Helping radar to see more clearly
Graphic processing units (GPUs) are the powerful electronic circuits that have opened the way for artificial intelligence and cryptocurrency trading. And they also have the potential to make military radars far more effective, according to the results of a collaboration between Saab and AI computing company NVIDIA.
Rotary wing drones have become a major battlefield threat in recent years. As well as enabling aerial reconnaissance, small drones can be easily weaponised and used with devastating effect on assets on land and at sea. To remain effective, armed forces need ways to clearly identify this growing aerial threat.
Now, a collaboration between Saab and AI computing company NVIDIA has demonstrated the effectiveness of graphic processing units (GPUs) in signal processing for radar. By supplementing or completely replacing conventional central processing units (CPUs) in such applications, GPUs have the potential to increase radar resolution and performance. This could allow radar operators to detect drones earlier and with far greater precision, to detect smaller drones, and to better discern between individual UAVs.
In a presentation given at the NVIDIA GTC 2025 conference for AI developers in California earlier this year, a Saab team revealed that GPUs were able to process radar signals up to ten times faster than CPUs during testing.
“The GPU-based system gave a significant performance increase and there's potentially more to come,” says software architect Jacob Lundberg who was part of a Saab team working on the project. “It opens the way for more signal processing opportunities that were previously out of reach.”
“The GPU-based system gave a significant performance increase and there's potentially more to come,”
Made for parallel processing
Two Saab teams received training from NVIDIA in mid-2024 – one working on using large language models in defence, and Jacob’s team who were focused on the use of GPUs in radar.
GPUs and CPUs differ in the way that they process data. CPUs are versatile and rely on a relatively small number of cores or processing units to perform sequential tasks. GPUs, meanwhile, have a larger number of cores, making them suitable for parallel processing – that is performing many tasks at once.
CPUs have traditionally been used in signal processing and they deliver exceptional results in a range of Saab radar systems. But in some ways GPUs are more suited to the task. With an advanced radar system like Saab Giraffe 1X, hundreds of elements in the antenna capture signals which must be processed to create a moving radar picture. And managing many processing task at once is something that GPUs do well.
The team knew that GPUs had been tested at Saab about a decade ago and been found unsuitable for this kind of signal processing. But with recent advances in technology, they wondered if modern GPUs could work better.
Up to 10 times faster
In a series of tests, they compared the performance of GPUs and CPUs in radar signal processing applications. In some instances the processors were connected to a Giraffe 1X radar antenna. Algorithms are used to enable the processors to read the signals from the antenna. The team parts of the existing CPU algorithms in NVIDIA Cuda for use by the GPUs. Completely new algorithms for the GPUs were also tested.
In all cases, replacing CPUs with GPUs increased processing speed by at least four times and as much as 10 times. In real-world operations, such a processing speed boost could translate to increased picture resolution and better performance, as well as reducing hardware requirements. Operators might be better able to identify each and every drone in a swarm and continuously track their movement.
While more work is needed, the team’s results suggest there is enormous potential to improve radar performance with GPUs.
NVIDIA is also a leading supplier of artificial intelligence (AI) hardware and software. Held this year in San Jose, California, the NVIDIA GPU Technology Conference (GTC) is a global artificial intelligence conference for developers, engineers, researchers, inventors, and IT professionals. It aims to showcase how AI and accelerated computing are helping to solve some of humanity’s most complex problems.
The presentation is available at the GTC website