Real-Time Multi-Channel Synthetic Aperture Radar Imaging on Heterogeneous Clusters

Imaging has been a computationally intensive task in Synthetic Aperture Radar (SAR) systems. This paper presents a heterogeneous cluster approach that makes a trade-off between scalability, portability and performance. The cluster model enables a very flexible pattern of scaling the work size by distributing data to independent nodes. In a particular node, the algorithm can be implemented on both CPUs and GPUs without any modifications to the code. In terms of performance, this model achieves real-time processing running on various GPU platforms and near real-time processing on several CPU machines. It is an advanced model in the system design of the imaging module in SAR ground stations in terms of reducing cost and promoting performance and flexibility.