A Deep Neural Network Approach for Spatiotemporal Throughput Prediction in LTE Network
| dc.contributor.advisor | Dereje, Hailemariam (PhD) | |
| dc.contributor.author | Redamichael, Jemberu | |
| dc.date.accessioned | 2022-02-15T07:03:56Z | |
| dc.date.accessioned | 2023-11-04T15:13:05Z | |
| dc.date.available | 2022-02-15T07:03:56Z | |
| dc.date.available | 2023-11-04T15:13:05Z | |
| dc.date.issued | 2021-10 | |
| dc.description.abstract | In the mobile broadband era, whereby high-resolution video services are emerging, user throughput is a key metric to ensure the satisfaction of user experience. Monitoring and assuring the desired level of user throughput is thus a major concern of mobile network operators (MNOs). Since different services and areas have different bandwidth requirements, it is difficult to guarantee a good user experience using traditional Key Performance Indicator (KPI) based network monitoring and analysis. To that end, a proactive approach that takes into account spatiotemporal dimensions is required. The enhancement of Deep Neural Network (DNN) algorithms and MNO’s big data can be leveraged to solve spatiotemporal modeling problems. Due to its ability to capture spatial patterns, Convolutional Long Short-Term Memory (ConvLSTM) is widely used in video prediction and later applied to various spatiotemporal prediction problems. Although MNO's network management system (NMS) can provide cell-level average user throughput data, spatial mapping of NMS data is challenging mainly due to the varying coverage of cells. In this thesis, a drive test-based coverage data analysis is conducted to overcome this challenge. By applying this analysis, cell coverage in a spatial grid of dimension 100m × 100m is derived. Next, a technique that maps Long Term Evolution (LTE) average downlink (DL) user throughput data (obtained from NMS) to each grid is adopted. Finally, the ConvLSTM algorithm is applied to build a model that could predict grid-level user throughput. The developed model provides a 3-hour future DL user throughput prediction with Root Mean Square Error (RMSE) of 2.02, Mean Absolute Error (MAE) of 1.52, and Mean Absolute percentage Error (MAPE) of 14.68. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/30090 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | LTE | en_US |
| dc.subject | DL user throughput | en_US |
| dc.subject | DNN | en_US |
| dc.subject | ConvLSTM | en_US |
| dc.subject | prediction | en_US |
| dc.subject | spatiotemporal | en_US |
| dc.title | A Deep Neural Network Approach for Spatiotemporal Throughput Prediction in LTE Network | en_US |
| dc.type | Thesis | en_US |