Transcription
Very much, Tom. Okay, so to temper some expectations, basically all of the really interesting stuff that's going to be interesting to people who don't care much about nerdy control stuff, Tom has already revealed, setting some expectations right from the get-go. So, we soldier on anyway.
My name is Stewart. I am the ADCS engineer at Alba Orbital. I got asked at lunch, "What does ADCS stand for?" So, to anyone out there who doesn't know, it means Attitude Determination and Control Systems. Basically, everything that's in charge of, "Where is the satellite pointing? Does it know where it is? And and and what direction it's looking?" That, that is the sphere in which we're talking here.
Um, and one of the limitations, I'm sure everyone in the room is very aware, when you're designing any small satellite, one of the challenges, the primary challenges, is how do you get everything, something so complex, many people have mentioned it today. How do you get something so complex to fit into something so small? If you've read any of the, uh, ADCS archi- literature, academic literature for CubeSats, you'll see they're going on and on and on about how size, weight, and power are the big issues when you're designing something so small. So, when we're talking about PocketCubes, those, those impacts, those problems, those constraints are compounded.
Um, and the question then becomes, as Tom shared earlier, "Can we actually do it?" And it's PocketCube Conference's worst-kept secret, the answer is, of course, yes. But let me go into a little bit of detail about how we got there and what we did to do this.
So, this is the first downlinked, full-resolution image that we brought back. This was from U2, all active pointing from a PocketCube. These are the Austrian Alps, for anyone who's interested. And I would like to go into a little bit of detail about what went on behind the scenes of the ADCS to get us from images that looked like this, or this, or this, to something that looks a little bit more like the, something a little bit more interesting, a little bit more sellable, frankly.
And one of the things that, uh, I'm going to have to caveat is, I know the ADCS isn't a silver bullet, it won't solve all of your PocketCube problems, but I am biased. There was a lot of work that went into getting these satellites and all the different subsystems, the electronics, the optics, the thermals, all of it required and was necessary to get these images back. But as I say, I'm a very biased control systems engineer. So, that's what I'm going to talk to you a, a little bit today.
Um, so, having a look at where we were before last year's PocketCube Conference, actually before 2025, from an ADCS perspective. We had about 54 valid ADCS telemetry packets, which meant we had no insight into how our control system worked on orbit. What was actually going on. We had zero ability to turn the actuators on to actually rotate the satellite. The actuators are the components that, that do the turning, that do the pointing of the sat. We hadn't even switched them on at this point last year. And that was predominantly because we didn't have enough power to do so.
As you can see from the graph on the left there, that shows kind of our voltage and current values. We were sitting, bouncing around our, kind of, minimum voltage value of about 7.4 volts. After that, the satellite conks out. And on the right, you've got the temperature graph. As a few people have heard mentioned today, temperature's not so good for the batteries. So, we drop below a certain voltage, that below a certain temperature, the voltage just conks. We just cannot power our satellite if we're sitting at these voltage levels and at these temperatures.
And the result of the electrical power problem was, we couldn't detumble the satellite. It comes out of the deployer spinning every which way. And until, as it's been mentioned, until you can arrest some of that momentum, still the satellite, almost, you can't really do too much regarding things like GPS or, or starting to actively point in any direction, all of which needs some actuation.
Um, and then finally, most importantly for the, the Unicorn-2 satellite use case, was you just had, we just had a limited ability to take images. We just were constantly trying to keep the satellites alive, never mind trying to take images with the things.
So, where are we now? As of today, we have over 20,000 ADCS data packets. I thought that was going to be a really impressive number until Tibor mentioned his 2 million. Thanks for that, Tibor. But this gives us so much insight into exactly what's going on. We understand our ADCS way more now than we did this time last year. We, we see how it operates in orbit and we've been able to optimize based off of that information.
We've validated that we can detumble the satellites. This was, I believe, Unicorn-2 Q. You've got the purple graph, which is kind of the total spin rate of the satellite. You can see it comes out of the deployer spinning, you know, not too fast, and then over time it accelerates. This is because in orbit, there are, will be disturbances, there are torques that get applied to the satellite that gradually spin it up over time. And as you can see, it got to, you know, started accelerating quite quickly. But as we started to test our detumbling mechanism, the B-dot algorithm that we had on board, you can see it starts to come down to, eventually, at the end of the graph, you can see we're kind of bouncing around zero. The spikes, for anyone who's interested, are kind of where the satellite goes into night time, it goes behind the Earth, and we weren't detumbling at that point to try and conserve power. So, it spikes up a little bit through the night. But as soon as it, we saw the sun again, we were able to charge, you know, we brought the, brought the tumbling rate back down.
Um, and as a result, we solved the power problem. We have a positive sun-pointing power budget on our Unicorn-2 satellites as of now. As you can see, we're kind of bouncing around our, our max battery level of 8.4 volts, which we're very happy about. We spent a lot of time trying to get ourselves into, into that place. And as a result, we were able to take over 100 targeted earth-pointing images over the last, as Tom says, 12 months. So, there's been some serious progress made from last year's PocketCube Conference from an ADCS perspective.
And I want to talk a little bit about how we did that, what was involved. Um, and, and all of the different ways in which we improved the Unicorn platform. So, forgive me for taking us back to the classroom just for a moment. I promise I'm going to keep this as simple as I can. Anyone who's studied control theory will know this diagram well. This is a basic, very simplified diagram of what goes on in a classical controller.
On the right, you've got your satellite doing satellite things, spinning around, doing its, doing its stuff, having a certain behavior. The sensors then measure that behavior, that's the idea of them. They give a, a reading of that behavior to the controller. The controller knows exactly what it wants the satellite to be doing and uses the sensors to figure out what is it actually doing, compares the two, and then uses that to tell the actuators, "Right, you've got to tell the satellite to do something different," or, "Hey, okay, we're all good. Don't worry about it, man, you're fine."
Um, that is a, kind of, brief overview of how a feedback controller works. And it also gives me a nice little structure into which I can kind of explain the different sectors that we focused on over the last year as we were trying to improve the Unicorn platform.
So, we'll start with the actuators, a little bit. We, on board the satellite, have three reaction wheels. And these, these actuators do most of the heavy lifting when it comes to the pointing. Um, they are pointing in the three primary inertia axes of the satellite. And the major upgrades we needed to do related to these were predominantly software. We had very little precision control over these things this time last year. We were, you know, working to try and improve that, get more accurate reading of how fast these things were actually spinning. If you don't know how fast they're spinning, you've no idea how to control them properly.
Um, and also, we were unable to, you'll see in the maximum speed rating there, you'll see this can do plus or minus 12,000 RPM. We couldn't get the thing to change direction at all. So, a lot of work went into improving the, the software used to drive these things so that we increased the operational envelope of our reaction wheels and hence our pointing capability from that.
And then to assist the reaction wheels, we also have three magnetorquers, as you'll have heard these mentioned in a couple of the other talks as well. Um, and these are essentially just electromagnets, copper wire wrapped around an iron core. And they allow, what they're predominantly used for is detumbling. Everyone's been mentioning that. And the reason for that is they're way faster as an actuation source than the reaction wheels. They can respond to your rotation rate of your satellite. And then beyond that, they basically just provide an assist to the reaction wheels when they start spinning up a little bit too quickly.
Um, and what, what we did to improve these was a lot of work went into, with our manufacturing team, improving the manufacturing consistency of these, of these components to get them looking nice and pretty like this. Um, and then also, there was a lot of modeling went in, a lot of theoretical understanding to try and figure out exactly what was going on on a physics level, um, with these things. I'll explain a little bit about why that helped us going forward. But what it did allow us to do was give a brief evaluation of, "Are these things powerful enough?" Turns out they're way oversized for what we need them for. There's probably an area of improvement, optimization, going in the future to size these down a little bit, scale it back, and make things easier for our perm manufacturing team.
Onto the sensors. So, these are the three kind of basic sensors we have on board. We've got a three-axis gyroscope, also a three-axis magnetometer. There is an accelerometer, we don't use it. Um, and LDRs. So, the gyroscope essentially just measures the rotational rate of the satellite. The magnetometer is there to tell us what the geomagnetic field is at any given location and orientation. Uh, and then the LDRs give us a real rough guide of where the sun is roughly on this face. We have six of them on the satellite and they, uh, allow us to kind of start turning certain faces towards the sun. And you'll see why that's important in a moment.
But, the real star of the show regarding sensors for the Unicorn are these sun sensors based on the Hamamatsu PSDs. We've got the guys from Hamamatsu here. Go down and check their stall if you haven't already. Uh, and the reason these are useful is they give us a really accurate sun vector, um, by which we can orient the satellite, by which we can turn and, uh, point in a direction that we know we can trust. Uh, the problem with them is they've got a pretty narrow field of view comparatively. So, um, you need to get the sun in roughly the right area, and then these will give you, get you the rest of the way. They'll give you a really accurate fine pointing, uh, once the sun is in view. And we've two of these on the satellite, one on the same face as the solar panels, and then one on what we call the Z negative face, which is opposite where the camera lens is. Um, I'll explain a little bit about why that matters in a moment.
And then finally, we have our control modes. So, um, we already knew and had pretty high level of confidence in our detumbling control mode. We were pretty, pretty confident in it. We have an air bearing Helmholtz cage setup, uh, test setup at the lab. By the way, if you're interested in coming to see this, I would recommend signing up for the tours on Friday. Uh, come along, have a look. It's a lot of fun. Um, but basically, that allowed us to test our detumbling, uh, algorithm and make sure that we were definitely going to, uh, you know, arrest enough of that momentum I was talking about earlier to then start pointing, uh, in any given direction.
We also did, on board the Unicorn, previously have a solar-pointing mode, um, but it needed a bit of work to get it up and running, to get ourselves to a point where we were comfortable to use it. And that basically just involves pointing the solar panel sun sensors at the sun, funnily enough. It uses a direction cosine matrix-based approach, for anyone who's interested, to do that.
And then one of the things that we added, uh, additionally for this last launch, this last set of satellites, was what we called an anti-solar pointing mode. Uh, and this was basically just taking that Z negative sun sensor I mentioned earlier and pointing that at the sun. And what that allowed us to do was we could then time when we were taking images so that the sun, the satellite, and our target were in a line. And it meant that if we, we knew if we were pointing the back end of the satellite at the sun, the target should be pointing rough, or the camera should be pointing roughly in the direction of the target.
This was done in quite a rush before the last launch, um, to try and get something that was simple, something that was vaguely reliable. Um, but we did also realize that we would need to turn rather rapidly from solar pointing. So, we were charging right up until the moment that we decided, "Right, now we want to take a picture." We had about 30 seconds to get rotated round. So, we had to tune the controller so that it allowed, it allowed it to do that, um, you know, safely and without completely overshooting the whole thing.
And in order to, uh, make sure that we had some confidence in all of these control modes, uh, we, there was a lot of work done with an intern who came. He did some excellent work with us, actually. He developed this simulation that you're seeing the results of here. On the top, we've got kind of, uh, a demonstration of the B-dot detumbling algorithm. Uh, and then on the bottom was a kind of very initial, very coarse, did a lot of work, uh, nadir control mode, pointing the satellite directly at the center of the Earth. Um, and then on top of that, we also had our air bearing test setup, which we could also test these with, just to provide that extra level of, uh, confidence, that everything was as we hoped it would be.
But, as is always the way, there are always faults, there are always things that don't quite go, uh, the way that you'd hoped. Uh, and this was true, as Tom mentioned, for Unicorns O, P, and Q. Um, the first major one was our clocks drifted. We have two RTCs on board the Unicorn, one for the onboard computer and one for the ADCS. We experienced drift for both of them. That obviously meant we had issues scheduling when we were going to take pictures, um, and kind of meant sometimes we weren't 100% sure exactly what the satellite was pointing at, based on our, uh, quickly thrown-together anti-solar pointing control mode.
Um, we also had, we experienced what we think was an I2C failure, uh, on board, uh, Unicorns 2 O and P. Uh, and it was the ADCS clock and both of the sun sensors that were on this I2C line. Um, and we're pretty confident it was cold exposure that did for these ones. Um, the reason being is we left both O and P in the OTV for an extra few weeks after 2Q deployed. Uh, and it, it seemed a bit too much of a coincidence that both of the satellites that got left in space for 4 weeks, uh, both experienced this failure mode. So, we're, we're working on, uh, improving that for the future.
And lastly, uh, we saw a rather interesting, uh, failure mode in Unicorn 2P in that it spontaneously spun up to around 18 radians a second, which is pretty damn quick. Um, we aren't actually sure what caused this. We don't know if there was some kind of impact. We don't know if there was some fault in the, in one of the actuators, but we didn't witness anything as, after we managed to spin it down, that would suggest it was an actuator fault.
Interestingly enough, for anyone who knows what the Dzhanibekov effect is, we're pretty confident that within the, our data, we found evidence that that was happening to, to our satellites, which is kind of a cool side thing that we weren't really expecting to show. Um, anyone who doesn't know what that is and wants to, come talk to me after.
Um, but one of the things that, uh, the modeling of our magnetorquers allowed us to do is we also developed a mode that would, instead of what normally happens, which is we switch the magnetorquers off, take a reading from our magnetometers, and then switch the magnetorquers back on. That would allow us to, you know, get a reading of the Earth's magnetic field without interference from the magnetorquers. Because we'd done some analysis of the magnetorquers, we developed a mode where we could just read through it. We could just ignore what the magnetorquer was do- doing, almost, estimate exactly how much of an effect it should have on the magnetic readings, and so just read through and let the magnetorquer work 100% of the time. That increased its efficiency, increased its effectiveness, and was what enabled us to spin down from what was otherwise a basically catastrophic failure mode. We would not have been able to spin down from 18 radians a second otherwise.
Um, so that was a, a pretty cool, uh, side, side effect of what went down.
Now, you've heard a number of times, we have a couple of Unicorns about to be deployed in the next, well, next few days, really. Um, so, what are the upgrades that we've implemented for the ADCS for these satellites? Well, instead of that direction cosine matrix, uh, control method that we, I mentioned earlier, we've switched over to a quaternion-based one. The reason for this is it just allows us to ensure that we're definitely rotating through what is geometrically the shortest possible distance between orientations, rather than mucking around with different orthogonal reference frames.
Um, we've applied some of the satellite position determination. Up until this point, the Unicorn has had no idea where it is in space in itself. We've known, but it hasn't. We've known, but it hasn't. Um, and so on board the satellite is a TLE propagation method, um, and, uh, a GNSS chip, as Tom mentioned. And one of the great things about GNSS is the timing on it is very accurate. So, we're hoping that this provides us with a backup in case we experience the RTC problem, clock drift again.
Uh, we've also started implementing a static attitude determination mode. Um, it's a TRIAD algorithm, for anyone familiar. Um, basically, this allows the satellite to determine in itself where it's pointing. Um, but in order to do that, in order for it to know exactly what direction it's looking in, we need both, uh, a sun model and a geomagnetic, uh, field model on board the satellite so it can determine where the magnetic field of the Earth should be pointing and where the sun should be, given any particular moment in time. And then associated with that are many, many, many, many, many reference frame transformations. Anyone familiar with the maths will know exactly what's involved in that. I'm not going to bore you all with it, but it's in there.
>> [snorts]
>> Uh, and lastly, we've implemented an extended Kalman filter. Uh, another, uh, student at the university came and did his master's project with us for that over the last year, and he did a great job. Uh, so, we're not implementing this in anger straight away. We've got it on the satellites, uh, and we will turn it on, hopefully, but the idea is we turn it on and then it doesn't affect the control system at all. We just read from it and just make sure that it's giving us what we expect it to give us. And then the idea is, in the future, we, we use it properly to estimate the attitude of the satellite. And the reason that's useful is it is less reliant on, or less susceptible to noise in all of the sensors. The TRIAD algorithm's just not that great for that purpose.
Looking forwards into Unicorn 2T and beyond, I would like to do some more analysis of how accurate exactly we can expect these images to be. Um, you know, we have a lot of images that were pretty close, but not quite on. I would like to really dig deep into the theory of why that's happening the way it is. We've got a lot of sensors on board that need calibrating. We want to improve those calibration methods, um, specifically for the magnetometer and the sun sensors. Uh, and also, we have some pretty coarse estimates of what our inertia tensor and magnetic residual dipole should be for the satellite. I want to have a look at some methods that we could do where we estimate what those are based on the data we're getting back from the satellite, which would be a pretty cool way of improving the numbers we're using to run our tests on the ground, I think.
Obviously, employing the common filters. Um, we're also pretty keen on implementing a star tracker into the satellite fairly shortly. So, that's something that we're working on and looking towards. Cuz at the moment, as a, number of people have mentioned, if you haven't got a sun, at the moment, certainly with Unicorn, we haven't got an attitude determination method. So, um, instead of using our filters, we're going to try with a star tracker and see how we get on.
Uh, and as Tom mentioned, we're going to be improving the optical payload going forward, and that's just going to require a whole new ADCS system. So, uh, we'll be working on that as well.
I, I showed at the start, um, kind of the first fully, full resolution image that we downloaded from the satellite. This is the image of Toronto that Tom showed. Um, and from an ADCS perspective, it's probably arguably the most successful, uh, of our images. The reason being is that's pretty much bang on where we were aiming. That's the closest to the target that we actually got. Um, and it was the result of a lot of days of effort and work and estimating and trying new things and learning by failing. But it's not my favorite image of the lot.
My favorite image is the one that Tom showed, which was U of T Q's last one. And the reason I love this is because U of T Q was the satellite that was, without a doubt, our most successful from an ADCS perspective. It, with there was leaps forward in terms of what Unicorn 2 was able to do as a result of U of T Q. Uh, and I think this image just represents all the work that went into that, both in the past, and also the success that we've got coming up for the next set of launches. So, I figured I would leave us with that image. Thank you very much for your attention.