For decades, the amount of compute power on satellites has remained relatively consistent and minimal. Since satellites have historically generated and transmitted information to those back on Earth, they didn’t need significant compute resources or processing capabilities.

As Bob Beachler, Vice President of Product at Untether AI recounted, “[When] I started in the [satellite] industry 35 years ago [there was] very rudimentary on-satellite compute. I mean, we're talking about eight-bit CPUs. Your smart speaker has more intelligence than what was going up in those birds…”

But that is all starting to change as the edge computing trend occurring in terrestrial networks makes its way into space.

Driven by demands to reduce the amount of data being transmitted across terrestrial networks and remove data transfer latency from processes and operations, technologists have been working to push compute resources and operations closer to where data is being generated.

Now, entrepreneurial startup companies are embracing the same approach with satellites, increasing the compute power of spacecraft in an effort to process data more quickly and efficiently. As Edward Ge, Co-Founder of Aethero explained, “When we say edge computing in space, we're talking about actually processing and computing the data onboard the satellite, as it's being generated.”

Ge noted the stark difference from how satellites operated in the past. “Before, we had satellites with very low computing and data processing capabilities, and as the satellites would collect data, they have to send all of the data to the ground station [for analysis],” he told Constellations.

This evolution is due in large part to advancements in compute technologies, which have drastically reduced the size and power requirements of computers. Considering edge computing’s ability to reduce the amount of data that needs to be transmitted, adding computers to satellites could become more cost and size effective than building small satellites with numerous transmitters.

“These computers aren't big. We aren’t talking about some massive laser terminal like those on optical relays. We're talking about a relatively small, plug-in compute module that can be deployed on something as small as a CubeSat,” explained Ge. “The power cost and space requirements to support these computers is also lower than using a higher-powered Ka-band transmitter or upgrading your pointing accuracy to support laser links, making it much easier to slide a compute module in...”

This new approach to data aggregation and analysis could be revolutionary to the satellite industry, driving the advancement of innovative space capabilities, and enabling more rapid decision-making for those that rely on satellite data for their operations.

Learn more, act faster

There are numerous organizations that rely on satellite data to inform decision-making. Unfortunately, the data that they receive from satellites is often slowed down due to latency and the small windows in which Earth Observation (EO) data satellites can transmit data back down to Earth. As Ge explained, “In a typical 90-minute orbit, a commercial satellite may only be over a ground station for 20 minutes. Oftentimes, that number is really between five and fifteen minutes…”

The ability to transmit important EO data back to Earth is limited to the time when satellites are strategically located above ground stations. This means that some data can wait long periods before it can be transmitted – a challenge that is growing as EO sensors and cameras improve in quality.

“…when you look at your newer EO satellites [with capabilities like] hyperspectral imaging, synthetic-aperture radar, and thermal remote sensing, these sensors generate data at a rate measured in gigabytes per second,” Ge said. “Companies oftentimes have to shut off the sensor or dump potentially valuable data and insights from the satellite just to make room [for] new data or to [transmit all of] the data in the relatively limited window that they have over the ground stations.”

Edge computing can offer a solution to this challenge by creating an environment in which advanced AI applications can analyze the data being generated on the satellite. This results in less raw data needing to be transmitted back to the Earth. Instead, satellites only need to transmit alerts, findings, or what is considered the most important data.

“Rather than sending the large amount of data that a reconnaissance satellite [generates] to get number crunched terrestrially, you could now do that on the satellite. It reduces the amount of bandwidth you need to beam home,” Beachler said. “For example, you can have the satellite counting cars, people, houses, tanks, airplanes, or whatnot on the street, and [conduct] object detection, recognition and classification. The satellite would then just send the relevant data back down rather than sending all of the data.”

But the benefits of smarter, more compute capable satellites extend beyond generating more timely, actionable insights for those that rely on EO data. Smarter, more compute-capable satellites could increase autonomy – opening the door to future capabilities that could revolutionize the space and satellite industries.

Self-driving satellites that can learn

By increasing the compute power in satellites and making satellites capable of gathering, processing and acting on data, the need for human intervention in satellite operations decreases. This is an essential step in the adoption of artificial intelligence (AI) in satellites and the increase in satellite autonomy.

This means satellites could process information, gain actionable insights, and alter their own operations to more effectively accomplish their missions. That level of autonomy could have a massive impact on the effectiveness of Rendezvous and Proximity Operations.

“We're talking about much more autonomous and adaptive decision making for services like on-orbit servicing,” Ge said. “[Edge computing in satellites] could enable autonomous navigation…for orbital transfer vehicles, orbital fueling, and in-space assembly.” Beachler agreed, “With AI-enabled satellites, you could have real-time collision avoidance and object recognition on the satellite, itself.”

Edge computing satellites with more compute resources could even gain the ability to work together towards a common mission. “NASA recently experimented with a swarm of satellites…multiple different satellites all operating as a single, cooperative swarm towards one set mission objective,” said Ge. “That's not really possible with today's paradigm, unless you put computing on these satellites and enable them to cooperate and coordinate with one another without having to data down to the ground.”

So, are these smarter, more compute-capable satellites making their way into space?

Anticipated, but untested

Edge computing is widely seen as an enabler for many advanced technologies – especially within the EO industry, where new imaging solutions are generating massive amounts of data.

“The newer generation of space companies, particularly those in very data-dense spectrums, like hyperspectral, thermal and synthetic-aperture radar are very active when it comes to [adopting edge computing],” Ge explained. “[They’re] figuring out they can't efficiently send all their data back to Earth. They need to process it on the satellite, and just send insights or organized files back.”

However, adoption could grow slowly as some satellite companies wait for edge computing technologies to become more mission tested.

“The biggest issue we've seen with [the adoption of] edge computing in space is that so many edge computing products are without a very proven track record that could [only result from] multiple missions and tons of data being accumulated,” Ge explained. “[Satellite operators] don't want to take this relatively new, novel, but untested technology and put it on their multi-million-dollar missions.”

In other instances, technologies that have long been used in terrestrial networks and data centers simply have yet to be hardened and tested for use in space – an environment where temperature, radiation and other factors are much more harsh and unpredictable than on Earth.

“[There is] typically a three-to-five-year lag [to take] what is commercially available today for terrestrial applications [and launch it into] Low Earth Orbit. That happens with many semiconductors that are going into satellites,” Beachler said.

Despite these challenges, edge computing in space would seem inevitable, if only for its ability to evolve satellites from spacecraft that become archaic and outdated over time into spacecraft that become better and more capable over time. As Beachler explained, “New AI models come out every day. From when you launched [a satellite] to its end of life, you could be updating the AI in that satellite hundreds of times. You're continually making the satellite better.”

Explore More:

What Role Could AI Play in Satellite Network Management?

Bringing ‘Earth-Like’ Computing Infrastructure to Space

PODCAST: Satellite Industry Softwarization, On-Board AI and Flying Routers in Space