Computational Complexity Reduce for the Machine Learning Based RFF via Dimension Reduction
Download as PDF
Wei Liu, Xinchen Xu, Haitao Wu, Fuquan Huang, Feiyi Xie, Honghao Liang, Jianfang Song, Hong Wen
In recent years, massive terminal connection and data transmission require 5G fast and high security access becoming very challenging. Fast and low cost authentication methods are neccessary. Radio frequency ﬁngerprint (RFF) as a non-cryptographic authentication method that is based on devices’ hardware information is an asymmetric authentication method under the edge computing system and almost all the computing load will be beared by the edge centre, which means that this method is low cost for the terminals and may meet the requirements of high speed, low power consumption and high security of 5G network. The existing best research on RFF authentication method is the method of RFF identification based on machine learning, which presenst high computing burden to the edge centre. In order to solve the problem of high computational complexity and time consuming, a principal component analysis (PCA) algorithm based on multi-resolution analysis and ReliefF was proposed. Experiments show that the scheme can achieve better recognition performance with lower complexity. This paper first introduces the principles of ReliefF and PCA algorithms, then introduces an improved algorithm based on multi-resolution analysis and reliefF algorithm, and proposes a principal component analysis algorithm based on the reliability of symbol features and sample selection. Finally, based on the data set analysis, the high recognition rate and low signal-to-noise ratio after data reduction are verified.
Complexity; Dimension Reduction; Identification; Machine Learning; RFF