Statistical Exploration of Automatic Neighbor Relation Process
Lumpo, Roni-Markus (2024)
2024
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-12-16
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2024121511147
https://urn.fi/URN:NBN:fi:tuni-2024121511147
Tiivistelmä
The scale of mobile networks is expanding nowadays in regards of the amounts of base stations and bandwidth. Therefore more and more coverage areas are needed, which will increase the need for handover processes between signals from one coverage area to another. The automatic neighbor relation process is used to automate the neighbor relations of the cells of mobile networks. The cells need to identify neighboring cells in order to conduct the handover process without interruptions in the signal.
This thesis is about exploring the neighbor relations by combining network and time series analysis using a real-life dataset. The main point of the thesis is to figure out how trivial the automatic neighbor relation process is and how predictable the relations are. The work began by inspecting the dataset itself and filtering out the relevant data. A network is formed from an edge list obtained as a result of the filtering. The network is divided into hierarchical levels and blocks by using a nested stochastic block model. After that, the time series related to each block are obtained and a seasonal decomposition is used to obtain the residuals of each time series. The normalized block residuals are then compared to the normalized residuals related to the largest connecting component’s time series, which displays a strong daily seasonal pattern.
The residual comparison is conducted by using Smirnov-Kolmogorov two sample test, which is determining if the two samples are coming from the same distribution. The results of this comparison shows that larger blocks in the higher hierarchical levels are display similar behaviour as the largest connecting component and the smaller blocks in lower levels start to lose the similarity in behaviour. This can be translated into information that independent nodes act rather randomly, but the communities in the network act consistently when compared to the activity of the largest connecting component.
The results would indicate that predicting a single relation is not a trivial task, but predicting the amount of relations is rather simple because of the strong seasonal pattern in the block and network level. The predictions of relations of single cells could benefit from the fact of the close physical proximity of the neighboring cells and base stations. The close physical proximity could be utilized as an attribute for the prediction.
This thesis is about exploring the neighbor relations by combining network and time series analysis using a real-life dataset. The main point of the thesis is to figure out how trivial the automatic neighbor relation process is and how predictable the relations are. The work began by inspecting the dataset itself and filtering out the relevant data. A network is formed from an edge list obtained as a result of the filtering. The network is divided into hierarchical levels and blocks by using a nested stochastic block model. After that, the time series related to each block are obtained and a seasonal decomposition is used to obtain the residuals of each time series. The normalized block residuals are then compared to the normalized residuals related to the largest connecting component’s time series, which displays a strong daily seasonal pattern.
The residual comparison is conducted by using Smirnov-Kolmogorov two sample test, which is determining if the two samples are coming from the same distribution. The results of this comparison shows that larger blocks in the higher hierarchical levels are display similar behaviour as the largest connecting component and the smaller blocks in lower levels start to lose the similarity in behaviour. This can be translated into information that independent nodes act rather randomly, but the communities in the network act consistently when compared to the activity of the largest connecting component.
The results would indicate that predicting a single relation is not a trivial task, but predicting the amount of relations is rather simple because of the strong seasonal pattern in the block and network level. The predictions of relations of single cells could benefit from the fact of the close physical proximity of the neighboring cells and base stations. The close physical proximity could be utilized as an attribute for the prediction.