Machine Learning-Assisted Calibration for Ray-Tracing Channel Simulation at Centimeter-Wave and Millimeter-Wave Bands
Cheng Yi, Weiqi Chen, Qi Wu, Haiming Wang
Abstract
Ray-tracing (RT) channel simulation is widely used in channel modeling and prediction. The electrical parameters of multiple types of materials play an important role in the accuracy and reliability of RT channel simulations. The conventional electrical parameters calibration methods are based on global optimization algorithms and they are computationally exhaustive. A machine learning-assisted optimization method is proposed for efficient electrical parameters calibration. The surrogate model replaces time-consuming RT simulations and is then coupled with a genetic algorithm to rapidly identify the optimal results. Based on the importance of the rays and the parameters to be optimized, the calibration process is designed to be two-stage, which further accelerates optimization convergence. Case studies for centimeter-wave and millimeter-wave indoor channels are conducted and they both show high accuracies after calibration while having optimization speedup of more than 6 folds.