Using machine learning to chase astronauts

The radio connection is terrible with lots of hissing and crackling, but then suddenly a woman's voice can be heard: "This is November Alpha One Sierra Sierra". This series of words is the radio call sign of the International Space Station (ISS) in the USA. The voice is that of astronaut Shannon Walker, who is currently orbiting the earth at a speed of 27,500 kilometres per hour. Walker is talking to a class of schoolchildren via radio connection – yet in principle anyone can receive messages by the ISS if the space station is moving across the horizon in their area.

And in principle anyone can answer them, too: that's the plan of Brad Tarratt, who is playing the recording in his home office in Madison Heights, Michigan. Behind two screens and a tower of computers, we see a signal generator and a frequency counter, together with a small soldering unit and something that looks like a container of coloured paint. "That's my lunch," says Brad and laughs – today it's chicken sausage gumbo, a stew made with a spicy roux that is typical for the southern states. Brad is an amateur radio enthusiast whose other hobby is astronomy, and he is also team leader for Embedded Software Engineering at umlaut. He describes his job profile as "Swiss Army Knife" – because he "can do pretty much anything", something his colleagues will testify to. Essentially, he designs software for computer systems that are embedded in a larger cosmos of devices. For example, the sensor systems in a car, or the control units in a test rig.

A telescope made from things bought at the DIY store

Chasing astronauts is actually his hobby, although it's never that easy to completely separate Brad's work from his free time activities. "It probably started with binoculars and a lounge chair." From this vantage point, Brad observed the stars as a child of just three or four years old. At the age of five, he got his first computer, and when he was ten he got a dusty grey Intel 286 which he programmed a music synthesiser for. When he left school, he spent some time doing house clearances of old properties, repaired computers in an electronics shop, and then got a job at a university, teaching the employees of an automotive factory how to use a PC. "I taught at a college for a while, never went to one, though."

In his mid-twenties, Brad got a job working in network operations at a cable company, later he moved to a phone company: "I was doing break/fix all day: my website is down, my phone service is not working. I learned stuff, but I had no creative input into the process of my job." To make up for this, he spent his free time in the i3 Detroit hackerspace where he built a Dobsonian telescope – a room-high telescope made of plywood, a cardboard tube usually used for pouring concrete and a couple of optical instruments. He taught himself how to operate radio equipment and started designing a device that could determine the direction that certain radio waves are coming from. He never finished the project but did spend a fair amount of time focusing on one component: software for the "BeagleBone" single board computer that enabled transmission of radio signals from the BeagleBone’s pins. Which he found could – with very precise timing – be used to alter the time on radio controlled “atomic clocks”.

From amateur radio project to a test rig

This BeagleBone was in fact used some years later in the prototype of the "umlaut CAN tool", a hardware module that simulates the output of various different components on a test rig – for example, the components of a car. Brad was heavily involved in its development; in the current version, the BeagleBone has been replaced by a custom printed circuit board. Most of Brad’s skills had originated from hobby projects in this way, a personal exploration of some idea that seemed fun at the time.

His work on the astronaut tracker was also this sort of project: "I learned there’s the possibility to talk to an Astronaut in space – the second I heard it’s something you can do I wanted to do it." Brad wrote a script that downloads orbital data of the ISS and then calculates when it is next due to pass over Madison Heights. The radio unit is then automatically switched on and tuned in to the right frequency. This requires constant adjusting for doppler shift as the space station crosses the sky, due to the incredible speed at which the space station travels. Another problem: the ISS team usually only transmit messages for contacts that have been pre-arranged, such as Shannon Walker's link-up to the school. Astronauts don't often reach for their radio devices in their free time. “When an unscheduled contact will happen is unpredictable, so you just have to be listening," says Brad. The tracker was added to the collection of unfinished projects.

Perhaps a little cosmic assistance was needed to tie up the loose ends in Brad's life. One evening in 2014, there was a clear sky and he set up his huge red Dobsonian telescope next to a telephone mast, directly in front of the garage entrance of the i3 hackerspace. After a full work day, his plans for the evening were to show the telescope to a couple of friends and perhaps look for the Ring Nebula in the constellation of Lyra. This is considered to be a challenge for city-dwelling amateur astronomers because so little light is received from it in comparison to, say, the planets. So they got started: down a bit from the bright star Vega, then close off the field of view a little. And there it was, the Nebula appeared in the corner of our eye, like a ring of smoke. "Absolutely amazing," says Brad. And on that same evening, there was also something else that was amazing: he met his future wife, fellow i3 member Nikki. Who as a child had gone skywatching with her dad using a self-constructed telescope that had looked an awful lot like this one: a huge, red Dobsonian.

Image recognition for the voices of astronauts

The fact that their daughter, who is now five years old, was named after the constellation of Lyra is no great surprise. But the fact that, just a short time later, Brad got into a conversation in the i3 hacker space with Nate Bezanson, a hardware designer from umlaut, makes us think that there had perhaps been a little cosmic assistance. He saw that Brad "seemed to be able to do anything" and was constantly learning new things – so he brought him along for an interview.

What does he like about his work nowadays? "I never start the day with a bad feeling," says Brad. He can use the things he learns in his free time in his work – and vice versa. When he was developing a tool for one of umlaut’s customers that would follow the eye movements of car drivers, this resulted in him finding the missing puzzle piece for his astronaut tracker. "We were able to track eye movement using machine learning: Are they looking at the road, the phone, out the window?" says Brad. The same principle of image recognition can also be used on diagrams that Brad can generate from audio waves. With 15,000 sound snippets from radio messages from the ISS, Brad was able to train a neural network. The result: the script was now able to differentiate between noise, human speech, and image transmissions from the station, and thus automatically alert him whenever it heard anyone speaking. “I was myself very surprised at how well it works. In fact it always works.”

Brad's astronaut tracker has been in use for a good year now; he gets a Telegram notification on his mobile phone whenever there is a radio message that is within range. He has already received quite a few, mainly pre-arranged link-ups with schools, such as the cosmic greeting from Shannon Walker. There was one single unplanned connection that he missed – “I’m pretty sure I was in a meeting.” Since then, he has been hoping for a further opportunity, before the ISS is taken out of service, which could as early as 2028. There remains only one question: what would he talk about if someone up there were to answer his radio call sign? “To be honest: No idea,” says Brad and laughs. He hopes that his daughter will be nearby if this happens. “Maybe she can do the talking.”