الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 65 |  |  |
| | | from | 65 | | |
Abstract
Cognitive Radio Network (CRN) can be thought as an emerging topic to tackle the problem of spectrum underutilization and scarcity. One of the main challenges in CRNs is the detection and identification of the nearby mobile users. To achieve full situation awareness, CRs must perform spectrum sensing to detect the existing primary users (PUs). Next, a refined analysis is performed to classify the detected users. Cooperative spectrum sensing scheme among mobile users can be used to determine the usage profile of wide spectrum bands in a large geographical region. In a large mobile crowdsensing environment, the key step is to assign the sensing tasks among mobile users to maximize the spectrum sensing performance while reducing the cost incurred by the mobile users during the sensing process. Moreover, by classifying the modulation schemes used in surrounding transmissions, a secondary user (SU) can identify the existing users in the system and adjust his/her transmission parameters accordingly. In this work, we propose two genetic algorithm-based approaches to solve the NP-hard problem of spectrum sensing task assignment among mobile users. The first algorithm uses a centralized genetic algorithm (GA) scheme to maximize the spectrum sensing utility function. The second algorithm uses an island genetic algorithm to assign the sensing tasks among mobile users in a distributive way. Simulation results show that both algorithms achieve comparable spectrum utility measure to the one obtained by running the recently proposed particle swarm optimization and greedy approximation algorithms while reducing the running time of the algorithm by a significant factor. In addition, the island algorithm massively outperforms both algorithms in the running time by running the algorithm independently at each sensing location and exchanging the necessary information for the overlapping locations, removing the bottleneck of having a central spectrum profiling unit to assign the sensing tasks among mobile users. Then, we propose a multi-task learning (MTL) approach to recognize the modulation scheme used among a specific set of analog and digital modulations. This approach uses a deep convolutional neural network (CNN) to extract the necessary features in order to classify the different modulation schemes. The MTL is used to separately train the modulation classes that normally cause a considerable confusion and therefore improve the overall classification accuracy. Our results on the RadioML dataset show that the suggested architecture achieves higher overall classification accuracy compared to the recently proposed Convolutional, Long Short Term Memory (LSTM), Deep Neural Network (CLDNN). Our classification accuracy of 86.97% at 18 dB SNR outperforms the state-of-the-art with 5% relative improvement.