View previous topic | View next topic

GPS & Packets

Page 1 of 2
Goto page 1, 2  Next

Tomas
775894.  Sat Jan 15, 2011 6:30 pm Reply with quote

Ok so I just watched QI History episode. I only saw a few minutes of it but they were discussing measuring time to high accuracy. After this, Stephen explains that GPS devices send a signal to at least 3 satellites and the time taken is measured. This is not true. A GPS device is a receiver. The satellites send a signal down and this is received by the GPS device. GPS devices do not transmit any data to the satellites at all, ever.

Next he said that transmission of data on the internet uses the same method of time measurement for reconstructing packets in the correct order. That is also not true. Packets have a frame number which describes the order in which they should be put back together. If you're using TCP, part of the process is identifying packets that got lost on the way and asking for them again. Packets are left to make their own way through the network so it is highly likely that they will arrive in a different order to that in which they were sent.

Edit:
If the data is transmitted by UDP then I believe the time is measured. UDP is used for realtime communication eg webcam, so if a packet arrives too late it is pointless processing it because it will be showing information that is no longer relevant. While this could be the point Stephen was making, I don't believe the show made the distinction.

Now, correct me if I'm wrong (but I don't think I am).

 
Posital
775902.  Sat Jan 15, 2011 6:55 pm Reply with quote

Tomas - you have it right - although I have no idea what his Fryness said.

 
Jeeves
775903.  Sat Jan 15, 2011 6:55 pm Reply with quote

You are quite right about GPS. I rarely find I have anything to question QI over, but Stephen's description of how the positioning system works was quite poor.

In fact, I believe every single thing he said about the system to be false. The receiver indeed does not send any signal to the satellites but operates quite passively by reading the signals broadcast from the satellites.

Also, a user's position is found by comparing the relavtive time-of-arrival of each of the satellites' signals and does not use a relay-type method Stephen described. In fact as far as I know the GPS satellites are quite incapable of communicating amongst themselves and are all controlled from the ground.

I believe he also mentioned that improved atomic clocks can give you better accuracy performance (1m was mentioned), again incorrect as most of the error in the GPS positioning is due to the signal passing through the ionosphere, and is not the fault of the atomic clocks at all.

Also, its 4 satellites needed for a fix, not 3.

As I say, I've never had reason to question the veracity of the show before, as it's one of my favourites. But I thought it was going to turn out to be a practical joke that every thing Stephen said on the subject was wrong!


Last edited by Jeeves on Sat Jan 15, 2011 7:00 pm; edited 2 times in total

 
Tuppy
775906.  Sat Jan 15, 2011 6:57 pm Reply with quote

I noted this gps boo boo too. Now people who think they know things because they saw it on QI will argue to the ends of the earth that their TomToms send signals to remote satellites, and that the therefore the CIA know where they've been.
Good grief, I don't think I've seen such a huge error on QI before.
Elves, you're sacked.

 
Posital
775907.  Sat Jan 15, 2011 6:57 pm Reply with quote

I thought it was three for a fix - and four if you needed height...

 
Tomas
775913.  Sat Jan 15, 2011 7:09 pm Reply with quote

In regards to 3 or 4 satellites needed for a fix, I believe it is explained here: http://www.maptoaster.com/maptoaster-topo-nz/articles/how-gps-works/how-gps-works.html


Last edited by Tomas on Sat Jan 15, 2011 7:14 pm; edited 1 time in total

 
Posital
775915.  Sat Jan 15, 2011 7:10 pm Reply with quote

BTW - improved GPS can be gained from WAAS/EGNOS which uses another satellite to communicate clock and ionosphere errors in GPS. Most GPS devices use this now.

 
Jeeves
775920.  Sat Jan 15, 2011 7:23 pm Reply with quote

*Edit:
I've just read the maptoaster article above, and it does a pretty thorough job of describing what I'm blabbing about :)
*

To fix your position with GPS you are basically making an observation of four unknown parameters, Latitude, Longitude, Height and Time (of your own local clock that measures the time of the signals). So you need four observations to fix four unknowns.

If you already have one of these (like knowing you are at sea level, or having your own atomic clock) then you may only need three observations to fix the others.

Mathematically speaking a 3-satellite fix puts you somewhere on the line of intersection of two hyperbolic paraboloids. This describes an elliptical curve in space, so even if you have your Height there may still be an ambiguity as the ellipse may cut surface of the Earth at more than one point.

Sorry to those for whom maths is a foreign language....


Last edited by Jeeves on Sat Jan 15, 2011 7:30 pm; edited 1 time in total

 
ponder
775925.  Sat Jan 15, 2011 7:30 pm Reply with quote

Tomas wrote:
Next he said that transmission of data on the internet uses the same method of time measurement for reconstructing packets in the correct order. That is also not true.


I believe the SDH (Synchronous Digital Hierarchy) networks underpinning it use atomic clocks to ensure the links are synchronized, and maybe take the signal from GPS. Since that technology was designed for the telephone days of constant bitrate streams (rather than packets), one link trying to feed data into another at a slightly higher rate than it was running at would be, well, problematic. Stephen's talk of packets was bizarre though, as that layer of communications doesn't need synchronisation.

 
tchrist
775990.  Sat Jan 15, 2011 11:57 pm Reply with quote

ponder wrote:
I believe the SDH (Synchronous Digital Hierarchy) networks underpinning it use atomic clocks to ensure the links are synchronized, and maybe take the signal from GPS.

Atomic clocks are neccesary but not sufficient: you must also take both general and special relativity into account to get the needed accuracy. Amazing but true.

--tom

 
Criffer
776151.  Sun Jan 16, 2011 12:22 pm Reply with quote

Tomas wrote:

Next he said that transmission of data on the internet uses the same method of time measurement for reconstructing packets in the correct order. That is also not true. Packets have a frame number which describes the order in which they should be put back together. If you're using TCP, part of the process is identifying packets that got lost on the way and asking for them again. Packets are left to make their own way through the network so it is highly likely that they will arrive in a different order to that in which they were sent.

Edit:
If the data is transmitted by UDP then I believe the time is measured. UDP is used for realtime communication eg webcam, so if a packet arrives too late it is pointless processing it because it will be showing information that is no longer relevant. While this could be the point Stephen was making, I don't believe the show made the distinction.

Now, correct me if I'm wrong (but I don't think I am).


You're quite right that TCP has no concept of absolute time, but neither does UDP. Both are layers on top of IP concerned with the treating routable IP packets as a single data stream The difference between the two is that TCP is designed to be lossless - packet loss is countered with retransmit at the TCP layer, whereas UDP leaves the question of handling packet loss or reordering to a higher layer. The TCP guarantee is that the data will arrive, intact and in order, although timing is not guaranteed. UDP guarantees that the data will either arrive intact, or not at all. Although UDP makes no guarantee of timing, it doesn't wait for retransmit or reordering.

When used for realtime media, such as the mentioned webcam, UDP is usually used as the transport layer for RTP, the Realtime Transport Protocol. RTP does include timestamps, but they are only relative to other packets in the stream, and do not have any notion of absolute time. The RTP timestamp is used in dejittering. Packets which arrive are held in a jitter buffer until playout time; if a packet is still not available in the jitter buffer when playout time comes, then the packet is marked as lost, and some strategy is needed to cope with the loss. For audio this usually involves synthesising audio to try to fill in the gaps and avoid high-frequency clicks which would be heard if there was just silence in that time period. For video, the lost frame data is concealed somehow, and the video must resynchronise somehow (either with a keyframe request or a periodic keyframe). The jitter buffer size is adjusted based on the competing requirements to a) wait long enough to receive any packet that is going to arrive, and b) minimise buffering latency.

In no situation involving UDP or TCP is an accurate clock required. High frequency transmission on a single link (such as Gigabit Ethernet) appear to require both ends to agree on the time, but this is not true. As long as the physical layer protocol has sufficient transitions to allow for clock recovery at the receiver (e.g. Manchester encoding), both ends can be running with very different clock sources. Clock drift is corrected by continually adjusting the tick rate based on the received transition rate. Data can be transmitted across different clock domains by inserting a buffer.

Stephen Fry seemed to be very wrong on how the Internet Protocol works, which is quite disappointing.

 
tchrist
776193.  Sun Jan 16, 2011 2:03 pm Reply with quote

Criffer wrote:
In no situation involving UDP or TCP is an accurate clock required.

Thanks for posting that summary of UDP, RTP, and TCP timing issues and non‐issues.

Criffer wrote:
Stephen Fry seemed to be very wrong on how the Internet Protocol works, which is quite disappointing.

Compared with the completely bizarre inversion of who’s the sender and who’s the receiver in GPS, I found the IP confusion only mildly disconcerting. (The things we call GPS units are always receivers, never senders.)

For one thing, Stephen’s a bright enough sort that I’m rather astonished he got the basic mechanics of GPS communication terribly wrong. I’m also quite disappointed that so major a guff went unnoticed by whoever the team was that put together the broadcast show using footage from the live one. That sort of blunder should have been left on the cutting room floor, not preserved to mislead posterity.

It almost makes you wonder whether Stephen’s Elf‐notes might have been the source of the error, and that because the notes and show agreed, nobody caught it during post‐production. Other possible explanations exist, but I still boggle that nobody on staff, be that Stephen or otherwise, spotted the obvious nuttiness of claiming all umpty‐gazillion GPS units sent active signals up to the satellites. It just doesn’t make any sense.

First, that setup could never scale, given how many GPS units there are that would need responding to. And second, it would require two messages per satellite (ground to sky and then back down again) instead of just a single one‐way message broadcast from the satellite to all listeners. Imagine the latency! There’a also the matter of power, given that I’ve seen tiny GPS receivers no bigger than a matchbox. Those little guys would never be able to power the uplink signalling required — they aren’t suitcase‐sized satellite phones!

For another thing, the botched discussion on atomic clocks and GPS was disappointing because it was the perfect occasion for Stephen to point out that we could never measure it accurately enough without taking Einsteinian Relativity into account. It really is Quite Interesting, being a real-world application of relativity, one impossible to get right otherwise. Those don’t come up very often, so it was a missed opportunity.

It’s also QI because it’s both kinds of Einsteinian Relativity. It takes Special Relativity to account for the atomic clocks on the moving satellites ticking 7 µsecs/day slower from the perspective of us here on the ground. And it takes General Relativity to account for the satellites’ clocks appearing to tick 45 µsecs/day faster than ours down here deeper in the more massive Earth’s gravity well, where timespace is necessarily more curved than it is further out.

Alas!

I’m betting Stephen himself would find the whole GPS–Relativity connection quite interesting, provided it were presented to him in an “Isn’t this neat?” sort of way, not as a “Boy were you a blithering idiot!” slap‐down.

--tom

 
Spiritman
776858.  Tue Jan 18, 2011 8:24 am Reply with quote

I think it was in an episode of Country Tracks where they featured a combine harvester, running on auto-pilot, which was accurate to just a few inches. At around £450K, I imagine part of the cost was for a guidance system much more accurate than the common-or-garden GPS we normally use. I'd be quite interested to know if it's the same system the military use.

 
Moosh
776994.  Tue Jan 18, 2011 1:07 pm Reply with quote

Hmm. To be honest my first guess wouldn't be to use GPS for that combine harvester. I'd use a system where the machine could "see" the fence around the field, then it's a fairly simple matter to get it to move around and cover the entire field. Alternatively you could just program it to just follow a set path, "forwards 1 mile, turn round, forwards 1 mile...", which again, doesn't really need GPS.

 
Neotenic
777005.  Tue Jan 18, 2011 1:42 pm Reply with quote

Actually, the GPS-guided combine harvesters were featured in Andrew Marr's 'Britain From Above' series.

 

Page 1 of 2
Goto page 1, 2  Next

All times are GMT - 5 Hours


Display posts from previous:   

Search Search Forums

Powered by phpBB © 2001, 2002 phpBB Group