Coverage Prediction Based on Spatial Interpolation Techniques: The Case of UMTS Network in Addis Ababa, Ethiopia

Cellular network coverage prediction is a cornerstone of mobile network operators and service providers in order to provide good services to users. Without coverage provisioning, it is meaningless to talk about service or Quality of service (QoS) provisioning. Coverage planning is a complex task for operators during deploying Radio Access Technology (RAT). This is because it needs to consider multiple and correlated network parameters as well as environmental conditions that are out of their control. It is impossible to completely avoid the existence of coverage holes in cellular networks during the planning phase. Therefore, coverage prediction processes are usually required during the operational phase. Traditionally, the cellular coverage estimation performed through drive tests, which consist of geographically measuring different network coverage metrics with a motor vehicle equipped with mobile radio measurement facilities. The collected coverage measurements through drive test are accurate but limited to roads and other regions accessible by motor vehicles. Drive tests cannot be conducted in the whole region of the network due to many obstacles such as buildings, lakes, and vegetation. Therefore, the drive test is quite inefficient means to solve the coverage problems and cannot offer a complete and reliable picture of the network situation. In this thesis, the performance of two spatial interpolation methods namely, Inverse Distance Weight (IDW) and Ordinary Kriging (OK) were evaluated to select which method is best for Universal Mobile Telecommunication System (UMTS) network coverage prediction using the Common Pilot Channel Received Signal Code Power (CPICH RSCP) collected from drive test. The experimental analysis was performed on a sample data collected from drive test UMTS network in Addis Ababa Ethiopia. Two general interpolation methods were employed with different parameters. The first method is IDW with various powers and number of neighbors and the second method is OK with Gaussian, Spherical and Exponential semivariogram models with different numbers of neighbors. The performance of estimation those algorithms were evaluated through the cross-validation, coefficient of determination (R ), Mean absolute error (MAE) and Root Mean Square Error (RMSE). The test results showed the two coverage prediction methods are able to predict coverage. However, based on the Exponential model of semivariogram with an optimal number of neighbors the OK method estimated with an error of prediction 4.84 RMSE whereas the IDW estimated 5.33 RMSE with a percentage difference of 17%. This shows that OK is more accurate than IDW. The OK method can infer the missing RSCP data and generates a more accurate coverage map than the IDW algorithm. This could probably be OK was able to take into account the spatial structure of data. Therefore, this thesis proposes the OK method as the optimal interpolation model to build a radio coverage map for cellular coverage prediction and hole detection purposes.