الفهرس 
Civilian ApplicationsOnly 14 pages are availabe for public view
 |  | from | 181 |  |  |
| | | from | 181 | | |
Abstract
Automatic Modulation Classification (AMC) is a technique to classify the modulated
signal by observing its received signal features without any prior knowledge of the
intercepted signal. It is the intermediate step between signal detection and demodulation. It is
a very important task for cognitive radio communication. This, in turn, removes the need for
sending end-to-end handshaking information between the transmitter and the receiver.
Therefore, spectrum efficiency is improved as no modulation information is needed in the
transmitted signal frame. The AMC techniques proposed in the literature are classified into
traditional techniques, which include Decision-Theoretic (DT) techniques, Feature-Based
(FB) techniques and advanced techniques that depend on deep learning.
In this thesis, we focus on advanced techniques. Deep Learning (DL) is implemented to
improve the accuracy of the AMC, due to its high capacity for representing features. Two
proposed methods for AMC are presented based on Convolutional Neural Networks (CNNs).
These networks are considered as Deep Learning (DL) tools. The classification is performed
in the presence of Additive White Gaussian Noise (AWGN) and Rayleigh fading, which is an
important task in several wireless communication applications.
In the first proposal, we present an accurate strategy for classification which combines
Gabor filtering of constellation diagrams, thresholding, and then a DL structure based on
basic CNN, AlexNet, or Residual Neural Network (ResNet 50). The Gabor filter can
effectively extract spatial information including edges and textures from constellation
diagrams. In terms of classification accuracy, the proposed AMC method achieves
competitive results.
In the second proposal, we adopt a strategy based on image decimation and sharpening
and thresholding with the help convolutional filters as feature extraction tools with the DL
structure. The same basic CNN, AlexNet, and ResNet 50 are used in the AMC process. The
objective of decimation is to reduce the computation cost of the AMC.
For both methods, we work on seven modulation types, which are BPSK, 4QAM,
8PSK, 16PSK, 8QAM, 16QAM, and 32QAM over the range of Signal-to-Noise Ratio (SNR)
from -10 dB to 30 dB. The performed experiments reveal that the suggested proposals
guarantee remarkable classification accuracy over AWGN and Rayleigh fading channels.