原文传递
Learning Robust Radio Frequency Fingerprints Using Deep Convolutional Neural Networks.
| 题名： | Learning Robust Radio Frequency Fingerprints Using Deep Convolutional Neural Networks. |
| 作者： | Gutierrez del Arroyo Perez, J. A. |
| 摘要： | Radio Frequency Fingerprinting (RFF) is the attribution of uniquely identifiable sig-nal distortions to emitters via Machine Learning (ML) classifiers. RFF is often pro-posed as an authentication mechanism for wireless device security, but techniques arelimited by fingerprint variability under different operational conditions. First, thiswork studies the effect of frequency channel for typical RFF techniques, which havepreviously only been evaluated using bursts from a single frequency channel withoutconsidering the effects of multi-channel operation. Performance characterization us-ing the multi-class Matthews Correlation Coefficient (MCC) revealed that operatingon frequency channels other than those used to train the models can lead to a dete-rioration in performance from MCC > 0.9 (excellent) down to MCC < 0.05 (randomguess), indicating that single-channel models should not be expected to maintainperformance in realistic multi-channel operation. A training data selection techniquewas proposed to create multi-channel models which outperform single-channel mod-els, improving the cross-channel average MCC from 0.657 to 0.957 and achieving fre-quency channel-agnostic performance. Second, this work introduced, developed, anddemonstrated the Fingerprint Extraction through Distortion Reconstruction (FEDR)process, a neural network-based approach for quantifying signal distortions. Coupledwith a simple Dense network, FEDR fingerprints were evaluated against four commonRFF techniques for N c = {5,10,15,25,50,100} unseen classes. The Dense networkwith FEDR fingerprints achieved best performance across all values of N c with MCCranging from 0.945 (N c = 5) to 0.746 (N c = 100), using nearly 73% fewer trainingparameters than the next-best Convolutional Neural Network. |
| 总页数： | 131 pages |
相关文献
- Transfer Learning in Convolutional Neural Networks for Fine-Grained Image Classification.
- Evaluation of Using Deterministic Heuristics to Accelerate Reinforcement Learning.
- Traffic Measurement and Vehicle Classification with Single Magnetic Sensor.
- Toward Automated Aerial Refueling: Automated Visual Aircraft Identification with Convolutional Neural Networks.
- Competitively Learned Attention Mechanism Prototypes for Network Intrusion Detection.
- Network Traffic Data Collection for Machine Learning Analysis.
检索历史
应用推荐
