IEEE - Aerospace and Electronic Systems - June 2022 - 36

Cognitive Radio for Aeronautical Mobile Telemetry: A Machine Learning-Based Approach
Figure 5.
Confusion matrix for modulation recognition across all SNRs.
RESULTS
Our experiments showed that ofthe classifiers tested, the quadratic
discriminant classifier performs the best in this application.
The quadratic discriminant classifier finds the class k,
whichmaximizes a quadratic function based on the covariance
matrix for the class k, the mean for the class k, and the percentage
ofsamples ofclass k in the training set. In contrast, a linear
discriminant classifier is very similar but assumes that the
covariance matrix for each class is identical. Since quadratic
discriminant analysis estimates the covariance matrix separately
for each class, it tends to fit the data better than LDA for
our problem.
Results for modulation recognition using the quadratic
discriminant classifier are plotted in Figure 4. For PCM/FM
and ARTM CPM, prediction accuracy is 100% across all
SNRs tested. For SNRs less than 6 dB, the performance for
the case when SOQPSK-TG was transmitted degrades.
Remarkably, these resultswere obtained using only themiddle
133 samples from each periodogram. The confusion matrix is
illustrated in Figure 5 and shows that all classification errors
weremisclassifying SOQPSK-TGas ARTMCPM.
For bitrate estimation, of the methods tested, the Gaussian
process regression model achieves the best performance.
The normalized root-mean-square (rms) errors for the bitrate
at each SNR are plotted in Figure 6. At low SNRs, the normalized
rms estimator error decreases with increasing SNR.
At higher SNR, the rate of improvement diminishes as the
additive noise ceases to be the dominant factor in estimator
performance. This model was trained on PSDs computed on
signals at 91 different bitrates. In the future, collecting data at
more bitrates could improve performance.
CONCLUSION
The three IRIG-106 standard CPM modulations can be identified
with excellent accuracy for SNRs relevant to
36
Figure 6.
Bitrate estimator: normalized rms error versus SNR.
aeronautical mobile telemetry. The modulation recognition
algorithm is based on a quadratic discriminant classifier
trained on the PSD. The accuracy is 100% across the entire
range ofSNRs for PCM/FM and ARTM CPM modulations.
For SOQPSK-TG, the accuracy is 100% above 6 dB. Because
the PSD is simple to compute, this modulation classification
is less resource intensive than approaches requiring higher
order statistical time domain features of the sampled data. In
addition, for a problem with three classes ofinterest, the quadratic
discriminant classifier requires less computational
overhead than convolutional neural networks or SVMs. To
solve the problem of estimating bitrate, a Gaussian process
regression model is tested. The estimation error is reasonably
small, even for our lowest SNRs ofinterest below5 dB.
The results demonstrate that trained machine learning
models could be implemented in a cognitive radio to perform
automatic modulation recognition and bitrate estimation.
What is noteworthy is that this performance is achieved using
low-complexity PSD estimates based on a relatively short
fast Fourier transform (FFT) length and using only the middle
133 periodogram samples. Given the ability of this simple
algorithm to successfully perform the task, there was little
motivation to explore more sophisticated techniques with
higher computational complexity.
In conclusion, machine learning can be successfully
applied to create a cognitive radio for use in aeronautical
telemetry. The approach is blind in the sense that the
bitrate is not required to determine the modulation type.
The approach has low complexity, making it suitable for a
hand-held device for use by flight line technicians to confirm
proper settings on a multimode transmitter.
REFERENCES
[1] Telemetry Standards, Range Commanders Council Telemetry
Group Std. IRIG106-20, Jul. 2020. [Online]. Available:
https://www.wsmr.army.mil/RCCsite/Pages/Publications.aspx
IEEE A&E SYSTEMS MAGAZINE
JUNE 2022
https://www.wsmr.army.mil/RCCsite/Pages/Publications.aspx
IEEE - Aerospace and Electronic Systems - June 2022

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - June 2022
