TinyAP: A Smart Access Point to detect KRACK on WPA2 Handshake in Wi-Fi using TinyML

Wi-Fi Protected Access 2 (WPA2) security protocol suite has been challenged by a recently discovered Key Reinstallation Attack (KRACK) that forces a supplicant to use an all zero-key for encryption. This has triggered the proposal of a new security protocol suite, WPA3. However, the existing Wi-Fi devices rely only on WPA2, leaving them vulnerable to KRACK. In this paper, we propose a TinyML-based solution, called TinyAP, to address the problem of detecting KRACK at the Wi-Fi router itself. Specifically, we have built up a framework of TinyAP where a trained TinyML model can be executed in a micro-controller for inference, and the micro-controller can be used as a plugin to access point (AP). We have used an open-sourced Wi-Fi network dataset containing KRACK trace and trained two-class classifiers using Multi Layer Perceptron (MLP) models. Then, we have converted these trained models into TinyML models using Tensor FLow Lite (TFLite) Micro libraries to be considered as TinyAP models. We have compared the performance of the TinyAP models in terms of their memory consumption, and size with the counterpart MLP models. The TinyAP models show a 21x reduction in memory usage for prediction and a 9x reduction in the trained model size on average as compared to their counterpart MLP models. The TinyAP model achieved an average accuracy of 98.63% in detecting KRACK attack.