72
Q. Wu et al.
Table 1. Prediction performance of three models
Performance
Model
SINR 0 db 4 db 8 db 12 db 16 db
LSTM
MAE 0.437 0.422 0.277 0.203 0.118
MAPE 0.796 0.950 0.623 0.456 0.278
BI-LSTM (noisy prediction)
MAE 0.565 0.413 0.267 0.168 0.109
MAPE 1.099 0.945 0.620 0.454 0.275
BI-LSTM (noiseless prediction) MAE 0.257 0.205 0.163 0.103 0.07
MAPE 0.592 0.548 0.361 0.248 0.146
Figure 4a shows the comparison of original signal and predictive signal based
on digital modulation when the SINR = 12 db. Similarly, Fig. 4b demonstrates
the case of FM modulation. We can find from the pictures that the difference
between the predictive signal and the original noiseless signal in the normal part
is much smaller than in the interference-containing part. So the difference of
predictive signal and original signal can be used for interference detection.
(a)
(b)
Fig. 4. a is the difference between predictive signal and original signal (digital modulation), and b is the difference between predictive signal and original signal (analog
modulation)
4 Interference Detection
4.1 Classifier for Interference Detection
The features calculated by Sect. 3 can be used for interference detection. In
order to overcome the factor of sample imbalance, select support vector data
description (SVDD) here.