Comparison of Statistical and Deep Learning Path Loss Model for Motherboard Desktop Environment

• Published in: 2022 16th European Conference on Antennas and Propagation (EuCAP) pp. 1 - 5
• Main Authors: Fu, Jinbang, Juyal, Prateek, Jorgensen, Erik J., Zajic, Alenka
• Format: Conference Proceeding
• Language: English
• Published: European Association for Antennas and Propagation 27-03-2022
• Subjects: Adaptation models; Deep learning; Loss measurement; measurements; Mixture models; path loss model; propagation; Propagation losses; Resonant frequency; THz communications; Wireless communication
• DOI: 10.23919/EuCAP53622.2022.9769371

In this paper, we characterized the path loss of the THz wireless channel in motherboard desktop environment and modeled it by both statistical (mixture distributions) and deep learning (multilayer perceptron) models. The performance of the two different model classes are compared and results show that mixture models captures the randomness of the channel by matching the PDF of measured path loss, which means that the statistical model can adapt to the changing environment. However, for the complex yet static motherboard desktop environment, the deep learning model outperforms the statistical models since it can also precisely describe the hidden patterns due to resonant modes and signal propagation in the static environment.