Abstract
The Internet of Things has attained rapid growth in the field of the device authentication. There are several research carried out in identifying the device to prevent intrusion of rogue devices in the network which thereby can access the private data of the users. Radio Frequency Fingerprint identification is one of the highly used device authentication techniques used. It takes into consideration transmitter hardware impairments. This RFFI framework extractor leverages the deep metric learning to train the extractors, which has a good capability to categorize and extract RFF data from the devices which were previously not seen. Wireless channels also have impact on the Extractors and is handles by exploiting channel independent feature and data augmentation technique. Different Devices and software were used to perform the experiments of the LORA devices.In this project we have presented a high-quality mage synthesis approach to evaluate the results of the extractors used. Denoising diffusion models is a latent model which is inspired from the nonequilibrium thermodynamics.
It’s a Markov Chain variational model inference to generate samples matching the data after finite amount of time. The transition is such models is when it learns to reverse the diffusion process. We have trained the model on different set of images to enhance its capabilities to improve the classification of different devices over the network.
The results obtained after the training and are used for classification and comparing it with the extractors 1 to examine how it perform. The results have successfully demonstrated that the proposed Diffusion model has shown improvements in some of the cases and achieved an accuracy of 98 to 98.90% in different scenarios. Thus, we can say that Denoising diffusion models can also be used in signal processing data for classification problem. The entire implementation of the project is developed in Python has been with the use of necessary libraries and algorithms.