Scholarlink

<p>Path loss prediction models are crucial for planning and deploying wireless communication networks. However, existing prediction models, such as empirical (e.g., Okumura-Hata) and deterministic (e.g., ray-tracing) models, have limitations. Empirical models lack universality, while deterministic models are complex for large-scale deployments. This research proposes a hybrid path loss prediction approach that combines convolutional neural networks (CNNs) with existing empirical models. Satellite images are used as input to the CNNs, which extract relevant features and generate correction factors for the empirical models. Two hybrid models are developed: CNN-FREESPACE and CNN-HATA. While CNN-FREESPACE leverages the free-space model, CNN- HATA utilizes the Okumura-Hata model, both incorporating CNN-derived correction factors for enhanced path loss estimation. The proposed hybrid models were compared with traditional empirical models using real-world data from television broadcasting transmitters in Adamawa and Bauchi states, Nigeria. The results demonstrate that the hybrid models significantly outperform the empirical models in terms of accuracy. The CNN-FREESPACE model exhibits the best performance, achieving an average mean absolute error (MAE) of 5.73&#xa0;dB and an average root mean squared error (RMSE) of 7.24&#xa0;dB for the test route in Adamawa State. Similarly, in Bauchi state, the CNN-FREESPACE model provided an MAE of 4.84&#xa0;dB and RMSE of 6.13&#xa0;dB. These results highlight how CNNs enhance path loss estimates by refining existing path loss models with site-specific corrections. The proposed hybrid path loss prediction models are especially amenable for several applications including digital TV coverage analysis, cellular network planning, and spectrum management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid path loss models using convolutional neural network and empirical models for path loss prediction in the ultra-high frequency

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Matthew K. Luka,
  • Okpo U. Okereke,
  • Ejike C. Anene

摘要

Path loss prediction models are crucial for planning and deploying wireless communication networks. However, existing prediction models, such as empirical (e.g., Okumura-Hata) and deterministic (e.g., ray-tracing) models, have limitations. Empirical models lack universality, while deterministic models are complex for large-scale deployments. This research proposes a hybrid path loss prediction approach that combines convolutional neural networks (CNNs) with existing empirical models. Satellite images are used as input to the CNNs, which extract relevant features and generate correction factors for the empirical models. Two hybrid models are developed: CNN-FREESPACE and CNN-HATA. While CNN-FREESPACE leverages the free-space model, CNN- HATA utilizes the Okumura-Hata model, both incorporating CNN-derived correction factors for enhanced path loss estimation. The proposed hybrid models were compared with traditional empirical models using real-world data from television broadcasting transmitters in Adamawa and Bauchi states, Nigeria. The results demonstrate that the hybrid models significantly outperform the empirical models in terms of accuracy. The CNN-FREESPACE model exhibits the best performance, achieving an average mean absolute error (MAE) of 5.73 dB and an average root mean squared error (RMSE) of 7.24 dB for the test route in Adamawa State. Similarly, in Bauchi state, the CNN-FREESPACE model provided an MAE of 4.84 dB and RMSE of 6.13 dB. These results highlight how CNNs enhance path loss estimates by refining existing path loss models with site-specific corrections. The proposed hybrid path loss prediction models are especially amenable for several applications including digital TV coverage analysis, cellular network planning, and spectrum management.