ABSTRACT
Prediction of path loss with least error between a signal source and a receiver is a fundamental problem in wireless network planning and optimization stage. During network planning, a reliable signal loss model is required for effective prediction of signal attenuation. However, the standard path loss models commonly explored in literature always have precision problems.
A hybrid optimization approach which employs Random Forest and Particle Swarm (RF-PS) is proposed for effective development of path loss models in this study. To achieve this, signal power measurements were made on Long Term Evolution (LTE) network at 2600MHz band using Test Equipment for Mobile System (TEMS) drive test tool. The measurement campaign was conducted at Agbor and Asaba in Delta State and Onitsha and Awka in Anambra State. The measurements were carried out between January and December, 2019. After the measurement, the signal power data were first routed through random forest for feature selection and dimensionality reduction process. Consequently, 100 set of trees were carefully engaged on the measured datasets to obtain the most informative and important subset of features which were later employed as input data to particle swarm optimization for further predictive modelling.
The performance of the proposed technique over standard Radial Basis Function- Particle Swarm (RBF-PS) and Particle Swarm (PS) optimization methods were evaluated based on Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Standard Deviation (STD), Mean Absolute Percentage Error (MAPE), Correlation Coefficient (R) and Relative Error (RE) statistical indexes. The results revealed that the proposed RF-PS optimization method attained the best prediction performance using the six statistical indexes. In terms of RMSE and RE, the proposed RF-PS, RBF-PS and PS optimization methods recorded the values of 2.09 - 4.48 dB, 2.69 – 6.16 dB, 2.69 – 6.38 dB and 3 - 11%, 8 – 11%, 9 – 12% across the four studied locations. These results imply that the proposed RF-PS models are promising models for path loss prediction than the standard models.
The proposed techniques can be employed by telecommunication operators to enhance Quality of Service (QoS) and planning of radio network at 2600MHz frequency band for 4G LTE network in the studied and other similar environments.
