Using machine learning to find optimal beam groups
Saukkonen, Samuli (2023)
2023
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2023121838061
https://urn.fi/URN:NBN:fi:amk-2023121838061
Tiivistelmä
The object of the thesis was to use machine learning to find optimal radio beams that work together without interfering each other. Stochastic hill climbing and genetic algorithm were used to find optimal beam groups from beams measurement data. Beams signal to noise ratio in the beam groups were used for verifying if the results were good.
A machine learning program was created that worked as a microservice. It was integrated to a cloud-based web service where users could start the microservice and view the available results.
Especially the genetic algorithm worked well. It was able to find working beam groups from the given data. However, a problem arose as the machine learning program used only measurement data’s azimuth results. As the beams can be directed both on azimuth and elevation plane machine learning could not handle this. Further update for this would be needed to get results to work with real use cases.
This thesis verified that machine learning can be used to find optimal beams groups. It also highlighted how time and money consuming creating machine learning projects can be.
A machine learning program was created that worked as a microservice. It was integrated to a cloud-based web service where users could start the microservice and view the available results.
Especially the genetic algorithm worked well. It was able to find working beam groups from the given data. However, a problem arose as the machine learning program used only measurement data’s azimuth results. As the beams can be directed both on azimuth and elevation plane machine learning could not handle this. Further update for this would be needed to get results to work with real use cases.
This thesis verified that machine learning can be used to find optimal beams groups. It also highlighted how time and money consuming creating machine learning projects can be.