AI-driven technologies are on the cusp of unlocking true autonomy in orbit, with the potential to enable spacecraft to operate independently and deliver more customized, intelligently managed data to Earth.
Terrestrial networks will also need to become more intelligent to handle the satellite communications market’s unprecedented expansion, driven by cheaper satellite bandwidth and smaller, more user-friendly terminals.
Software-defined revolution
The advent of flexible satellites with software-defined payloads over the last decade laid the foundations for bringing intelligence to orbit. A satellite’s capabilities no longer have to be locked in at launch, with fixed beam patterns and power levels targeting specific regions.
“The beam pattern was designed before launch, usually in the shape of one or more continents, with set power levels per beam,” said Quilty Analytics analyst Caleb Henry.
“If a new customer emerged outside of that footprint, that’s too bad. Or if a new customer emerged inside that footprint, but the beam was already saturated, that’s also too bad.”
Software-defined satellites, which can be reprogrammed on the fly, now make up a growing share of spacecraft orders, primarily from geostationary operators looking for flexibility to adapt as low Earth orbit (LEO) constellations, such as OneWeb and Starlink, reshape the market.
However, software-defined satellites vary widely in their flexibility.
At the basic level, there’s simple coverage redefinition, such as steering beams without altering their shape or capacity, said Nathan de Ruiter, managing director of Novaspace (formerly Euroconsult).
On the other end of the scale, full in-orbit reconfigurability includes the ability to adjust beam power and modify the radio waves used for communications.
Airbus Onesat, Thales Inspire, and MDA’s Aurora platforms meet the strictest definition of fully software-defined satellites, he added, though none ordered to date are expected to launch before 2026.
This timeline poses an obstacle for fully integrating advanced AI, noted Stuart Daughtridge, vice president of advanced technology at Kratos, which is providing the ground system for Thaicom’s first software-defined satellite, based on the Onesat platform and due to launch in 2027.
“To implement AI, you must already have a system and training data to be able to implement it,” Daughtridge said. “So since these [fully software-defined satellites really aren’t deployed yet, you can look at how you would implement the AI … you could do it with some simulated data, but you really can’t deploy it at any scale until you have systems up and running, and can get real data to ensure training.”
However, Daughtridge anticipates rapid AI development in orbit once more flexible software-defined satellites are launched, particularly to help manage the influx of available capacity.
“[I]f I’ve got to manage massive amounts more capacity, I need more automation,” he said, “and as soon as you start talking automation, you start talking AI.”
Software-defined satellites also require a more coordinated ground system to match their flexibility, Daughtridge added, with AI aiding in the management of satellite and terrestrial communications as a unified network.
Hands off the steering wheel
Today’s data and technology make it possible to leverage AI to control satellites in orbit, streamlining operations and reducing complexity for ground crews.
Canadian software startup Mission Control recently announced plans to test long-term spacecraft autonomy in partnership with Spire, which is providing a small satellite for a mission of at least a year to evaluate machine learning (ML) capabilities.
The payload, scheduled to launch no earlier than 2025, will include AI-powered software primarily to analyze imagery data from the satellite’s onboard cameras.
Notably, Mission Control also seeks to work with partners to analyze the spacecraft’s telemetry data, using AI to monitor the satellite’s health and ensure consistent performance throughout its operational lifetime.
“For a space exploration mission, there is limited data to train ML models before flight,” said Michele Faragalli, Mission Control’s chief technology officer.
“This means that the flight conditions might be completely different than the training conditions for such missions [so] re-training the model and updating it in flight, using data collected in-situ is an important step to ensure the ML model is performing as intended.”
The issue isn’t limited to space exploration. Spacecraft performance can degrade over time, and anomalies or unforeseen circumstances may arise due to changes in onboard performance or the environment.
Getting more out of GEO
Static beams from a non-software-defined geostationary satellite may illuminate an area at peak performance for 16 hours a day but leave capacity underutilized at night, affecting the operator’s cost per megabit.
Software-defined satellites ultimately help operators reduce total service delivery costs by increasing efficiency, said Cynthia Harty, senior vice president of corporate development at ground network specialist ST Engineering iDirect.
“And the ability to do that is all based upon this flexibility and utilizing those software-defined satellites as optimally as possible,” Harty said.
ST Engineering iDirect plans to release a software-defined ground system in September 2025 that would be capable of leveraging AI to assist network management tasks.
However, while outsourcing network management to AI could significantly boost efficiency, operators are generally hesitant to fully embrace these tools due to concerns over reliability and trust.
“Day one, we are not advocating automagically letting these two [ground and space network] systems just run unabated,” Harty said.
She envisions a “natural evolution where there are several use cases that are tested, tested again, and tested again before an operator is willing to flip that switch and have it be truly automatic and let the two systems talk together.”
In the meantime, she said ST Engineering iDirect is researching several anomaly detection algorithms for monitoring its ground networks, which could inform predictive analytics to protect against potent issues.
Technical Glossary
- Artificial Intelligence (AI): The simulation of human intelligence by machines, enabling them to make decisions, solve problems, and perform tasks autonomously. In the satellite industry, AI can help streamline operations, manage complex network systems, and improve data processing.
- Machine Learning (ML): A type of artificial intelligence that enables computers to learn and improve from experience without explicit programming. In space applications, ML can analyze data patterns to make predictions or optimize satellite performance based on real-time conditions.
- Predictive analytics: The use of data, statistical algorithms, and machine learning to identify the likelihood of future outcomes. In satellite networks, predictive analytics can help forecast potential issues, enabling proactive maintenance and improved reliability.
- Software-defined Satellite: A satellite with a flexible payload that can be reprogrammed after launch. Unlike traditional satellites with fixed configurations, software-defined satellites can adjust parameters such as beam shape, power levels, and frequencies in response to changing
- Beam pattern: The specific direction and area covered by a satellite’s signal. Traditional satellites have fixed beam patterns, while software-defined spacecraft can adjust these patterns to match evolving demand.
- Telemetry data: Data that is collected and transmitted by a satellite to monitor its health, performance, and operational status. This information is critical for maintaining optimal function throughout a satellite’s lifecycle.
This article first appeared in the November 2024 issue of SpaceNews Magazine.
