VeriPHY: Physical Layer Signal Authentication for Wireless Communication in 5G Environments

TL;DR

VeriPHY introduces a deep learning-based physical layer authentication method for 5G that embeds and detects unique device signatures within wireless signals, achieving high accuracy and stealthiness.

Contribution

The paper presents VeriPHY, a novel deep learning approach that embeds pseudo-random signatures in 5G signals for secure device authentication, ensuring high accuracy and low detectability.

Findings

Achieves 93-100% signature identification accuracy

Maintains low false positive rates and 28 ms inference time

Supports stealth signature generation indistinguishable from normal signals

Abstract

Physical layer authentication (PLA) uses inherent characteristics of the communication medium to provide secure and efficient authentication in wireless networks, bypassing the need for traditional cryptographic methods. With advancements in deep learning, PLA has become a widely adopted technique for its accuracy and reliability. In this paper, we introduce VeriPHY, a novel deep learning-based PLA solution for 5G networks, which enables unique device identification by embedding signatures within wireless I/Q transmissions using steganography. VeriPHY continuously generates pseudo-random signatures by sampling from Gaussian Mixture Models whose distribution is carefully varied to ensure signature uniqueness and stealthiness over time, and then embeds the newly generated signatures over I/Q samples transmitted by users to the 5G gNB. Utilizing deep neural networks, VeriPHY identifies and authenticates users based on these embedded signatures. VeriPHY achieves high precision, identifying unique signatures between 93% and 100% with low false positive rates and an inference time of 28 ms when signatures are updated every 20 ms. Additionally, we also demonstrate a stealth generation mode where signatures are generated in a way that makes them virtually indistinguishable from unaltered 5G signals while maintaining over 93% detection accuracy.