CSI-Based physical layer authentication is a promising candidate to achieve fast and lightweight authentication for wireless communication. However, the current methods usually cannot achieve initial authentication and are susceptible to channel noise. Besides, the current learning-aided physical layer authentication usually requires illegitimate channel state information (CSI) samples that are difficult to obtain. This letter proposes a newly deep-CSI-based authentication scheme to solve the above problems. We map CSI to a device’s location and further to its authenticated identity via deep learning in a static environment. Therefore, the proposed scheme does not require the cooperation of cryptography-based authentication to achieve initial authentication. The deep-learning-based authenticator with a confidence score branch is designed to learn the mapping relationship between the CSI and the identity. The confidence score branch can output a scalar that indicates whether the device is legitimate or not in the absence of illegitimate device CSI samples. CSI data are constructed as CSI images and implementation tricks are proposed to train the authenticator. Experiment results show that the authenticator performs well on all metrics and is robust to channel estimation errors.

中文翻译：

---

通过深度学习进行基于 CSI 的物理层身份验证


基于 CSI 的物理层身份验证是实现快速、轻量级无线通信身份验证的有希望的候选者。然而，当前的方法通常无法实现初始认证并且容易受到信道噪声的影响。此外，当前的学习辅助物理层认证通常需要难以获取的非法信道状态信息（CSI）样本。本文提出了一种新的基于深度CSI的认证方案来解决上述问题。我们将 CSI 映射到设备的位置，并通过静态环境中的深度学习进一步映射到其经过身份验证的身份。因此，所提出的方案不需要基于密码学的认证的配合来实现初始认证。具有置信度分支的基于深度学习的身份验证器旨在学习 CSI 和身份之间的映射关系。置信度得分分支可以输出一个标量，该标量指示在不存在非法设备 CSI 样本的情况下设备是否合法。 CSI 数据被构造为 CSI 图像，并提出了训练验证器的实现技巧。实验结果表明，验证器在所有指标上都表现良好，并且对信道估计误差具有鲁棒性。