Robust satellite antenna fingerprinting under degradation using recurrent neural network

Abstract

Antenna fingerprinting is critical for a range of physical-layer wireless security protocols to prevent eavesdropping. The fingerprinting process exploits manufacturing defects in the antenna that cause small imperfections in signal waveform, which are unique to each antenna and hence device identity. It is an established process for physical-layer wireless authentication with proven usage systems in terrestrial systems. The premise relies on accurate signal feature discovery from a large set of similar antennas and stable fingerprint patterns over the operational life of the antenna. However, in space, many low-cost satellite antennas suffer degradation from atomic oxygen (AO). This is particularly a problem for nano-satellites or impromptu temporary space antennas to establish an emergency link, both of which are designed to operate for a short time span and are currently not always afforded protective coating. Current antenna fingerprinting techniques only use Support Vector Machine (SVM) and Convolutional Neural Networks (CNNs) to take a snap-shot fingerprint before degradation, and hence fail to capture temporal variations due to degradation. Here, we show how we can perform robust antenna fingerprinting (99.34% accuracy) for up to 198 days under intense AO degradation damage using Recurrent Neural Networks (RNNs). We compare our RNN results with CNNs and SVM techniques using different signal features and for different Low-Earth Orbit (LEO) satellite scenarios. We believe this initial research can be further improved and has real-world impact on physical-layer security of short-term nano-satellite antennas in space.