Architecture Exploration of a Backprojection Algorithm for Real-Time Video SAR

Area of Science:

• Remote Sensing
• Computational Imaging
• Signal Processing

Background:

• Video Synthetic Aperture Radar (VISAR) systems require efficient processing for real-time applications.
• Backprojection algorithms are computationally intensive, posing challenges for high frame rates.

Purpose of the Study:

• To explore novel multi-GPU architectures for accelerating backprojection-based VISAR (BP-VISAR).
• To analyze and optimize BP-VISAR frame rates in non-overlapped and overlapped aperture modes.

Main Methods:

• Definition of processing data units for parallelizing backprojection.
• Proposal and evaluation of six distinct GPU-based architectures.
• Implementation of multi-stream backprojection on multiple GPUs for enhanced efficiency.

Main Results:

• Evaluated performance of architectures in both non-overlapped and overlapped modes.
• Achieved accelerated frame rates using sub-aperture processing and multi-stream backprojection.
• Demonstrated scalable frame rates with increasing GPU count for high-resolution imagery.

Conclusions:

• The proposed BP-VISAR architectures significantly enhance processing speed.
• Sub-aperture processing with multi-GPU acceleration is effective for high-resolution, real-time video SAR.
• The system can generate high-resolution frames at high frame rates (e.g., 73.5 Hz on quad GPUs).