In this thesis, we consider a cognitive radio (CR) network where the primary network's feedback information is utilized to design access schemes for the secondary network, to exploit the underutilized primary spectrum resources. Secondary users (SUs) identify the spectrum opportunities by sensing the spectrum for primary users (PUs) activities and by listening to the PUs feedback. The feedback signals monitored in this research work are the channel quality indicator (CQI) and automatic repeat request (ARQ) available in the PUs network. For detecting the PUs activities, SUs employ hard/soft energy sensing. The secondary access decisions are optimized to maximize the SUs service rate while maintaining the PUs queues' stability. The proposed systems are modeled as multi-dimensional Markov chains (MCs) that capture the number of packets in the PUs queues and the state of the one/two observed PUs' feedback signals. The performance of the proposed systems are evaluated by deriving the SUs service rate and the average PUs packet delay. We compare the performance of the proposed systems with other baseline systems utilizing different types of PUs' feedback signals. Results reveal the improvement in the SUs service rate and the PUs' delay of the proposed systems compared to the baseline systems. This improvement is mainly due to the fact that in our proposed system SUs have access to extra information, in terms of PUs feedback, as compared to other systems. Therefore, SUs in our proposed systems can have better inference on the PUs' activities; thus more collisions between the PUs and the SUs can be avoided, resulting in significant performance gains in terms of SUs' throughput and PUs' average delay. Finally, we propose a multi-layer perceptron (MLP) Q-learning based spectrum access scheme for cognitive radio networks. In the proposed scheme, SUs overhear the ARQ feedback available in the PUs' network and exploit it to learn the PUs behavior. Since the SUs observe only the PUs' ARQ feedback and have no information about the PUs packet arrival rates or the states of their queues, the system is modeled as a partially observable Markov decision process (POMDP). The proposed MLP Q-learning access scheme is used to solve this POMDP and find the best SUs' actions (channel access probabilities) based on the observed PUs' ARQ feedback and past experiences. The performance of the proposed scheme is shown to be on par with that of other feedback-based access schemes, with the added strength of only having partial information about the PUs and the primary network.


School of Sciences and Engineering


Electronics & Communications Engineering Department

Degree Name

PhD in Engineering

Graduation Date

Winter 1-1-2020

Submission Date


First Advisor

Seddik, Karim

Committee Member 1

Gadallah, Yasser

Committee Member 2

Abdelazeem, Sherif

Committee Member 3

Elsabrouty, Maha


137 leaves

Document Type

Doctoral Dissertation

Institutional Review Board (IRB) Approval

Not necessary for this item