Real-time Wireless Transmitter Authorization: Adapting to Dynamic Authorized Sets with Information Retrieval

As the Internet of Things (IoT) continues to grow, ensuring the security of\nsystems that rely on wireless IoT devices has become critically important. Deep\nlearning-based passive physical layer transmitter authorization systems have\nbeen introduced recently for this purpose, as they accommodate the limited\ncomputational and power budget of such devices. These systems have been shown\nto offer excellent outlier detection accuracies when trained and tested on a\nfixed authorized transmitter set. However in a real-life deployment, a need may\narise for transmitters to be added and removed as the authorized set of\ntransmitters changes. In such cases, the system could experience long\ndown-times, as retraining the underlying deep learning model is often a\ntime-consuming process. In this paper, we draw inspiration from information\nretrieval to address this problem: by utilizing feature vectors as RF\nfingerprints, we first demonstrate that training could be simplified to\nindexing those feature vectors into a database using locality sensitive hashing\n(LSH). Then we show that approximate nearest neighbor search could be performed\non the database to perform transmitter authorization that matches the accuracy\nof deep learning models, while allowing for more than 100x faster retraining.\nFurthermore, dimensionality reduction techniques are used on the feature\nvectors to show that the authorization latency of our technique could be\nreduced to approach that of traditional deep learning-based systems.\n