A physical layer authentication mechanism for IoT devices

When Internet of Things (IoT) nodes access the network through wireless channels, the network is vulnerable to spoofing attacks and the Sybil attack. However, the connection of massive devices in IoT makes it difficult to manage and distribute keys, thus limiting the application of traditional high-level authentication schemes. Compared with the high-level authentication scheme, the physical layer authentication scheme realizes the lightweight authentication of users by comparing the wireless channel characteristics of adjacent packets. However, traditional physical layer authentication schemes still adopt the one-to-one authentication method, which will consume numerous network resources in the face of large-scale IoT node access authentication. In order to realize the secure access authentication of IoT nodes and regional intrusion detection with low resource consumption, we propose a physical layer authentication mechanism based on convolution neural network (CNN), which uses the deep characteristics of channel state information (CSI) to identify sending nodes in different locations. Specifically, we obtain the instantaneous CSI data of IoT sending nodes at different locations in the pre-set area, and then feed them into CNN for training to procure a model for IoT node authentication. With its powerful ability of data analysis and feature extraction, CNN can extract deep Spatio-temporal environment features of CSI data and bind them with node identities. Accordingly, an authentication mechanism which can distinguish the identity types of IoT nodes located in different positions is established to authenticate the identity of unknown nodes when they break into the pre-set area. Experimental results show that this authentication mechanism can still achieve 94.7% authentication accuracy in the case of a low signal-to-noise ratio (SNR) of 0 dB, which means a significant improvement in authentication accuracy and robustness.