Cell distance predictor without timing advance : a machine learning approach
Rekis, Matti (2024)
2024
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-2024111428169
https://urn.fi/URN:NBN:fi:amk-2024111428169
Tiivistelmä
This thesis was commissioned by Enhancell Ltd., a company located in Oulu, Finland, that makes tools to test mobile networks. The author, Matti Rekis, is a long-time employee of Enhancell and has developed an algorithm to estimate the locations of radio base stations or cells.
A cell can be located by using GPS to find out the positions of the measurer and then determining the distances to the cell to which the measurer’s user equipment (UE) is connected. The UE in this case is a mobile phone. The distance to the cell can be solved with a mobile network parameter called timing advance (TA).
The goal of this thesis was to develop a cell distance predictor based on machine learning models to predict the cell distance in cases where the TA is not available. For instance, it is not available when using a cellular network scanner for cell detection, and it is not available in the UMTS network when using a mobile phone.
The machine learning models were trained by using cellular signal parameters useful for distance prediction purposes as the features and the distance to cell as the target. The training data was collected with measurement tools made by Enhancell, and for the training, Keras, an open-source library to train artificial neural networks, was used.
After the training and evaluation, the cell distance predictor was implemented in the location estimation module of Echo Studio. Echo Studio is a post-processing tool for mobile network data made by Enhancell that runs in a desktop environment.
A cell can be located by using GPS to find out the positions of the measurer and then determining the distances to the cell to which the measurer’s user equipment (UE) is connected. The UE in this case is a mobile phone. The distance to the cell can be solved with a mobile network parameter called timing advance (TA).
The goal of this thesis was to develop a cell distance predictor based on machine learning models to predict the cell distance in cases where the TA is not available. For instance, it is not available when using a cellular network scanner for cell detection, and it is not available in the UMTS network when using a mobile phone.
The machine learning models were trained by using cellular signal parameters useful for distance prediction purposes as the features and the distance to cell as the target. The training data was collected with measurement tools made by Enhancell, and for the training, Keras, an open-source library to train artificial neural networks, was used.
After the training and evaluation, the cell distance predictor was implemented in the location estimation module of Echo Studio. Echo Studio is a post-processing tool for mobile network data made by Enhancell that runs in a desktop environment.