Orbital Edge Computing: Software, Hardware, and System Support for Computational Satellites

 Decreasing costs of deploying space vehicles to low-Earth orbit support large constellations of satellites and enable new, Earth-observation applications that provide global, real-time insights. However, old operating strategies for these systems do not scale as satellites proliferate. Orbital edge computing, which filters data on satellites before transmission, addresses these new challenges with autonomy in orbit. Space systems need a new operating strategy with more autonomy at the edge as satellites proliferate. 

Edge computing is an emerging paradigm aiding responsiveness, reliability, and scalability of terrestrial computing and sensing networks (e.g., cellular and the “Internet-of-Things” or IoT). This work defines and characterizes orbital edge computing (OEC) — or, edge computing in space — as a means to address new system challenges arising from the proliferation of satellites in low-Earth orbit (LEO). 

Individual satellites have long experienced a downlink bottleneck: sensors onboard a space vehicle have the opportunity to observe more data per orbit revolution than can be downlinked per orbit revolution on average. A more-crowded LEO exacerbates this issue with contention among satellites for limited downlink time. Orbital edge computing autonomously filters observations on the satellite to prioritize high-value data for downlink and to use intelligent early discard to omit low-value observations before transmission. 

Because existing filtering applications execute slowly on existing and near-future satellite hardware, a computational bottleneck prevents direct deployment of such software to the space edge. This work presents a system (Kodan) to transform machine learning applications for a target satellite to increase the data value density of filtered, downlinked observations. This work also presents computational nanosatellite pipelines (CNPs) to address the computational bottleneck by distributing work across many nanosatellites. 

Satellites cannot recharge from a power grid and must harvest all energy from the environment. High?bitrate communication across hundreds of kilometers and autonomous analysis of sensor data both need significant energy. Thus, computational nanosatellites experience an energy bottleneck; harvested power limits OEC system utility. Subsystem scheduling within this physical constraint improves performance. 

Earth-observation satellites orbit at altitudes hundreds of kilometers above the planet surface. Volume, mass, and cost constraints limit camera sensor capability. As a result, orbital edge computing applications must be co-designed with the data quality of sensor observations for target objects. 

To evaluate orbital edge computing solutions, this work develops and releases the open-source cote simulation software. This simulator models orbital mechanics, radio communication, energy harvesting and storage, onboard computing, sensors, Earth rotation, ground stations, and other system parameters to evaluate constellation designs. This work also develops and releases the open-hardware and open-software Tartan Artibeus satellite as a platform for deploying and evaluating OEC concepts in space