I was in the audience for these presentations last week, and I sat down with Neinavaie and Khalife for a couple hours to discuss their work. I’m also engaged in very similar research myself, which I presented at the same conference.
So... not directly affiliated with the original authors, but very familiar with the work. Can I answer any questions people might have?
Accuracy will depend on whether/to what degree SpaceX commits to providing a positioning, navigation, and timing (PNT) service.
The technique used in the article is quite powerful and general, but it does have certain drawbacks too. One advantage is that it is relatively insensitive to errors in the ephemeris. The path we are exploring relies more heavily on the ephemeris, as you speculate; but these can be really quite excellent, even in the sub-meter range. Actually the bigger source of uncertainty for absolute positioning using non-Doppler techniques will probably be clock errors. For Doppler, integration time/time-to-first-fix is a sticking point, unless you have 8+ satellite signals, IIRC.
We think accuracy in the low single digit meters is not too much of a stretch (note that I am making a claim about a rather different technique than the one in the article... nothing worth doing is ever totally straightforward!).
I understand that SpaceX ephemerides they publish are more accurate than the typical two-line element set. But the TLEs using the SGP4 propagator typically have errors on the order of 10s to 100s of meters [0] (and more, the further you are from the ephemeris time). Is the estimation really enough to get to down to the meter or sub-meter accuracy? Is that why it really requires 8+ satellites to get a good fix, i.e. it takes that many measurements to reduce your variance and GDOP?
I’m assuming they needed to have very accurate orbit information for each satellite. Was this provided to them by SpaceX, or was it derived from the signal?
Yeah, I think they said they used the TLEs. There are better ephemerides posted by SpaceX on space-track.org under contract w/ the govt, who doesn’t really want to be responsible for tracking all those objects.
2. I would not expect that to work very well without an IMU, but you can do some pretty remarkably good odometry using wheel rotations, vision, and/or automotive radar if you’re dead set against using an IMU.
3. I think they were measuring instantaneous Doppler and integrating to get phase. They only looked in a 1 MHz passband, and they didn’t have a very directional antenna, so they must have been acquiring something pretty narrow-band compared to the full downlink packets.
Processing time would also be highly dependent on what the processing was done with and once the algorithm was established it could be ported to something less generic than Matlab. For example, an FPGA or ASIC for potentially quite short/low-power processing times in not a lot of volume.
The article states they used a small antenna to detect the satellite signals but how small is small? After all, the SpaceX "Dishy" is fairly large, heavy and power hungry. Would this technique really be possible with an antenna small enough to strap around the neck of an animal? What about something that could fit in my phone?
You can do satellite comms on something that small (maybe an antenna a few centimetres square) if you're only interested in moving a few bits per second or maybe kilobits per second. In the transmit direction you're generally limited by how much power you can pump out (the smaller the antenna, the less antenna gain you have and the more power you need to produce from your transmit chain). In the receive direction, you have a figure called G/T which has to do with the ratio of the antenna gain to the noise temperature of the receiver chain (which includes thermal noise picked up to the universe, plus noise introduced at each gain stage etc.).
The simpler the modulation, the less power you need generally, but you can move less data in a given amount of frequency spectrum, so that's the tradeoff.
Their antenna was a CotS LNBF. Think just the bit on the end of the arm of a DirecTV dish, not the dish itself. Typical aperture maybe 10cm, full unit about hand-sized.
"Dishy" size is because its a two-dimensional phased-array antenna. It has a lot of small antennas on it to steer the energy released from the antenna directionally through interference patterns. If you're only going to receive and the signals are powerful enough you don't need that big of an antenna to receive signals at these frequencies.
The particular signal they used can be captured with a $40 low-noise block w/ feedhorn (I’m using the Visiblewave Hybrid LNB), a 12-16 VDC biasing circuit, and a garden-variety SDR. I speculate, based on my own studies, that the signal they caught is the Starlink “Idle” tone. I don’t see that signal when the downlink is busy in my own experiments, but I can’t be certain yet. Different latitude, different equipment, different assumptions.
The broadband downlink signals I’m interested in require a bit more effort— I’m using a 75cm offset parabolic dish on an azimuth/elevation turret with some nice high-end servos w/ built-in 12-bit encoders. Up until yesterday, we could only capture 60 MHz of real-time bandwidth; but now we have the device we need to capture closer to the entire 2 GHz span of the 10.7-12.7 GHz downlink in one fell swoop.
With all that, I still only get about 11 dB of SNR. I’ve seen hints that I might be able to get closer to 20 dB, which I could trade off for a smaller dish.
11 dB is pretty impressive, especially if you are just looking at the Doppler shift in the carrier.
Which actually begs the question, can you decode anything (e.g. satellite ID) from the downlink/idle signal? If not, how do you figure out which satellite a given signal came from?
My first guess would be that you can fit an orbit to the Doppler shift, find the closest TLE set from space-track.org, and assume that the signal came from that satellite. Then propagate the orbit to the observation epoch and that should give you a fairly accurate guess of the satellite's actual location. Or do you do something completely different?
What you describe with finding the closest TLE is pretty much what they did. We use a highly directional antenna (~1-2 degree beam) and some tracking software, so it’s not ambiguous.
I’ve been able to demodulate and reconstruct about 100 subcarriers’ worth of signal, but until recently I couldn’t record the entire bandwidth. Now we’ve got our hands on one of those awesome RFSoCs with 2 Gsps ADC, so it might be possible to decode more. But unknown error correction, bit interleaving, so it’s a tall order. We’d have to get lucky.
I’ve asked them about this, but they said they were too busy rolling out the internet service to consider any alternatives at the moment. I’ll keep hounding them until they relent!
> These orbiting internet nodes fly closer to Earth’s surface than the satellites in the government-run global navigation systems, such as the U.S. Department of Defense’s GPS, the European Union’s Galileo, China’s BeiDou, or Russia’s Global Navigation Satellite System. Because militaries operate many of the systems, the signals can be targets for jamming or scrambling—spurring researchers to look for alternatives.
> Because the satellites of internet megaconstellations fly so much closer to Earth than those used in traditional GPS, their signals can be thousands of times stronger. This could make them easier to pick up in dense jungle environments where GPS sometimes struggles to determine position, Langley suggests. The brighter signals might also allow for better data collection of animals outfitted with tracking collars, he adds.
My guess from the article language is some kind of doppler shift analysis compared with ephemeris data and ground reference stations? Is that correct? Is it better than 100m?
>By measuring this effect for six Starlink satellites flying overhead on different trajectories, the scientists could pinpoint their location on the ground to within 7 to 8 meters
Is GPS a general synonym for GNSS these days? I was always under the impression that GPS is just one specific GNSS among others like Galileo, GLONASS or Beidou. This article is describing a new concept for a GNSS system that has no relation to GPS? Or what am I missing?
GPS stands for 'Global Positioning System', but if you instead think of it as 'global positioning system', then all these satellite systems are GPS, because yes they are literally all a 'global positioning system', they're just not the 'Global Positioning System'.
That's what happens when you give something specific a very generic name. People shouldn't do that.
And on the phone, it might as well be cell networks or other sources that tell you the position - not GPS - yet people call it GPS. The phone calls it location service.
This might be a US thing, but probably even wider. "GPS" seems to have taken on the same effect as "kleenex" here. What makes it frustrating is that you can still do research in academic papers, white papers, etc on "GPS" alone!
>SpaceX wasn’t involved in the research and can’t do much to try to block such techniques, though it’s unclear whether it would try.
well, government can always mandate to add artificial randomly floating frequency shift to break that Doppler based calculation to supposedly block the Taliban's ICBMs from precise navigation.
I understand that Russia's "GPS" system ( https://en.wikipedia.org/wiki/GLONASS ) has been fully operational for a while now. The Taliban (or North Koreans, or others non-friends of the U.S.) don't need to count on U.S.-run geolocation services for their Important Stuff.
SpaceX will (and does) have good coverage but the accuracy will not be to anywhere what people have come to expect from GPS. At least not without major design changes to focus on this as a product from SpaceX.
If you know the orbits of a few satellites, and you know how those sound from your current spot, you can roughly figure out your location. This isn't a new concept in the slightest, scientists were doing it in the 1950s.
Ultimately though, to get extremely accurate positioning data you need to be very synchronized (like, atomic clocks?). You'll also need updated data of where these satellites actually are as things flying tend to drift. This is especially true of things in LEO.
Is this the "differential" technique in which you have a number of ground stations observing each satellite, so you can compute the positions of the satellites, and the device that wants to know its own position both observes the satellites and communicates with the ground stations so that it knows where the satellites are? (Obviously the communication with the ground station doesn't have to be low-latency or line-of-sight. If you had line-of-sight communication with the ground stations you could just use them instead of the satellites!)
Or does this work without the "device" having to communicate with ground stations?
This is really interesting, if SpaceX allows organizations to subscribe to higher levels of GPS-like accuracy, I wonder if it would shut down other country’s attempts to launch their own GPS competitors.
On the other hand, I can imagine the Air Force not wanting this to happen, as high precision location data can be used for missile and ICBM delivery. However, if it is an inherent property of the low orbit network, there is no point in blocking it from consumers.
Good question. Also consider that most of the threats to GPS are ground based jamming. It is fairly easy and cheap to make a jammer out of hobbyist parts and knock out a large area of GPS coverage. However, just because someone jams GPS for an entire city doesn't mean that they are able to jam the LEO satellites flying over that city.
The Pentagon is (plausibly) Starlink's by-far most valuable potential customer. It seems pretty safe to assume that Starlink was designed to keep working after anti-satellite weapons, jammers, hackers, etc. knocked out GPS.
I don't have any concrete knowledge to say Starlink satellites rely on GPS, but based on what I know about satellites with thrusters and autonomous formation flying capabilities, they almost certainly have one or more GPS receivers onboard.
My understanding is that the SpaceX ground stations uplink the orbital parameters of other satellites and/or debris and it is mostly up to the satellites to avoid a collision. You need very precise onboard location knowledge for that. Also, I bet they are using GPS time to synchronize the fleet.
Of course you can still synchronize and maneuver the satellites manually from ground stations if necessary. And they probably do that when a satellite is misbehaving. But overall the Starlink operations concept is probably tailored to mostly automated operations.
So IMO they would definitely have big issues if GPS went away suddenly.
> My understanding is that the SpaceX ground stations uplink the orbital parameters of other satellites and/or debris and it is mostly up to the satellites to avoid a collision.
That is almost certainly not how that is implemented. Why would they run the orbit maneuver planning in space when they could do the whole thing comfortably on a simple server PC on earth?
> But overall the Starlink operations concept is probably tailored to mostly automated operations.
I agree with that. They are most definietly not hand-flying their birds. But this autonomy is most likely implemented on the ground.
> That is almost certainly not how that is implemented. Why would they run the orbit maneuver planning in space when they could do the whole thing comfortably on a simple server PC on earth?
That's a great question, but SpaceX is, apparently, doing much of the processing onboard[1].
Probably because the purpose of SpaceX is to build a city on Mars, and Starlink is a prototype for a communications and positioning system to be put in place around Mars before there are any facilities on the Martian surface. Remember this is an Elon Musk company.
It works up to 3000km with full precision, you just need a receiver without CoCom limits. (actually, precision should be better without atmosphere). It also works all the way up to geostationary orbit, but with some limitations.
The commercial GPS receiver are just software disabled not to work at certain speed and altitude. There used to be a receiver you can buy that does not have those restriction. You can also build your own GPS receiver and not have the restriction.
GPS orbits at 20,200km from the surface.
Starlink orbits at 550km from the surface. So, relatively, they're basically on the ground, just moving really fast!
I assume regulations might prevent this but technically I guess Starlink could add a little bit of extra data to their signals and make a pretty decent GPS alternative?
Can’t speak to the regulations, but essentially the answer is “yes”. Our publications can be found under the moniker “fused LEO GNSS”. Those of Zak Kassas and his students can be found on their website.
So... not directly affiliated with the original authors, but very familiar with the work. Can I answer any questions people might have?