Accelerating Spaceborne SAR Imaging Using Multiple CPU/GPU Deep Collaborative Computing.

Area of Science:

• Remote Sensing
• High-Performance Computing
• Signal Processing

Background:

• Synthetic Aperture Radar (SAR) technology generates massive datasets, posing challenges for real-time image processing.
• Current GPU-based methods often underutilize CPU capabilities, limiting overall efficiency.

Purpose of the Study:

• To develop a deep collaborative SAR imaging method utilizing multiple CPUs and GPUs for real-time processing.
• To enhance SAR imaging efficiency by integrating advanced CPU and GPU parallel computing strategies.

Main Methods:

• Proposed a task partitioning and scheduling strategy for deep collaborative CPU/GPU computing.
• Implemented the Advanced Vector Extension (AVX) method for multi-core CPU parallel imaging.
• Optimized GPU parallel imaging using streaming and parallel pipeline techniques to overcome memory and data transfer limitations.

Main Results:

• The deep collaborative CPU/GPU method significantly enhances SAR imaging efficiency.
• Achieved a 270-fold increase in SAR imaging efficiency on a single-core CPU compared to traditional methods.
• Realized real-time SAR imaging with an imaging rate exceeding the raw data generation rate.

Conclusions:

• The proposed deep collaborative SAR imaging method effectively addresses the challenges of real-time processing of large remote sensing datasets.
• This integrated approach maximizes computational resources by synergizing CPU and GPU capabilities.
• The method demonstrates the potential for achieving real-time SAR imaging crucial for various applications.