Path Loss Prediction of 5G in the 24.25-27.5 GHz Band based on Machine Learning

Abedeen, Zaid, Ekpo, Sunday ORCID: https://orcid.org/0000-0001-9219-3759, Ijaz, Muhammad ORCID: https://orcid.org/0000-0002-0050-9435, Raza, Umar ORCID: https://orcid.org/0000-0002-9810-1285, Alabi, Stephen and Han, Liangxiu ORCID: https://orcid.org/0000-0003-2491-7473 (2024) Path Loss Prediction of 5G in the 24.25-27.5 GHz Band based on Machine Learning. In: The Second International Adaptive and Sustainable Science, Engineering and Technology Conference: ASSET 2023 Proceedings. Signals and Communication Technology . Springer, Cham, pp. 19-28. ISBN 9783031539343 (hardcover); 9783031539350 (ebook)

Accepted Version File not available for download. Available under License In Copyright. Download (1MB)

Abstract

Millimeter-wave 5G signals require accurate path loss predictions due to the low spectral and energy efficiencies of pre5G networks. This paper proposes a hybrid machine learning technique comprising an environment classifier that determines the propagation environment using a convolutional neural network (CNN) in the TensorFlow machine learning framework and a path loss model using the XGboost model. The results of the evaluation demonstrate the model's exceptional accuracy in predicting path loss for the 5G n258 standard (24.25-27.5 GHz) band. Through extensive training and testing using a carefully constructed dataset, the model achieves a root mean square error (RMSE) under 1 dB when compared with the empirical 26 GHz band measurements. Moreover, the machine learning model demonstrates a low computational latency with the parameters sweep predictions of 0.23 seconds, yielding a 99.65 % decrease in execution time compared with the conventional methods.