neural network architecture, namely DenseNet121 and InceptionV3-the study aims to overcome the limitations of traditional spectrum sensing methods in highly dynamic and noisy wireless environments. The research highlights that, through rigorous hyperparameter optimization, these models achieved substantial improvements in detection accuracy, reaching 97.3% and 98.2%, respectively, compared to initial performance levels of 93.0% and 95.0%. These performance improvements were particularly notable in controlled scenarios where low signal-to-noise ratio frames were excluded, with 60% of frames containing little or no information-highlighting the critical role of signal quality in both training and evaluation. It is worth noting that the models were trained and tested on a large and diverse dataset, including synthetic signals and real-world data, simulating a wide range of practical deployment conditions. This comprehensive database supports the generalizability of the proposed approach and its real-world applicability. The study also confirms that the models demonstrated competitive performance in various test scenarios, and that their integration into future wireless systems could significantly enhance smart spectrum management and real-time communication reliability in modern networks.
本文介绍了一种使用先进深度学习模型提高5G和LTE信号频谱感知准确性的新方法,特别关注系统超参数调整的影响。通过利用最先进的神经网络架构,即DenseNet121和InceptionV3,该研究旨在克服传统频谱感知方法在高度动态和嘈杂的无线环境中的局限性。研究强调,通过严格的超参数优化,这些模型在检测准确性方面取得了显著提高,与初始性能水平93.0%和95.0%相比,分别达到了97.3%和98.2%。在排除低信噪比帧的受控场景中,这些性能提升尤为显著,其中60%的帧几乎不包含或不包含信息,这突出了信号质量在训练和评估中的关键作用。值得注意的是,这些模型在一个大型且多样的数据集上进行了训练和测试,包括合成信号和真实世界数据,模拟了广泛的实际部署条件。这个综合数据库支持了所提出方法的通用性及其在现实世界中的适用性。该研究还证实,这些模型在各种测试场景中表现出具有竞争力的性能,并且将它们集成到未来的无线系统中可以显著提高现代网络中的智能频谱管理和实时通信可靠性。