Power vs. Performance Evaluation of Synthetic Aperture Radar Image-Formation Algorithms and Implementations for Embedded HEC Environments (Ongoing Study)

摘要

The continuing miniaturization and parallelization of processing hardware has facilitated the development of mobile and field-deployable systems that can accommodate terascale processing within once prohibitively small size and weight constraints. Unfortunately, the added computational capability of these small systems often comes at the cost of larger power demands, an already strained resource in these embedded systems. This study explores the power issues in a specific type of field-deployable system, Mobile Radar. Specifically, we focus on a computationally intensive phase of Synthetic Aperture Radar, Image Formation (IF), and evaluate performance tradeoffs in terms of time-to-solution, output image quality, and power consumption for two different implementations, single- and double-precision, of two different IF algorithms, one frequency-domain based and the other time-domain based. Preliminary results show that in some CPU-based instances single-precision IF leads to significant reductions in time-to-solution and, thus, total energy consumption (over 50%) with negligible but possibly acceptable SAR image output degradation. In the near future, this ongoing study will reevaluate these results, i.e., SAR IF power consumption vs. performance tradeoffs with more sophisticated IF workloads and output quality metrics, finer-grain performance and power measurement methodologies, and more computationally powerful embedded HEC devices, i.e., GPGPUs.