Echo state network approach for radio signal strength prediction applied to cellular communication frequency bands in northen Namibia
| dc.contributor.author | Gideon, Kenneth | |
| dc.date.accessioned | 2018-05-26T16:12:37Z | |
| dc.date.available | 2018-05-26T16:12:37Z | |
| dc.date.issued | 2018 | |
| dc.description | A thesis submitted in fulfillment of the requirements for the Degree of Master of Science Electronics and Computer Engineering | en_US |
| dc.description.abstract | Reliance on mobile connectivity has led to demands for wireless spectrum capacity to grow on a daily basis resulting to congested networks. Ensuring acceptable levels of Quality of Service (QoS) for users in wireless communication systems, through continuous wireless network analysis using simulation tools based on radio propagation models has become increasingly prominent. To provide automated analytical model building, the use of machine learning methods has been considered to predict characteristics of the wireless channel. Thus, in this work, a method for predicting radio signal strength using Echo State Networks (ESNs) is proposed and applied to three different locations in Northern Namibia. This method aims at providing a better approach for radio signal strength prediction, which leads to improvements in wireless communication planning, design and analysis. Its performance is compared with the Support Vector Regression (SVR) method optimized for radio propagation modeling. Simulations are conducted in Python using propagation data measured from the three locations based on the following four performance metrics: goodness of fit criteria; error measures; computation complexities; and F-Test for statistical model comparison. Simulation results show that the ESN gives a better prediction accuracy in terms of the goodness of fit criteria and the error measures (i.e. average R2 = 0.82 and average mean absolute error (MAE) = 0.0312 for ESN compared to 0.648 and 0.0624 for SVR), but it is inferior to the SVR in terms of computation complexities (i.e. average training complexity of 410 ms and average testing complexity of 79.0 ms for ESN compared to 8.19 ms and 1.04 ms for SVR). In addition, results from the F-Test also indicates that the ESN provides a significantly better fit than the SVR. | en_US |
| dc.identifier.uri | http://hdl.handle.net/11070/2278 | |
| dc.language.iso | en | en_US |
| dc.publisher | University of Namibia | en_US |
| dc.subject | Echo | en_US |
| dc.subject | Network | en_US |
| dc.subject | Radio signal | en_US |
| dc.title | Echo state network approach for radio signal strength prediction applied to cellular communication frequency bands in northen Namibia | en_US |
| dc.type | Thesis | en_US |