Undergrad students deploy applications to geosynchronous satellite 22,236 miles above Earth

For many college students, satellites are faint dots crossing the sky on clear nights. These objects are background details from science fiction to reality, like GPS, satellite radio, and WiFi.

For University of Pittsburgh computer engineering students Dikchhya and Nischal Kharel, the compact objects hurtling through space and pelted by radiation are unique challenges.

Through a partnership with the University of Pittsburgh’s NSF Center for Space, High-Performance & Resilient Computing (SHREC) and Lockheed Martin, the two undergraduates had a unique opportunity to develop and deploy applications to a satellite in geosynchronous orbit (GEO), 22,236 miles above Earth.

Discovering SHREC

As the cousins were nearing graduation from the Swanson School of Engineering last spring, neither had concrete plans to attend graduate school. That changed after a conversation with Samuel Dickerson, associate professor of electrical and computer engineering.

“When Professor Dickerson learned that we both loved space and wanted to work at NASA, he asked if we knew about SHREC,” Nischal said. “Somehow, we didn’t.”

Dickerson amended that, introducing them to Alan George, Department Chair, R&H Mickle Endowed Chair, NSF SHREC Space Center Director, and professor of electrical and computer engineering. George provided Dikchhya and Nischal with the opportunity to earn credits conducting space research with Linus Silbernagel, a second-year Ph.D. student, and Evan Gretok, a SHREC postdoctoral researcher.

Solving real-world problems

On November 1, 2022, Lockheed Martin launched an In-space Upgrade Satellite System (LM LINUSS) into GEO, which enables satellites to orbit at the same rate as Earth’s rotation, making them appear stationary from the ground. Today, as the mission nears completion, the CubeSat satellite serves as a test bench.

In GEO, more than 20,000 miles farther out than satellites in low Earth orbit (LEO), LINUSS provides an opportunity to test code and applications in more challenging conditions.

“Lockheed Martin approached Dr. George about the possibility of building and running tests in a GEO satellite because of SHREC’s expertise in satellite hardware and software,” said Silbernagel. “We had some problems that Dikchhya and Nischal could help solve.”

The limits of memory

CubeSat satellites are incredibly small—some fewer than four inches—and thus constrained in their computational power, reducing function efficiency. Tiny onboard cameras, for instance, can capture wavelengths that the human eye can’t process but that create enormous files that waste time and energy to transmit back to Earth.

“I worked on a project called CNN JPEG, which a graduate student had started and that uses machine learning to compress data in satellite photos,” said Nischal. “I wanted to deploy the technology on LINUSS.”

The LINUSS satellite uses a ZCU102 computing board, and the team replicated Lockheed Martin’s satellite configuration on it.

“Lockheed Martin packages their apps in a certain way,” Nischal said. “We got the technology running on our test board and sent it to them to be transmitted to LINUSS.”

Unfortunately, although the app worked on Earth, it was too large to run in the satellite. “In classes, we’re taught about memory constraint, but it’s different when you actually run into it,” said Nischal, who has been working to optimize CNN JPEG.

Making it work in space

Satellites generate many images, but most are irrelevant. They depict stretches of ocean or forest, or just clouds. “Satellites send back many images of little interest to researchers, wasting memory and bandwidth, which creates delays,” said Dikchhya.

To solve this problem, Dikchhya set out to run a machine-learning model on LINUSS. The model, trained by Gretok, enables autonomous onboard classification on a constrained system. The application reduces massive images to tiles and categorizes them to help eliminate redundant data. “My job was to make it work in space,” Dikchhya said.

She did just that.

Using the ZCU102 board, she programmed and packaged the application and sent it to Lockheed Martin.

“Mine is a smaller model, so it ran without issues,” said Dikchhya. “After our final tests, the Lockheed Martin team sent it up to LINUSS, ran the application, and the model correctly classified the images. Watching something we built operate successfully 22,000 miles above Earth was a major milestone and an exciting opportunity.”

“It would be impossible for students to do something this big without Lockheed Martin’s partnership,” said Silbernagel. “Dikchhya and Nischal got to join the team remotely during the testing and deployment. It’s the first time Pitt undergraduate students have run apps in GEO.”

“This project represents an essential aspect of SHREC,” said George. “Through university and industry partnerships like this one, students get these incredible opportunities that propel them into space research and prepare them for careers in space engineering.”

“When we started,” said Nischal, who like Dikchhya is now a SHREC graduate student, “we were undergraduate students with almost no real-world research experience. We encountered problems that others hadn’t, and we had to figure out how to solve them.”