One-shot Generative Distribution Matching for Augmented RF-based UAV Identification

TL;DR

The paper tackles UAV identification in constrained RF environments using RF fingerprinting and proposes one-shot generative augmentation of transformed RF signals. It provides a theoretical bound on distributional distance via Wasserstein metrics and validates it with GPDM, a one-shot, patch-based method, showing robust improvements in low-data regimes over conditional GANs and VAEs. Empirical results on a DFT-based UAV RF dataset demonstrate that GPDM yields classification performance between baseline and ground truth while CGAN/CVAE struggle with limited data and outliers. This work advances low-data learning for RF sequences and offers a practical framework with public data and code for augmented UAV identification in challenging RF environments.

Abstract

This work addresses the challenge of identifying Unmanned Aerial Vehicles (UAV) using radiofrequency (RF) fingerprinting in limited RF environments. The complexity and variability of RF signals, influenced by environmental interference and hardware imperfections, often render traditional RF-based identification methods ineffective. To address these complications, the study introduces the rigorous use of one-shot generative methods for augmenting transformed RF signals, offering a significant improvement in UAV identification. This approach shows promise in low-data regimes, outperforming deep generative methods like conditional generative adversarial networks (GANs) and variational auto-encoders (VAEs). The paper provides a theoretical guarantee for the effectiveness of one-shot generative models in augmenting limited data, setting a precedent for their application in limited RF environments. This research contributes to learning techniques in low-data regime scenarios, which may include atypical complex sequences beyond images and videos. The code and links to datasets used in this study are available at https://github.com/amir-kazemi/uav-rf-id.

Figures (7)

• Figure 1: Sequence $\mathbf{X}$, subsequence $\mathbf{SX}$, and projection $\mathbf{S}^T\mathbf{SX}$ for $d=4, d'=2$: (a) $\tilde{\Omega}_S=\Omega_S$ is the set of all subsequences of length $d'=2$ which includes six subsequences, (b) and (c) $\tilde{\Omega}_S\subset \Omega_S$ is the set of substrings of length $d'=2$.
• Figure 2: A one-shot generative model takes in real sequences $\mathbf X$ and outputs synthetic sequences $G(\mathbf X, \mathbf Z)$. The distance metric employed is optimal transport, specifically the Wasserstein distance, applied between the distributions of sequences and subsequences, as detailed in the accompanying formulation.
• Figure 3: An $n\times n$ matrix $\mathbf Y$ with $n' \times n'$ substrings, vectorized as sequence $\mathbf X$ with projections as $\mathbf S^T \mathbf{SX}$, where $n=3$ and $n'=2$.
• Figure 4: Zero-padded $\mathbf Y$ is $(n+2n'-2) \times (n+2n'-2)$, where $n=3$ and $n'=2$. The central $n\times n$ elements are repeated $n'\times n'$ times each. The repetition of other elements can be substituted arbitrarily (marked by ?), as corresponding elements in $\mathbf Y_{padded}$ are zero in both real and generated data and do not contribute to optimal transport.
• Figure 5: Mean, as well as mean plus ten times the standard deviation of $W/\delta$ (shaded area), calculated over five folds for different scenarios to verify its compliance with the theoretical upper limit in Eq. (\ref{['2Dbound2']}), where P denote the percentage of utilized data.

Theorems & Definitions (12)

• Definition 1
• Definition 2: Distance Metric for Distributions
• Definition 3
• Lemma 1
• proof
• Theorem 1
• proof
• Corollary 1
• proof
• Corollary 2