Carnegie Mellon University

Earth from space

April 13, 2020

Orbital Edge Computing

Krista Burns

As humanity moves into outer space, scientists and engineers must come up with innovative technology to get them there. In a recently published paper, Brandon Lucia, assistant professor of electrical and computer engineering, and Bradley Denby, a Ph.D. student, explore one such innovative idea: orbital edge computing with nanosatellite constellations. 

The paper recently received the Best Paper Award at the 2020 Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS).

A satellite constellation is a large group of satellites that work together. Constellations are widely used today with various applications including GPS. Today, however, satellites are not managed as a constellation, instead being individually remotely controlled by a human operator.  Satellites are directed by an operator to collect information using sensors (like taking pictures of Earth) and they then send their sensor data back to a radio base station somewhere on Earth.  Sending data to Earth from a satellite costs a huge amount of time and energy, and the model of manual control and sending all of the data collected on orbit makes today’s satellite operations extremely expensive. 

The cost goes up as satellites – now appropriately called nanosatellites – become smaller. Smaller satellites are cheaper and we can send more of them into space.  As we create ever-larger and more promising constellations, we also increase the human cost to manually control the satellites and the time and energy cost to beam their sensor data back down to Earth.  Growing constellations create a problem on Earth, too.  Together, the satellites of a constellation collect and send down an enormous amount of raw data. As the number of satellites increases, eventually, ground stations can’t keep up. The problem is in the operating model of satellites, which, today, expect data to be processed by computers in computing clouds on Earth.  As constellations hit the limit of data downlinking, Lucia and Denby propose that the relationship between satellites and computers must change.

Lucia and his group propose that, instead of sending the data back to Earth, satellites in a constellation process their sensor data in orbit, a new model that they call “orbital edge computing”. With orbital edge computing, the satellites don’t send large amounts of raw data to Earth. Equipped with machine learning capabilities, they analyze the sensor data, looking for signals of interest to any number of different applications and send down only the data that are actually valuable. Lucia’s team’s new software and hardware satellite control system supports orbital edge computing and ensures that a constellation avoids collecting redundant data. Using these techniques, the data size can be considerably reduced, and the data that’s left can be sent much faster and cheaper.  A constellation of thousands can be supported by a single operator and a few small radio ground stations on Earth to collect sensed data.

According to the paper, Lucia expects to see tens or hundreds of thousands of small, sensor-equipped satellites launched in the near future, which he says will be a perfect fit for the new orbital edge computing—and might even need it.