الفهرس 
Chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 180 |  |  |
| | | from | 180 | | |
Abstract
This thesis is concerned with improving the spectrum utilization efficiency
and providing efficient spectrum sensing methods for improving the performance
of spectrum sensing in cognitive radio (CR) systems. Firstly, a new practical
cooperative spectrum sensing system is used for measuring the utilization of the
Wi-Fi 5 GHz band and switching to the 2.4 GHz Wi-Fi band. The practical system
is simulated. The simulation and measurement results are compared with previous
related measurements obtained in Singapore, Barcelona, North Dakota United
States of America (USA) and Germany.
Two proposed spectrum sensing models are presented in this thesis. The first
model is an efficient adaptive multistage spectrum sensing model for CR system.
The proposed model consists of an Energy Detection (ED) stage and a Wavelet
Denoising (WD) stage. The proposed model adopts only a single ED stage at high
SNR and two sequential stages (ED + WD) at low SNR. The proposed model
reduces the detection time as the second stage is activated when the SNR is low. It
achieves the probabilities required for detection and false alarm and achieves
higher sensing accuracy compared to other methods, even at low SNR values.
The second model treats the spectrum sensing as a classification based on a
deep learning Convolution Neural Network (CNN). This model works on the
spectrogram images of the received signals as the input of the CNN. It uses
various signal data and noise data at different low Primary User (PU) SNRs to
train the network. We conduct extensive experiments with different CNN layers to
verify the performance of the proposed model and reach the optimum number of
layers, which gives a high detection accuracy for PU signals at low SNRs.
The proposed model can distinguish between PU signals and noise after
training, and can determine the presence/absence of the PU signals with high accuracy. The simulation results show that the proposed model outperforms the
previous single-stage, and two-stage spectrum sensing methods and the previous
deep learning CNN models in terms of spectrum detection accuracy and spectrum
sensing time in the case of low SNR of (-5 to -20) dB.