TK1111 : Optimal Resource Allocation in Hybrid Optical-Wireless Networks
Thesis > Central Library of Shahrood University > Electrical Engineering > MSc > 2026
Authors:
Abstarct: Abstract
This study addresses the problem of optimal resource allocation in a hybrid wireless–optical network baxsed on RF and FSO lixnks. The objective is to design and evaluate a frxamework that jointly performs active-lixnk selection, power allocation, and bandwidth allocation as a single decision under time-varying channel and traffic conditions. To this end, the FSO channel is modeled by accounting for atmospheric attenuation and turbulence, while the frxamework also allows the effects of pointing errors to be incorporated. The RF channel is modeled using Nakagami-m fading. The dynamic control problem is formulated as a Markov Decision Process (MDP) and solved using the Proximal Policy Optimization (PPO) algorithm. The state vector combines FSO/RF channel information with queue status, traffic conditions, and available resources, while the control action determines the lixnk selection and the normalized fractions of power and bandwidth.
The proposed frxamework is fully implemented in MATLAB. In addition to physical-laxyer validation, packet/queue behavior and Quality of Service (QoS) metrics are evaluated. The PPO policy is compared with baxseline methods, including FSO-only, RF-only, SNR-threshold-baxsed selection, and greedy rate selection. To ensure a fair comparison, common channel and traffic realizations are used, and the final statistical evaluation is performed over 20 independent seed pairs.
The final results show that the improved PPO achieves a throughput of 86.357 ± 0.345 Mbps, an average delay of 7.767 ± 0.027 ms, a 95th-percentile delay of 33.350 ± 0.275 ms, and a packet drop rate of 12.709 ± 0.193%. On average, this policy utilizes 58.42% of the available power and 62.69% control problem is formulated as a Markov Decision Process (MDP) and solved using the Proximal Policy Optimization (PPO) algorithm. The state vector combines FSO/RF channel information with queue status, traffic conditions, and available resources, while the control action determines the lixnk selection and the normalized fractions of power and bandwidth.
The proposed frxamework is fully implemented in MATLAB. In addition to physical-laxyer validation, packet/queue behavior and Quality of Service (QoS) metrics are evaluated. The PPO policy is compared with baxseline methods, including FSO-only, RF-only, SNR-threshold-baxsed selection, and greedy rate selection. To ensure a fair comparison, common channel and traffic realizations are used, and the final statistical evaluation is performed over 20 independent seed pairs.
The final results show that the improved PPO achieves a throughput of 86.357 ± 0.345 Mbps, an average delay of 7.767 ± 0.027 ms, a 95th-percentile delay of 33.350 ± 0.275 ms, and a packet drop rate of 12.709 ± 0.193%. On average, this policy utilizes 58.42% of the available power and 62.69% of the available bandwidth, while reducing the number of lixnk switches to approximately one event. Although the greedy rate method provides higher raw QoS performance, this improvement comes at the cost of full resource utilization and substantially more frequent switching. Therefore, the primary advantage of PPO in this study is not absolute superiority across all QoS metrics, but rather its ability to achieve a balance among QoS, resource efficiency, and switching stability.
Keywords:
#RF hybrid network #resource allocation #deep reinforcement learning #PPO #Quality of Service (QoS) #adaptive switching_ Keeping place: Central Library of Shahrood University
Visitor:
Visitor: