Deep Learning Modeling of Power Delay Profile in Motherboard Desktop Environment

In this paper, we present a multilayer perceptron (MLP)-based THz channel model for line of sight propagation in a motherboard desktop environment. A six layer decoder-like MLP network was designed and trained with augmented measured data. The power delay profiles predicted by the MLP model match the relevant signal structures closely while uncovering hidden structure not explained by conventional ray-tracing models. The validity of our model’s predictions is demonstrated by comparing with a baseline of the measured data at different transceiver heights.