A GPU-Oriented Onboard Imaging Framework for High-Resolution Sliding Spotlight SAR on an Embedded GPU Platform

Area of Science:

• Remote Sensing
• Signal Processing
• Computer Engineering

Background:

• Conventional spaceborne Synthetic Aperture Radar (SAR) systems face communication burdens and processing delays due to ground-based data downlink.
• Onboard SAR imaging is limited by power, memory, thermal, and size constraints of embedded processors.
• High-resolution sliding spotlight SAR demands advanced processing due to complex range-azimuth coupling and migration, challenging existing onboard methods.

Purpose of the Study:

• To develop a GPU-oriented ω-k imaging framework for high-resolution sliding spotlight SAR suitable for embedded platforms.
• To overcome the computational and memory limitations of embedded GPUs for complex SAR imaging algorithms.
• To enable efficient onboard processing of high-resolution SAR data on platforms like the Jetson AGX Orin.

Main Methods:

• A GPU-oriented ω-k imaging framework was designed for sliding spotlight SAR.
• Hybrid data partitioning and an asynchronous multi-stream pipeline were implemented to optimize memory access and computation.
• Customized GPU kernels were developed for Stolt interpolation and FFT-based spectral centering.

Main Results:

• The proposed framework achieves well-focused imaging results with consistent impulse response characteristics.
• For a large simulated SAR dataset (32,768×32,768), an end-to-end imaging time of 32.17 seconds was achieved on the Jetson AGX Orin.
• The framework demonstrated efficient execution on the embedded GPU platform, validating the optimization strategies.

Conclusions:

• The developed GPU-oriented ω-k imaging framework is feasible for onboard high-resolution sliding spotlight SAR.
• The optimization strategies effectively address the challenges of memory and computation on embedded GPU platforms.
• This work paves the way for advanced onboard SAR processing capabilities in spaceborne systems.