International Journal of Science, Engineering and TechnologyAuthors: Dr.M.Vijaya Maheswari, Dr. Mohammad Shahbaz Khan
Abstract: The growth of Internet of Things (IoT) has created some security issues, especially in the authentication process that is often associated with very limited computation capability. This paper suggests a lightweight machine learning approach for IoT device authentication based on edge intelligence. The suggested approach utilizes small scale convolutional neural networks that run on edge gateways to generate unique features based on the information obtained from channel state information and radio frequency fingerprinting. No cloud connection is needed due to this continuous authentication approach. Quantization and pruning are used to make sure the model is lightweight and energy efficient while having good authentication capabilities. Experimental validation on a testbed made of Raspberry Pi shows that good performance of authentication can be achieved using a lightweight model and fast inference per cycle of authentication. Comparison to previous cryptographic and machine learning methods proves that better performance can be achieved in terms of both energy usage and protection from attack like replay and impersonation.
DOI: http://doi.org/10.5281/zenodo.23095607
