Scholarlink

<p>In this research, we propose a novel hybrid transfer convolutional network (HTCN) model for efficient and secure channel estimation in sixth generation (6G) multiple input multiple output (MIMO) systems, where conventional channel estimation schemes fail to adapt to non-linear propagation environment and are susceptible to adversarial attacks. In this context, the proposed HTCN model is developed through transfer learning to expedite the feature extraction and reduce training time, and defensive distillation to improve resistance against adversarial perturbations. The methodology involves training the HTCN model on a large dataset obtained by generating channel matrices in MATLAB’s 5G toolbox and testing under different adversarial environments. Experimental results demonstrate the HTCN model's low MSE (mean squared error) values of 0.0256 for training and 0.0281 for testing which implies high accuracy and generalization. Furthermore, the model demonstrates significant robustness against adversarial attacks, where the MSE only increased marginally. As a result, this research contributes to the state of knowledge in channel estimation in next-generation wireless networks by providing a secure and efficient solution to both the performance and security challenges.</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 transfer convolutional networks and defensive distillation for secure and energy efficient channel estimation in MIMO systems

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Venkata Subbamma Thippareddy,
  • Arvind Chakrapani,
  • Srihari Kannan

摘要

In this research, we propose a novel hybrid transfer convolutional network (HTCN) model for efficient and secure channel estimation in sixth generation (6G) multiple input multiple output (MIMO) systems, where conventional channel estimation schemes fail to adapt to non-linear propagation environment and are susceptible to adversarial attacks. In this context, the proposed HTCN model is developed through transfer learning to expedite the feature extraction and reduce training time, and defensive distillation to improve resistance against adversarial perturbations. The methodology involves training the HTCN model on a large dataset obtained by generating channel matrices in MATLAB’s 5G toolbox and testing under different adversarial environments. Experimental results demonstrate the HTCN model's low MSE (mean squared error) values of 0.0256 for training and 0.0281 for testing which implies high accuracy and generalization. Furthermore, the model demonstrates significant robustness against adversarial attacks, where the MSE only increased marginally. As a result, this research contributes to the state of knowledge in channel estimation in next-generation wireless networks by providing a secure and efficient solution to both the performance and security challenges.