Architectures for Automatic Classification ofModulation in an Environment with DopplerFading and Impulsive Noise
Automatic Modulation Classification, Deep Learning, Cyclostationary Analysis
The automatic classification of modulation (AMC) allows identifying the modulationof the received signal, being a key part for the development of cognitive radio devices thatadapt the type of modulation to the characteristics of the communication environment. Forthis reason, several researches on AMC have been developed. And one trend is the useof architectures based on deep learning. While powerful feature descriptors have beendeveloped over time. Thus, we propose a comparison between methods based on deeplearning and on cyclostationary descriptors, in an environment with Doppler fading, withnormalized Doppler frequency fromFdTS=0.002, and impulsive noise, characterized bya distributionα-stable, withαparameter assuming values in the set{1.3,1.7,2.0}, andconsidering a Geometric Signal-to-Noise Ratio (GSNR) varying from 0 to 15 dB.