关键词: 无线安全, 伪AP检测, 信道状态信息, 无线设备指纹, Siamese网络

Abstract: With the popularity of mobile Internet, a large number of devices are connected to the Internet through wireless access points. However, wireless security has been one of the hot research topics because of the frequent wireless network attacks. Currently, rogue AP attack is one of the popular methods of wireless network attacks. The current methods for detecting rogue AP have flaws such as features being easily forged and high deployment costs. A new method for detecting rogue AP based on wireless device fingerprinting was proposed. The features used in this method can hardly be forged, and the training data only need the data of legitimate AP for rogue AP detection. CSI data of legitimate AP were collected, and wireless device fingerprints based on CSI phase information were extracted after data preprocessing. Then, sample pairs were generated based on the extracted device fingerprints, the Siamese network was trained by the sample pairs and AP fingerprint database was created. During the detection phase, the AP to be tested was compared with the corresponding fingerprints in the fingerprint database by the Siamese network to detect rogue AP. The experiments compared the accuracy rate of legitimate AP test set classification and rogue AP detection rate of Siamese network based on traditional deep neural network structure and convolutional neural network structure under different time windows. The experimental results showed that the method proposed in this paper has good prospects, and the Siamese network based on convolutional neural network has a higher detection rate.

Key words: wireless security, rogue AP detection, channel state information, wireless device fingerprinting, Siamese network