Efficient Time-Domain Simulation of Sub-Terahertz Multipath: A Hybrid Architecture and Multi-Tiered Validation

Abstract

As wireless communication systems transition toward sub-terahertz (sub-THz) and terahertz frequencies to support 6G data rates, deterministic channel modeling becomes computation- ally bottlenecked. At sub-millimeter wavelengths, common indoor textures such as drywall roughness and carpet weave act as complex diffuse scatterers rather than simple specular re- flectors. Standard full-wave numerical methods, such as the Finite-Difference Time-Domain (FDTD) method, are computationally prohibitive at these macroscopic scales, while efficient ray-based techniques typically rely on approximations to estimate surface roughness effects. This thesis presents a novel, efficient time-domain hybrid computational electromagnetic sim- ulator designed to deterministically resolve sub-THz multipath scattering without the need for high-performance computing. The three-phase architecture integrates a two-dimensional FDTD engine with analytical ray launching and Rayleigh-Sommerfeld scalar diffraction prop- agation. A Simultaneous Grid Subtraction technique is also implemented to eliminate nu- merical dispersion during cylindrical Total-Field/Scattered-Field (TF/SF) source injection. The hybrid simulator is validated against commercial full-wave solvers (HFSS) and empir- ical measurements utilizing a 75–115 GHz Vector Network Analyzer (VNA) in a modified Naval Research Labs (NRL) Arch configuration. Time-domain Monte Carlo simulations of stochastically generated rough surfaces, modeled using Perfect Electric Conductor (PEC) boundary conditions, demonstrate excellent agreement with physical measurements, achiev- ing a 0.97 correlation in the extracted Power Delay Profiles. The simulator accurately predicts macroscopic channel statistics, matching empirical Mean Excess Delay to within 2.05 ps and Rician K-Factor to within 1.55 dB. The localized FDTD domain partitioning achieves an 86% reduction in active grid cells, shrinking the static memory footprint from 2.54 GB to 346 MB. This optimization yields a 16x reduction in execution time compared to a full-wave equivalent, executing sub-THz Monte Carlo campaigns in under 10 hours. This work provides a computationally viable, deterministic alternative to statistical models and ray-based deterministic models, bridging the gap between macroscopic simulation efficiency and microscopic electromagnetic precision.