Real-Time Machine Learning Incorporating Radio-Frequency Fingerprint Augmentation for Secure Wireless Communication
Abstract
Project Objectives: Networked devices collect and transmit data wirelessly for remote computing. In the near future, the number of wireless devices equipped with sensors for facilitating the automation of assorted information processing capabilities is expected to increase significantly. However, each wireless device interacting with other sensors, individuals, unmanned vehicles, or the environment has potential security vulnerability. These vulnerabilities include malicious target injection (spoofing), missed targets through broadband jamming, effects of intentional interference, or unauthorized replacement (counterfeiting). The critical need is to develop self-healing systems that can rapidly verify signal integrity based on unclonable radiofrequency (RF) security primitives while mitigating the impacts of false signals and malicious injection. To bridge the gap, this project will develop algorithm-hardware co-designed wireless communication interfaces that consider augmentable RF signatures, along with generative learning models to verify signal integrity and assure information security. Methods: The objective of this research will be attained by pursuing the following specific aims: Aim #1 RF signature augmentation incorporating combinatorial device primitives: Reconfigurable RF fingerprints will be co-designed with wireless transceivers driven by a generative model to augment physicallayer security primitives. The higher accuracy of authentication can be achieved by integrating the prominence of each transmitterÕs unique security primitives, which are updated in the temporal space to counter cyclostationary feature detection measures in the adversarial receivers. Aim #2 A brain-inspired generative model for RF signature generation and authentication: Self-supervised learning methods will be utilized to encode received signals to an encoded map using variational autoencoders, which attempt to extract signal features and reconstruct the original device-selection bitstream map. The reconstruction error will be used to assess the confidence level of the input signal so that only signals above a certain confidence threshold will be authenticated. Aim #3 Rapid integrity verification with cross-checking and hash functions: The tree hash algorithms will be developed to efficiently perform hash functions on the executed results for integrity verification. An extended output layer with dynamically permuted classes is added to the Bayesian inference model to cross check the results. Such self-protection mechanism will enable trusted machine intelligence. The Significance of the Proposed Effort to the Advancement of Scientific Knowledge : This research will establish a holistic approach that integrates a brain-inspired generative model and combinatorial randomness to rapidly perform device authentication and generate high-quality timestamped signatures using physical-layer security primitives to ensure trustworthy communication in extreme-complex battlefield environments. The situation-aware systems can adaptively augment and update the signatures in the spatial-temporal space to dynamically establish and maintain its capability under diversified situations. Lightweight machine learning modules with self-protection mechanisms that can counter adversarial manipulation and persistent attacks will allow operations under resource-constrained conditions. If successful, this research will enable secure and assured wireless communication with integrated authentication measures. The results are expected to facilitate a paradigm shift of how authentication and trustworthiness can be ensured to demonstrate the trust in dynamic wireless environments.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 09, 2023
- Source ID
- W911NF2310073
Entities
People
- Vanessa Chen
Organizations
- Army Contracting Command
- Carnegie Mellon University
- United States Army