AI Based Realistic Multi-Hop Wireless Simulation

1 online resource (51 pages) : PDF

2021

University of North Carolina at Charlotte

Wireless multi-hop networks and software-defined networking (SDN) are emerging technologies in wireless communications for deploying cost-efficient programmable network backbones. However, the physical testbeds do not provide scalable environments, accelerated development, and testing. Simulations are a cost-effective approach to overcome the bottlenecks of the physical testbed. Furthermore, Wireless multi-hop networks tend to have a high environmental impact which leads to significant performance issues such as low SNR and throughput in single-channel links. Adopting a multi-channel multi-radio(MCMR) setup can reduce signal noise and increase overall channel utilization for wireless multi-hop networks. Existing wireless simulators fail to simulate dominant factors like co-channel and adjust channel interference while simulating a realistic physical layer to close the gap between the actual physical layer and multi-channel multi-radio topology simulation. This thesis attempts to reduce the gap between the physical layer from the real-world testbed and simulators. We propose a high fidelity physical layer simulator (FedEdge Simulator) which uses dynamic link scheduling and trace-based channel modeling to simulate a realistic physical layer for wireless multi-hop networks. The simulator supports the integration of custom-built machine learning model’s to model the channel accurately. In addition, the simulator can act as a learning environment for Reinforcement learning in wireless multi-hop networks and transfer knowledge from simulation to reality. Finally, To illustrate the reduced reality gap betweensimulation and reality, we set up our experiment to integrate our FedEdge simulator with the existing framework of FedEdge. We also show that our design of realistic simulations could help in knowledge transfer in reinforcement networking.

masters theses

Computer science

M.S.

Computer Science

Wang, Pu

Lee, Minwoo Jake
Dorodchi, Mohsen

Thesis (M.S.)--University of North Carolina at Charlotte, 2021.

This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.

Copyright is held by the author unless otherwise indicated.

pothuneedi_uncc_0694N_12939

http://hdl.handle.net/20.500.13093/etd:2927