Machine Learning for RF Fingerprinting Extraction . . .
201
Constellation side length
1.05
1.1
1.15
1.2
1.25
1.3
Constellation
diagonal length
1
1.005
1.01
1.015
1.02
1.025
Fig. 10. View of constellation side length and diagonal length with SNR = 9.07 dB
and constellation angle as input features. Figure 10 is the front view of Fig. 6,
which presents the dot plot using constellation side length and diagonal length
as input features.
Table 3 records the average accuracy when multiple features are considered as
the classification features with SNR = 9.07 dB and three classifiers. As shown in
the table, the classification performance of three features combination performs
better than two features combination. Compared with single feature input in
Table 2, the performance superiority of bagged tree achieves better stability than
other classification.
Table 3. Classification rate of several types of features for QPSK with SNR = 9.07
dB and three classifiers
QPSK
Bagged tree (%) Weighted KNN Fine Gaussian SVM
Box dimension and
fractal dimension
97.2
94.7
85.6
Constellation side
length and
constellation
diagonal length
97.5
95.8
88.9
Phase noise
spectrum and
constellation angle
97.7
96.7
90.9
Box dimension,
fractal dimension,
and phase noise
spectrum
98.5
98.3
95.5
Constellation side
length constellation
diagonal length and
constellation angle
98.2
98.0
94.9