How GPUs Are Delivering a 10,000x Improvement in Missile-Tracking Systems
Getting a 2-3x performance boost with an upgraded GPU is a matter of delight. Applying GPU technology to get a 10,000x performance improvement in a missile-tracking phased array radar system is a matter of national security.
Yet that’s exactly what the real-time systems and software engineering firm FishEye Software has done. In a talk at our recent GPU Technology Conference, Ted Selig, chief operating officer at the Boston-based company, described how GPU acceleration helped overcome the limitations of conventional systems.
By doing so, FishEye has shown a 10,000-fold gain in price-performance with a system that operates at real-time speeds, tracking scores of missiles.
Phased-array radars identify and track incoming missiles, as well as outgoing interceptor missiles, and can send messages to the outbound ones to help guide their flight. The systems project electromagnetic waves — like a bat or dolphin uses sonar waves, except at the speed of light. And they’re about five stories tall, housing a large antenna that can be electronically adjusted to focus and broaden the beam used to identify objects.
Conventional phased-array radar systems are built with custom hardware and very expensive supercomputers, which create all sorts of complications. Custom hardware is pricey and hard to source. Supercomputers have substantial and expensive cooling and power demands. And because they’re built to spec, there aren’t often spares on hand if something goes wrong.
Difficult to modernize and maintain, a traditional system will often endure a typical two-year validation cycle and come out the other end behind the technology curve. Plus, testing is challenging and costly, requiring travel to radar sites, often in remote reaches of the world.
Managing Massive Complexity
A key challenge behind these systems is the array of thousands of antenna on a radar’s face. A phased array system slightly changes the signal from antennae in relation to their neighbors to locate, identify and track objects. Coordinating these thousands of antennae and extracting data to inform human decisions is a massively complex undertaking.
Using about $10,000 in off-the-shelf technology, FishEye’s system outperforms custom radar hardware at a fraction of the price of a traditional system, which can cost upwards of $3 million. And the performance difference isn’t even close.
A conventional system might ably simulate two radar objects in real time. FishEye’s system, which currently uses GeForce GTX 660 and 780 Ti GPUs for processing, can calculate 400,000 glints — the small, electromagnetic sparkles that reflect and make up objects seen by radar — per second.
Selig points out that the GPU-based system presents an opportunity for defense departments to save millions of dollars in radar lifecycle costs, while pushing radar capabilities to new limits using previously impractical scenarios.
This high-fidelity and real-time view enables the system to better understand what’s in the sky and distinguish clutter from targets. The tech is off the shelf, so upgrades are easier, reducing the time to obsolescence. And because the system is so fast, FishEye can accurately simulate the front-end data collection without having to travel thousands of miles to remote locations.
More and higher fidelity testing in the lab lowers costs and gets problems worked out quickly and early in the process. This lets customers better understand exactly what they want, on a faster schedule, before they buy and build.