Ultra-high fidelity radio frequency propagation modeling using distributed high performance graphical processing units: A simulator for multi-element non-stationary antenna systems

摘要

A newly-invented, distributed, high-performance graphical processing framework that simulates complex radio frequency (RF) propagation has been developed and demonstrated. The approach uses an advanced computer architecture and intensive multi-core system to enable high-performance data analysis at the fidelity necessary to design and develop modern sensor systems. This widely applicable simulation and modeling technology aids in the design and development of state-of-the-art systems with complex waveforms and more advanced downstream exploitation techniques, e.g., systems with arbitrary RF waveforms, higher RF bandwidths and increasing resolution. The recent breakthroughs in computing hardware, software, systems and applications has enabled these concepts to be tested and demonstrated in a large variety of environments and early in the design cycle. Improvements in simulation accuracies and simulation timescales have been made that immediately increase the value to the end user. A near-analytic RF propagation model increased the computational need by orders of magnitude. This model also increased required numerical precision. The new general purpose graphics processing units (GPGPUs) provided the capability to simulate the propagation effects and model it with the necessary information dependence, and floating point mathematics where performance matters. The relative performance improvement between the baseline MATLAB® parallelized simulation and the equivalent GPU based simulation using 12 NVIDIA Tesla K20m GPUs on the Offspring High-Performance Computer (HPC) using the AirWASP© framework decreased simulation and modeling from 16.5 days to less than 1 day.