Synthetic Aperture Radar on the Jetson TX1

John MelonakosArrayFire, Case Studies, Computer Vision, CUDA, Image Processing 1 Comment

Researchers at Peter the Great St. Petersburg Polytechnic University have implemented a synthetic aperture radar processing on the Jetson TX1 Platform using ArrayFire as described in this paper.

The paper introduces SAR as “a remote sensing technique producing high-resolution radar images of the Earth’s surface. SAR technology allows obtaining wide swath radar images of objects at a considerable distance regardless of the weather and lighting conditions. It can be used by unmanned aerial vehicles and space satellites. Thus, SAR technology allows solving various problems, such as: detecting small objects (vehicles, airplanes, ships), assessing the state of railways, airfields, seaports, mapping an area, assisting in geological exploration, mapping vegetation, detecting oil spills and pollution as well as many other tasks.”

The synthetic aperture is the result of the successive reception of the electromagnetic wave by the real aperture at various positions relative to a target followed by a special signal processing as shown in the paper’s Figures 2 & 3.

The researchers chose ArrayFire to implement its SAR processing algorithms on the Jetson TX1, as well as reference implementations on desktop CPUs and GPUs.

From the paper’s conclusion: “Analysis of execution times shows that Jetson TX1 CPU is several times slower than a typical desktop CPU. On the other hand, Jetson TX1 GPU performance compares favorably to a higher-end desktop GPU. Both types of GPUs are more effective in parallelizing FFT-based algorithms than the algorithm that uses individual coordinates of trajectory points. In the latter case, desktop CPU is more efficient than desktop GPU, but Jetson TX1 GPU is still significantly faster than its CPU. This may be due to the different memory architectures.”

The radar image produced by the FFT-based algorithm is shown in Fig. 4 (note the significant distortions in the top-right corner, where the flight path is non-linear). The image processed using the navigation system data is shown in Fig. 5, depicting a much cleaner version of the scene.

For more information on synthetic aperture radar with ArrayFire, ask us anything at or schedule a phone call with a sales engineer.

Comments 1

Leave a Reply

Your email address will not be published. Required fields are marked *