Scholarlink

<p>With the evolution of sixth-generation (6G) wireless networks, a large number of edge intelligent devices are now connected to the Internet of Things (IoT). Owing to the open nature of wireless networks, the massive edge IoT devices need to continuously prevent spoofing and the intrusion of malicious IoT devices. The deep learning (DL)-based radio frequency fingerprint identification (RFFI) provides a promising zero-trust edge IoT security scheme by automatically extracting the radio frequency fingerprint (RFF) feature from the intrinsic hardware imperfections. To address the problems in the training overhead, data limitation, and scalability for the current DL-based RFFI, we considered an outdoor long-range (LoRa) edge intelligent network, where we combined the large language model (LLM) for the first time and proposed a BERT-LightRFFI framework to enhance zero-trust edge IoT security. Specifically, we pre-trained a BERT model with the unlabeled data via self-supervised learning and obtained a powerful RFF feature extractor. Then, we used the knowledge distillation to inherit the BERT learn-gene to the small BERT-Light model and then fine-tuned a classifier of the pre-trained BERT-Light model by using few-shot labeled wireless data. The time complexity, parameter quantities, and computational complexity were analyzed. In the experiments, we used a large-scale real-world LoRa dataset to evaluate the performance of the proposed framework and suggest some interesting insights. The results prove the proposed framework’s effectiveness, achieving an accuracy of 97.52% in the presence of multipath fading and Doppler shift, which is better than the previous benchmark methods.</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

Let RFF do the talking: large language model enabled lightweight RFFI for 6G edge intelligence

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Ning Gao,
  • Yi Liu,
  • Qifan Zhang,
  • Xiao Li,
  • Shi Jin

摘要

With the evolution of sixth-generation (6G) wireless networks, a large number of edge intelligent devices are now connected to the Internet of Things (IoT). Owing to the open nature of wireless networks, the massive edge IoT devices need to continuously prevent spoofing and the intrusion of malicious IoT devices. The deep learning (DL)-based radio frequency fingerprint identification (RFFI) provides a promising zero-trust edge IoT security scheme by automatically extracting the radio frequency fingerprint (RFF) feature from the intrinsic hardware imperfections. To address the problems in the training overhead, data limitation, and scalability for the current DL-based RFFI, we considered an outdoor long-range (LoRa) edge intelligent network, where we combined the large language model (LLM) for the first time and proposed a BERT-LightRFFI framework to enhance zero-trust edge IoT security. Specifically, we pre-trained a BERT model with the unlabeled data via self-supervised learning and obtained a powerful RFF feature extractor. Then, we used the knowledge distillation to inherit the BERT learn-gene to the small BERT-Light model and then fine-tuned a classifier of the pre-trained BERT-Light model by using few-shot labeled wireless data. The time complexity, parameter quantities, and computational complexity were analyzed. In the experiments, we used a large-scale real-world LoRa dataset to evaluate the performance of the proposed framework and suggest some interesting insights. The results prove the proposed framework’s effectiveness, achieving an accuracy of 97.52% in the presence of multipath fading and Doppler shift, which is better than the previous benchmark methods.