Truly astonishing amount of information stored
Inmarsat seems to store the complete communication it handles. This was not a debugging session but a postmortem analysis.
As air and sea searches continue to try and identify and retrieve wreckage from Malaysia Airlines MH370's presumed crash site in the southern Indian Ocean, Inmarsat has briefed Reuters on how it estimated the path the flight took. Since relatively early in the search for the flight, reference has been made to the seven “pings …
Inmarsat seems to store the complete communication it handles. This was not a debugging session but a postmortem analysis.
I would imagine that one ping per hour per aircraft is not going to require NSA-quantities of storage?
There is no information that this collection is limited to a ping handshake. The last attempted handshake was described as incomplete. I do not read it as telling that there was no reply from the airplane at all. Further analysis would not be possible w/o data.
I know the people who did this work, and also know a lot about the Inmarsat systems. Basically when Inmarsat receives a message over their satellites, the radio management system forwards it with an additional header that includes (among other things) the frequency that the message was received at. Inmarsat know the nominal frequency that is used for the messages, so they are able to work out the Doppler shift. The final step is to construct a flight-path model that reproduces the observed Doppler shift - this took a lot of number crunching but was relatively straight-forward.
@alannorthhants A really nice summary that should be in the article itself. Thanks.
Finally, some real information in the link from the article! Too bad it's still a pretty long chain from InMarSat through 2 other agencies to a Facebook page. I assume the Annex referred to is just the 3 diagrams?
But the chart raises several questions:
Why does the Malaysian release say that 6 pings were received after the aircraft disappeared, while the chart shows 7 (not counting the final partial ping)?
Why does the chart show measured data points that are clearly not hourly when all sources until now have said that the pings were on a regular hourly schedule at 11 minutes after the hour?
It looks like those first 3 data points after 2am local time are from a single ping message where the velocity changed rapidly throughout the ping, giving them 3 separate readings. They think it's because the aircraft was turning. How long does a ping take?
If that's the case, the Malaysians are correct in saying that there were 6 pings, and everyone has been wrong in assuming the pings were hourly. The interval appears to be slightly longer on average and slightly random.
As I understand it, determining whether the aircraft was on the northern* or southern track depended on the fact that the satellite 'wobbles' slightly north and south of its nominal geostationary location. This slight movement would have been additive on one track and the opposite on the other (and the magnitude changes between hourly pings). It's this analysis that has enabled Inmarsat to eliminate the northern path.
* The northern track was never very plausible, because it crosses too many dodgy areas where one would imagine a rogue radar return flying at 35,000 feet would have attracted attention. But then there aren't many completely plausible explanations of this event.
Thank you. I was about to post and ask how the Doppler Effect could determine whether it was going left-to-right or right-to-left from the satellite's viewpoint.
I don't see how that would work. Knowing the satellite wobble can only tell you about the motion of the satellite, not the motion of the aircraft.
Perhaps it's more to do with the fact that if the aircraft flies to the south it crosses over the equator, so initially it is pointing slightly more toward the satellite as it approaches the equator and then slightly more away after it crosses the equator. Whereas on the northern route it starts out north of the equator and continues north.
But looking at the geometry here, two things stand out:
There are several key assumptions required about what the aircraft actually did in order to correlate this analysis with a real world path. One can imagine other alternative aircraft paths that could fit the data equally well if it didn't actually fly north or south.
There are few data points, and the random motion of the aircraft in the air has got to be almost on par with the small velocity differences that InMarSat is trying to read. Where are the error bars on this chart? Where is the statistical analysis of likely this fit is?
I read that they compared the doppler shift with the known flightpaths of ``many'' other flights, and found the ones most consistent with the emasured results
just to point out that you can spot a large object under the sea from great height but be right over it in a boat and not see it - such are the vagaries of the fractal nature of waves.
Its not intuitive and can confuse people as to why its so sodding difficult to find a 20 meter long white object that someone dropped a marker on from a couple of thousand feet. The marker only needs to be wind blown a few yards away from the object and no amount of eyes will pick it up.
Not sure how many of the search boats have sonar. Should be easy enough to make cheaply these days?
So not exactly an unknown subject.
Doppler shift is given by 2 things.
Motion of the comm sat (known very accurately)
Motion of the plane.
Subtract out the frequency shift of the satellite and you know if the plane is going away or toward you, or possibly some more complicated path.
What's surprising I think is that Inmarsat actually collects that much fine detail about the signal, as well as the messages it transmits and receives.
Anyone who designs or has to support an RF system soon learns to keep as much debug info as possible. Things like frequency offset and power levels are important if you want to know why a link is going bad as it can point to a failing transmitter (if one customer is having problems) or a failing transponder or ground station (if all show similar problems).
As it is gather per-message/ping and only a few tens of bytes of data to append at most, it is hardly NSA-sized storage.
One system I worked on could keep up to a week's worth of 'raw' RF (i.e. as digitized and before any significant processing) on a cheap consumer-style NAS so you could go back and analyse any problems if they were reported promptly.
The doppler effect will given by 3 things - the third being the difference in the gravitational field which will be slightly lower for the satellite thus time will pass a bit quicker for it. Not much, but when you're getting into parts per trillion it may be approaching the arse end of a significant figure.
> Inmarsat actually collects that much fine detail about the signal, as well as the messages it transmits and receives
I have seen this explained, in that the spectrum is very expensive, and crowded, in that band, so there is a negotiation of frequency between the co-moving satellite and aircraft, such that the transmission occurs right in the middle of the band where it's meant to be. So there's a reason that the frequency is measured.
I just read that if the satnav sats don’t take gravity into account theres a drift of 7 miles a day?
In one sense I am not surprised that it does, because its easy to collect. And not very large. to store.
Its the sort of info you would store in order to trace and understand any system malfunctions. Frequency drift in the aircraft transponders for example.
Like log files on (internet) servers, there's a ton of stuff stored you seldom need, but when you DO need it is remarkable what you can reconstruct from it.
"What's surprising I think is that Inmarsat actually collects that much fine detail about the signal, as well as the messages it transmits and receives."
No, they just collect the precise frequency the signal was received on. That is useful for any diagnostic etc and is enough for doppler calcs.
Doppler is not just something GPSs have to account for (ie. it is not just something that gets in the way), it is also a very useful signal.
The doppler measurements themselves are a better source of velocity data that doing differences between positions. Doppler is also a useful input into filtering algorithms etc.
To the layman, measuring to 0.08% sounds pretty amazing, but GPS etc require far better than that to track satellites.
For all practical purposes, that gravitational difference is zero. We're dealing with a few ppm, not pptrillion.
Atmospheric distortion will have far greater impact that gravitational effects.
"If we ignore the Reuter's “19th century physics” angle, that's still an impressively accurate frequency resolution: taking the cruising velocity of the aircraft as around 250 m/second, its impact on a 1 GHz signal would be a change of around 0.08 per cent, even if MH370 had been flying directly at the satellite."
Where do I start on this?
1) As any 19th century physicist will tell you, Fo=FV(V +/- Vs)
2) So, the percentage change in observed frequency Fo is unrelated to the absolute value of the actual frequency F, it is simply (1 - V/(V +/- Vs)). 100%. No need to mention 1GHz at all.
3) V, the velocity of the waves, = c in this case, i.e. ~300e6 m/s. Vs, the velocity of the source, is about 250m/s. So, the percentage change is 8.3e-7.100= 8.3e-5 = 0.000083%. The article is out by three orders of magnitude.
I have to agree.
The Doppler effect is 0.83ppm (not 830ppm) if the relative velocity is 250m/s.
From the article's link to the Malaysian Ministry of Transport report, there is a nice graph at the bottom of the measured shifts of the different pings. The maximum shift is 250Hz for the last ping. At 1GHz this is 0.25ppm indicating a relative velocity between the satellite and aircraft of 75m/s.
As the satellite isn't quite stationary relative to the earth, the Doppler effect going north would have been slightly different to that measured if it was going south. See the graph, again.
So Reg, a fundamental schoolboy error!
Now you know how Stephen Fry feels.
Except that he is not a reporter on a tech publication.
The doppler calculations were easier than a Gatso one, no? Thought that Gatso guns used frequencies in the 10's of GHz range (24 rang a bell) yet were dealing with speeds as low as 30mph?
I don't know anything about how the Gatso actually works. But clearly there are other methods of measuring an objects speed other than Doppler. Ping it every 10ms and see how far it has traveled in that time, for example.
The Gatso takes two pictures a short time apart. You travelled 5 feet in 0.1 secs in a 30 mph area? (As measured by the markings in the road.) That'll cost yer!
But yes, it is triggered by Doppler radar; and the reflected signal has a shift representing twice the speed of the reflector.
You make the mistake of thinking that a frequency shift of 8 parts per 10,000 is somehow difficult to detect. In fact it's easy because, when combined with the reference signal, you get a difference signal in the megahertz range. That's what makes good old British fine-you-for-1-mph-over radar work. The effect is exactly the same as the one that creates "beats" when two musical notes that are close in frequency sound together.
The bigger mistake is thinking it's 8 parts per 10,000 when it's actually 8 parts per 10,000,000. See my post above.
El Reg used to report scientific stories much better than the national or local press. But their standards are slipping.
Shifting or Dopplering.
Exactly so. Inmarsat satellites MUST have a very stable very accurate clock source in them and by down mixing that and the received signal you up with a beat frequency which IS easily measured.
Its how all doppler radar etc is done.
Geo- "stationary" satellites actually do a small daily "tidal" figure of eight loop due to the gravitational effects of the sun and moon. This loop shape and timing vary over the month. This adds to the relative velocity equation, I don't know how much but this is the critical bit. (In fact, if you think about it, the doppler effect would be the same for mirror image north- and southbound planes with a truly geostationary satellite.)
It is unlikely that the exact satellite path is known accurately but this isn't actually required. The trick would be to scatter plot the frequency shifts for pings from different planes at around each time of interest against some actual known velocity components and find where the MH370 pings fit. You would have to eliminate the east west component. From the statements made it appears they used an approach of matching similar known flight paths rather than a fully mathematical/statistical all data analysis. My guess is that the effect would be small but detectable, with confirmation coming from correlation across the the six or so pings on the terminal leg of the MH370 flight.
Given the large distances I'm surprised the pings weren't picked up by more than one Inmarsat satellite?
Or have I missed something?
The Sydney Morning Herald reports the area offering the best prospects to locate wreckage is probably in the vicinity of the Southeastern Indian Ridge, a zone of underwater volcanoes that hasn't been surveyed for more than 20 years.
Nice quiet place for a lair then. Stromberg teams up with Blofeld?
Biting the hand that feeds IT © 1998–2017