Combined Optimization of Feature Reduction and Classification for Radiometric Identification

• Published in: IEEE signal processing letters 2017-05, Vol.24 (5), p.584-588
• Main Authors: Jia, Yongqiang, Ma, Junhu, Gan, Lu
• Format: Article
• Language: English
• Subjects: Dimensional reduction; Feature extraction; Gallium nitride; linear regression classifier (LRC); Modulation; Mutual information; Optimization; quadratic mutual information (QMI); Radio frequency; radio frequency (RF) fingerprint; radiometric identification; Radiometry
• ISSN: 1070-9908; 1558-2361
• DOI: 10.1109/LSP.2017.2683523

Recently, dimensionality reduction for radiometric identification has attracted more attention. Previous research works generally considered dimensionality reduction and radio fingerprint classification separately, which resulted in poor performance. The reason is that the feature set after dimensionality reductions may not be suitable for classifier. In this letter, a new radiometric identification method based on combined optimization of dimensionality reduction and fingerprint classification is presented. The proposed method attempts to find an optimal dimension-reducing projection matrix by minimizing the classification error and maximizing the quadratic mutual information between the reduced low-dimensional features and the class label simultaneously. Since both the uncertainty of the true class label of the reduced low-dimensional features and the classification error of the classifier are considered, the proposed method can obtain better results in radiometric identification applications. Experiments on real data sets demonstrate that the proposed method not only outperforms the other methods with a higher accuracy of radiometric identification, but also has a better robustness against noises. The experimental results indicate that the proposed method can identify six emitters made by three different manufacturers with the classification accuracy over 95.08%.