Acceleration of finite difference time domain modeling using GPU and transfer functions with application to channel modeling

Abstract

Next generation communication technologies aim to use broadband and/or high frequency systems for commercial communications, including utilizing mmWave frequencies and ultra-wide bands. Channel modeling at these frequencies is currently the focus of extensive measurement campaigns. Application of the Finite Difference Time Domain (FDTD) method at mmWave frequencies is suitable for modeling the broadband system, but several challenges remain between it and practical implementation. This thesis shows practical, simple GPU implementations suitable for FDTD modeling using MATLAB for accelerating large problems, such as those found at mmWave frequencies. Additionally, it’s shown that transfer functions can be utilized within the FDTD method to allow for simulation of arbitrary length signals within ordinary simulation times, that can achieve better than -30dB of error between transfer function and direct simulation approaches.