Machine Learning-Based Path Loss Models For Heterogeneous Radio Network Planning In A Smart Campus

ABSTRACT

An easy-to-use and accurate multi-frequency path loss model is a necessary tool for heterogeneous radio network planning and optimization towards achieving a smart campus. The learning ability in artificial intelligence may be exploited to reduce computational complexity and to improve prediction accuracy. In this research project, an optimal heterogeneous model was developed for path loss predictions in a typical university campus propagation environment using machine learning approach. Radio signal measurements were conducted within the campus of Covenant University, Ota, Nigeria to obtain the logs of signal path loss at 900, 1800, and 2100 MHz. Different path loss prediction models were developed based on Artificial Neural Network (ANN) and Support Vector Machine (SVM) learning algorithms. The prediction accuracy and generalization ability of the ANN-based model, which has seven input nodes (distance, frequency, clutter height, elevation, altitude, latitude, and longitude), single hidden layer with 43 neurons and logarithmic sigmoid (logsig) activation function, and a single output neuron (for path loss variable) with tangent hyperbolic sigmoid (tansig) activation function, was found to be the best when compared to the prediction outputs of SVM-based model, and popular empirical models (i.e. Okumura-Hata, COST 231, ECC-33, and Egli). The ANN-based path loss model was trained based on Levenberg-Marquardt learning (LM) learning algorithm. The prediction outputs of the ANN-based path loss model has the lowest Root Mean Square Error (RMSE) of 4.480 dB, Standard Error Deviation (SED) of 4.479 dB, and the highest value of correlation coefficient (R) of 0.917, relative to the measured path loss values. This finding was further validated by the results of Analysis of Variance (ANOVA) and multiple comparison post-hoc tests. In essence, ANN-based path loss model was found to be the optimal model for heterogeneous radio network planning, deployment, and optimization in a smart campus propagation environment.

Keywords: Path Loss Model; Heterogeneous Radio Network; Artificial Neural Network (ANN); Support Vector Machine (SVM); Radio Network Planning and Optimization (RNP/O); Smart Campus