Scholarlink

<p>The advantages of Machine Learning(ML) in fitting nonlinear problems have led to its widespread application and development in various fields, and it has become one of the currently popular technologies. In the field of communications, individual identification of signals by studying the collected signals is currently a popular research. However, the received information contains not only signals, but also complex channel information. The two types of information are mixed with each other, causing the simpler channel feature information will be selected for classification according to the classification task. In this way, the recognition effect varys if the channel information changes, which is incorrect. In response to the above problems, this paper using unmanned aerial vehicle(UAV) datasets and Minimum Mean Square Error (MMSE) as the basis, proposes a Signal Recognition Combining MMSE and Multi-Loss Convolutional Network (SMMCN) to accurately identify UAVs. The network obtains signal and channel information by training two different feature extraction layers, and peels off the originally mixed information, thereby reducing the impact of channel characteristics on signal classification, enabling the network to focus on the UAVs’ features when completing the task. Experiments have proven that this network can still extract signal features and achieve good recognition results even when there is a strong correlation between signal and channel features.</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

SMMCN: a Machine Learning Signal-Recognition Combining MMSE and Multi-Loss Convolutional Network

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Zherui Zhang,
  • Yingshen Zhu,
  • Wanyu Zhou,
  • Zheng Dou,
  • Hang Jiang,
  • Shuang Li

摘要

The advantages of Machine Learning(ML) in fitting nonlinear problems have led to its widespread application and development in various fields, and it has become one of the currently popular technologies. In the field of communications, individual identification of signals by studying the collected signals is currently a popular research. However, the received information contains not only signals, but also complex channel information. The two types of information are mixed with each other, causing the simpler channel feature information will be selected for classification according to the classification task. In this way, the recognition effect varys if the channel information changes, which is incorrect. In response to the above problems, this paper using unmanned aerial vehicle(UAV) datasets and Minimum Mean Square Error (MMSE) as the basis, proposes a Signal Recognition Combining MMSE and Multi-Loss Convolutional Network (SMMCN) to accurately identify UAVs. The network obtains signal and channel information by training two different feature extraction layers, and peels off the originally mixed information, thereby reducing the impact of channel characteristics on signal classification, enabling the network to focus on the UAVs’ features when completing the task. Experiments have proven that this network can still extract signal features and achieve good recognition results even when there is a strong correlation between signal and channel features.