ADVERSARIAL MACHINE LEARNING FOR PHYSICAL-LAYER AUTHENTICATION

In this dissertation, we propose the use of adversarial machine learning to characterize wireless radio transmitters for the purpose of physical-layer authentication. Wireless communication systems are quickly evolving to take advantage of autonomous networking for applications such as 5th generation mobile networks, Internet of Things, and vehicular-to-everything technologies. Robust and efficient network security mechanisms are necessary to protect the authenticity of the data and safeguard the integrity of the greater interconnected network. To this end, we leverage unique channel-dependent differences in received transmissions, known as channel state information (CSI), to make authentication decisions with machine learning algorithms. Many physical-layer authentication techniques are not effective when used in the presence of nefarious users who are able to spoof the underlying physical-layer authentication traits. Our approach uses adversarial learning to counter malicious actions such as spoofing against legitimate transmitter CSI, an already difficult characteristic to emulate. We simulated various radio frequency channel environments and our results indicate that the use of machine learning techniques can produce high authentication accuracy.Commander, United States NavyApproved for public release. Distribution is unlimited