Scholarlink

<p>This study presents a novel deep learning (DL)-based transmission power control (TPC) method for wireless sensor networks (WSN), focused on improving energy efficiency and network performance. By utilizing DL to estimate signal-to-noise ratio (SNR) for various modulation schemes, we enable the master sensor to optimize power consumption in connected sensors. The method involves training models for eight modulation schemes, including quadrature phase shift keying (QPSK), 8PSK, 16PSK, 32PSK, quadrature amplitude modulation (QAM)16, QAM64, QAM256, and QAM1024, and using a transfer learning approach to adapt popular image classification models like MobileNetV2, ResNetV2 (50, 101, and 152 layers), Xception, InceptionV3, and InceptionResNetV2 for SNR estimation. Based on these estimates, the master sensor adjusts the transmission power of the slave sensors to maintain optimal performance while saving energy. Results show that InceptionV3 delivers the best accuracy, making it ideal for our application. InceptionV3 successfully classifies SNRs between 0 and 20 dB for all modulation schemes, achieving an accuracy of over 90%, while lighter models, such as MobileNetV2 and Xception, yielded accuracies as low as 44% for complex, high-order modulation schemes like QAM256 and QAM1024. This result underscores the InceptionV3’s reliability in identifying different SNR levels. This approach can facilitate sensors in the network use power more efficiently while maintaining reliable communication.</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

Deep learning aided transmission power control for wireless sensor networks

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Merih Leblebici,
  • Ali Calhan

摘要

This study presents a novel deep learning (DL)-based transmission power control (TPC) method for wireless sensor networks (WSN), focused on improving energy efficiency and network performance. By utilizing DL to estimate signal-to-noise ratio (SNR) for various modulation schemes, we enable the master sensor to optimize power consumption in connected sensors. The method involves training models for eight modulation schemes, including quadrature phase shift keying (QPSK), 8PSK, 16PSK, 32PSK, quadrature amplitude modulation (QAM)16, QAM64, QAM256, and QAM1024, and using a transfer learning approach to adapt popular image classification models like MobileNetV2, ResNetV2 (50, 101, and 152 layers), Xception, InceptionV3, and InceptionResNetV2 for SNR estimation. Based on these estimates, the master sensor adjusts the transmission power of the slave sensors to maintain optimal performance while saving energy. Results show that InceptionV3 delivers the best accuracy, making it ideal for our application. InceptionV3 successfully classifies SNRs between 0 and 20 dB for all modulation schemes, achieving an accuracy of over 90%, while lighter models, such as MobileNetV2 and Xception, yielded accuracies as low as 44% for complex, high-order modulation schemes like QAM256 and QAM1024. This result underscores the InceptionV3’s reliability in identifying different SNR levels. This approach can facilitate sensors in the network use power more efficiently while maintaining reliable communication.