A New RF-PUF Based Authentication of Internet of Things Using Random Forest Classification

This paper presents a novel RF-PUF based authentication framework which exploits the intrinsic non-idealities in physical characteristic of a device/medium to generate a unique identity for wireless nodes. It also takes the advantage of Random Forest classification to securely identify the sender nodes based on their unique features extracting from already-existing modules in the receiver side. In contrast to the neural network-based schemes, our proposed approach incurs lower design complexity and overheads, while it no longer needs a large amount of preparatory and preprocessing works related to the learning process and adjusting the network parameters. Thus, the overall runtime required to preparing and testing of network is drastically lessened. The experimental results show that the proposed scheme can reach to 100% accuracy in the identification of 225 nodes when a forest network with 100 trees and depth of 20 is developed, posing a negligible overhead on the receiver side. This high accuracy can be nearly achieved even in the presence of channel variations as our approach has less sensitivity to environmental conditions.