Indian Space Program: News & Discussion - Sept 2016

[vina]

The accuracy required (in both stable frequency and latency (wrt to "standard" TAI) is about MILLION times more than what is typical needed (or achieved) on a typical computer or a cell phone.

Yes. What I am trying to understand is WHY is such accuracy required .

(*** side question *** then how does GPS on your cell phone work? if it does not have such an accurate clock ***? (I generally ask this question to test really top students to see if they can reason it out without looking up))

I was never a "top student" anywhere , but let me try it all the same.

Hmm. Normally I would have answered that it is "easy" . We know that what the GPS measures is time delay of arrival of signal. The satellite broadcasts a signal identifying a) Itself b) It's location in space and c) A timestamp when the signal was sent out. Now the receiver looks up the timestamp , and finds out the time at which it received it , and hence calculates the time the signal took for the signal to propagate from satellite to receiver and from that knowing speed of radio waves , we calculate the distance of satellite to receiver.

Now do this same from 3 satellites simultaneously, we will know the distance of the receiver from the 3 satellites, and since we know the positions of the satellites, the distance between the satellites themselves are known , and from solid geometry with the 3 satellites and the receiver forming a "prism", and since we know the measure of all the sides, we can locate the receiver exactly in 3d space.

Actually your Garmin or GPS on cellphone's clock is not synched with satellite.. in fact there is not even a very precise clock there.

And this is what made me realise that it wasn't as "easy" as I posted above. So, I put pen on paper and thought of something like this.
The satellite clocks are synchronised, and the signals they transmit are continuous and time synchronised as well . So, if the receiver is getting signals from 2 satellites, depending on the distance, if the time of arrival, of one is t, the other signal will lead or lag a bit. What we can measure by comparing the signals is the lead /lag in phase .

If the two satellites are A & B , if the receiver is at position C, since we know the distance AB , and we know that both the satellites have be visible on the horizon , we can solve numerically given the triangle ABC ( A,A + c*delta t, AB) , imposting the condition that angle ACB < 180 deg , and from that we get the distance from the satellites A & B .

So let me post my question again - if you have not heard this question before --- How does GPS on a cell phone can get your position even though it does not have an atomic clock??

Hmm. What am I doing wrong here ? I don't seem to need an atomic clock , or even a clock at all in the cell phone, nor do I need to know what the actual time at all ! All I need to know is the speed of radio waves, which is a constant!

This is not exactly true, as not only time but also the position of sats calculated by ephemeris (calculating orbit with very high precision) comes into play. For that one has to know the 'true time' to calculate the orbit.

But why would the satellite need an atomic time to find it's own position ? Theoretically, the satellite can have a "GPS receiver" on board and get signals from GPS ground stations that are visible , sort of inverse of how the GPS receiver on earth gets it from the satellite without needing any accurate clock (or even any clock possibly), and it can very accurately determine it's position in 3d space , just like the receiver does wrt to the satellite constellation.

All that it seems to require is that the clocks in the satellites are synchronised between themselves and not necessarily in absolute time with the "master clock" / atomic time on earth.

[putnanja]

If the times are off b/w receiver and sender, won't the distance be miscalculated? If time diff b/w satellite and cell phone is say 2 secs, that means the distance that is calculated will be wrong, and so is subsequent triangulation point? And so the triangulation spot will move off the actual location??

[shiv]

In practice it is the preciseness of the clocks which is the basis to measure the distance. You find the distances by measuring the time difference between two accurate clocks (not the other way around !!).

<snip>

Similarly frequency shifts (doppler shift as some people pointed out) etc depends on relative velocity -- again high accuracy is required and the accuracy in practice is measured by assuming the accuracy of the clocks..

****
Shivji - yes the clocks are precise enough that even the relativistic effects on slowing of clocks needs to be taken into account.

Thanks - I knew there was some issue in there. I also recall reading (and I put it in a link earlier) that the satellites carry clocks that are orbiting at thousands of kmph and that would make those clocks slow down relative to earth clocks due to relativity. I guess that would also require correction for accurate distance measurement

[shiv]

If the times are off b/w receiver and sender, won't the distance be miscalculated? If time diff b/w satellite and cell phone is say 2 secs, that means the distance that is calculated will be wrong, and so is subsequent triangulation point? And so the triangulation spot will move off the actual location??

Let me do some guesswork here.

For each satellite that the cellphone communicates with - the distance is dependent on the time taken for the signal to travel to the satellite.

But because the satellite clock is travelling at over 10,000 kmph it has slowed down relative to earth. That means that it will report receiving a signal at the wrong time and that wrong time gets sent back to the cellphone - doubling the error as the signal goes back. So (I am making a guess here) the satellite should have some "awareness" of the time factor by which it is reporting the wrong time. I guess that this "awareness" can only come from a periodic signal fed to the satellite from a very accurate earth-bound clock as to what the degree of error is. This is what vina has suggested (If I have understood it right)

So the satellite sends back could be a "corrected" signal compensating for the error.

[Lilo]

OT
I have never been initiated into the theory of relativity .Have zero understanding of it.
Can guru's point to an online physics lecture or offline book resource along with a basic problem set(like of irodov) for understanding of relativity ?
/OT

[shiv]

I know nothing of the theory of relativity - however as a schoolboy I was exposed to classic physics books by people like Perelman and George Gamov which I still treasure but regret are of no use to today's generation

If you get on a spaceship that travels very fast - maybe even 1/10 or 1/2 the speed of light your time/clock slows down while that of earth remains the same. So if you travel for 2 years and come back - you may find that earth time has travelled forward by 100 years and even your children have passed away while you have aged only 2 years. This sounds like rubbish but it has been proven - though not with humans.

There are a couple of other things that happen that I cannot explain or understand - but not relevant here. Your mass increases as your speed increases and becomes infinity at the speed of light. Also your length in the direction of travel becomes shorter and shorter as you go faster and becomes zero when you reach the speed of light. Do not ask me to explain

The long and short of it is that the atomic clocks that are travelling fast relative to the earth slow down compared to earth clocks. That means that given enough time if the earth clock is showing 8 PM the satellite clock will be showing 7:59 PM and 59.999something seconds This makes a huge difference in distances over 1000s of km

[LokeshC]

Actually if you change the reference frame to the person traveling fast (i.e. he is stationary and earth is moving back fast from him) he will also find that the clocks on earth have slowed down.

Relativity is mutual.

Its an interesting puzzle that took me quite sometime to grasp and even more time to convince myself that there are no inconsistencies.

[sudeepj]

again major inability to simply read before ranting using misplaced ego. I had given no solution, i was only asking the technical reason why a slave vcxo is not possible with ground-based master oscillator, there must be a good reason but i was asking what that reason was.

But seriously when you misdiagnose a deflating ego as anger, and being a hakim trying to argue with a hard core engineer -about engineering, is like me explaining you the dangers of FGM and why you should not be doing it.. So kindly shoo away, because in engineering you are a child talking to the big daddy.

PS: Bw of a vcxo is narrow.. Which is control system-101. And you think its "mumbo jumbo".. Too funny

In theory, this is possible.
1. As long as you have four signals originating from known locations, that have internally consistent clocks, you will be able to triangulate your position *and* derive the internally consistent clock.
2. This is simply the mirror image problem that GPS solves. In a normal GPS, you use four satellites with internally consistent clocks, to triangulate a position on the ground. In the mirror image problem, you will use four ground control centers with internally consistent clocks to triangulate a position in space. There will be some relativistic effects you will need to account for, but not a big deal.

Here are the problems:
1. When GPS was invented, the idea of being able to shoot down/disable satellites was laughable. Ground control centers on the other hand, will come under attack. If your ground control centers are disabled, there goes the entire system.
2. The accuracy of the atomic clocks is much much higher than what can be achieved with a controlled XO. They shift on very short time scales, because of temperature, acceleration, and sometimes, just because they want to. You will never be able to get an XO frequency that stable. There may also be unknown physics involved here. Who knows what unknown behavior the XO will present when in orbit.
3. As the Satellites move, they will move in and out of the view of your ground control centers, so you will need to maintain multiple ground control centers all over the globe, with technical staff to maintain them. At the very least, every one of your satellites must see four ground control centers at all time. More for redundancy and a good geometry (basically, you cant have four centers all located in Mumbai, they have to be separated far apart for the position fix obtained by the satellite in space to be a good fix with only a few meters of error). This may not even be physically possible.. consider the Pacific ocean and the poles.. and vast physical expanses controlled by hostile powers such as China.. Only very specific latitudes are inhabitable.. and vastly fewer accessible.

Its vastly cheaper to simply add a few more atomic clocks to the satellite for redundancy, than to maintain multiple ground control centers. A satellite qualified atomic clock will cost only a few hundred thousand dollars, or a few million at most.. spread over 10 years expected life of the satellite. A fraction of the cost of running a full ground control center. As navigation constellation satellites are built in numbers, the cost goes down. The cost of the atomic clocks is insignificant compared to the cost of the entire system. There is no point in introducing a lot more complexity into an already complex system to save a few million dollars.
4. ISRO perhaps messed up buying the atomic clocks from a European company, they should have used American sources who have much greater experience and much longer track record. As far as I remember, American Atomic clocks were cleared for export to India. I am not sure if this was L1 bidder or some other consideration. Significant American support was provided any way for the Navic constellation, so I am not sure why this component was not sourced from them as well.
5. Given the failures, statistically, one can expect a few more to fail and the constellation will not be stable for a while. But look at the bright side, our system is much better than the Chinese junk. Half of their satellites are bad at the moment..

[shiv]

Actually if you change the reference frame to the person traveling fast (i.e. he is stationary and earth is moving back fast from him) he will also find that the clocks on earth have slowed down.

Relativity is mutual.

Its an interesting puzzle that took me quite sometime to grasp and even more time to convince myself that there are no inconsistencies.

Well I guess we are travelling "relative" to the other guy. The first time I read about "proof" of this was long after I had read the theory in layman's terms. It was about subatomic particles created by particle collisions that had a very short lifespan but were moving very fast. Their lifespan (as observed by researchers) was longer than the predicted time and varied directly with velocity.

So as long as the reference point is "us" - time is slowing for the particle and "normal" for us (the researchers). I guess if the particle had stopped and looked at us before it died it would have been surprised to find that a timespan longer than its own life had elapsed on earth?

[sudeepj]

How does GPS on a cell phone can get your position even though it does not have an atomic clock??

Consider four satellites and a receiver. The local clock has an error term e. As the ranging measurements are made from the four satellites, they are time stamped with the local time at the receiver, with the error term e. All four measurements are made at different time instances, but they all have the common error term e. (We ignore the local oscillator drift at the moment).

This error term is common to all the ranging measurements made from the satellites. This common error term gets eliminated as we solve the non linear system of four equations.

[LokeshC]

Well I guess we are travelling "relative" to the other guy. The first time I read about "proof" of this was long after I had read the theory in layman's terms. It was about subatomic particles created by particle collisions that had a very short lifespan but were moving very fast. Their lifespan (as observed by researchers) was longer than the predicted time and varied directly with velocity.

So as long as the reference point is "us" - time is slowing for the particle and "normal" for us (the researchers). I guess if the particle had stopped and looked at us before it died it would have been surprised to find that a timespan longer than its own life had elapsed on earth?

You got it much faster than me. So the key is the guy who was traveling fast away from earth had to decelerate, slow down, stop and to turn back and come back to meet us. During this process, the clocks catch up and match exactly. He would have aged the same as us.

In case of the particle, it slows down, and before it stops, it will be dead. But from its point of view, it will look as if we are all standing still for the short time it existed.

Anyway. OT here. But interesting nevertheless.

[prasannasimha]

Moving Clocks alter their "measured time" wrt to an external observer directly as a consequence of special relativity. General relativity will slow done clocks nearer a larger mass This becomes manifest with either higher speeds or more precise measurements.
http://www.astronomy.ohio-state.edu/~po ... 5/gps.html

http://www.physics.org/article-questions.asp?id=77

[Amber G.]

Let me add to the following nice post..

1. As long as you have four signals originating from known locations, that have internally consistent clocks, you will be able to triangulate your position *and* derive the internally consistent clock.
2. This is simply the mirror image problem that GPS solves. In a normal GPS, you use four satellites with internally consistent clocks, to triangulate a position on the ground. In the mirror image problem, you will use four ground control centers with internally consistent clocks to triangulate a position in space. There will be some relativistic effects you will need to account for, but not a big deal.

Here are the problems:

...Its vastly cheaper to simply add a few more atomic clocks to the satellite for redundancy, than to maintain multiple ground control centers. ...

Yes, theoretically solving the "mirror image" problem is possible BUT a few commonsense points -

Unlike GPS in a car where you have to compute the position once in a while... A GPS sat has to broadcast the timing and position regularly while sats position is changing. You need very fast CPU (and error-free and efficient software) just to keep up. And statistically it may add to convenience but not over-all accuracy.. basis of accuracy is only 4 original ground stations.. sat does not add anything more in terms of accuracy.

(I once tried to demonstrate this point by having a Garmin GPS unit in a plane and it was not able to get the fix because plane was moving fast (much faster than a car) and cpu was not fast enough to calculate)

(As an example given by Vinaji - In Network protocol the S0 server standards are atomic clocks.. even S1 servers are accurate only up to a microsecond or so.. (your local computer may be as far away as a fraction of a second to a millisecond).. cost of faster cpu adds up (much cheaper to get another atomic clock than a slave).)

As I gave an example of sundial -- some may think it may be cheaper to built a sundial indoor (to prevent it from elements)..and in theory it is possible to build a sundial inside a planetarium and simulate a sun so that one can see the shadow on sundial and measure the time... but it sort of misses the whole point of using the sun as reference for accuracy.

I mean why go out to all that trouble to put an sat in stable orbit with its own atomic clock -- a slave can be used using other ground stations and theoretically any sat or a even hot air ballon with sophisticated electronics may be possible.

The point is: One uses the sats for simplicity and accuracy which ADDS to overall standards.. for this one try to use as stable (and predictable) orbit as possible, as stable clock (which does not depend on temperature, pressure etc) as possible. Cutting unnecessary corners here does not add anything of value.

[Amber G.]

How does GPS on a cell phone can get your position even though it does not have an atomic clock??

Consider four satellites and a receiver. The local clock has an error term e. As the ranging measurements are made from the four satellites, they are time stamped with the local time at the receiver, with the error term e. All four measurements are made at different time instances, but they all have the common error term e. (We ignore the local oscillator drift at the moment).

This error term is common to all the ranging measurements made from the satellites. This common error term gets eliminated as we solve the non linear system of four equations.

Thanks. This is why one needs at least 4 sats to get a fix. If your GPS had a calibrated atomic clock, you will need only
3 sats. (or only 2 sats, if you knew you are on earth's surface and knew the shape of the earth accurately )!

The reason I bring this to math class is for people to know and learn about geometry of hyperbola. If we take a plane and know the distance of a point from two fixed points, we can locate that point. If we know the difference of distances from two fixed point, the point can be anywhere on a hyperbola. (Thats the well known property of a hyperbola). Now if you have three fixed points, you can locate the intersection of these two hyperbola to get your position. In three dimension these three points are now changed to four.

Again the point is, by not having an accurate reference clock in GPS, you do not know/or calculate the distance between you and the satellite, but if you have two sats, you do know the time delay between these two signals, and hence the difference of distance between these two sats. Thus you know you are somewhere on a fixed hyperboloid. By finding intersection between three such hyperboloids you can determine your position,

Hope this may be of interest.

[Amber G.]

You got it much faster than me. So the key is the guy who was traveling fast away from earth had to decelerate, slow down, stop and to turn back and come back to meet us. During this process, the clocks catch up and match exactly. He would have aged the same as us.

.

I may be reading you wrong but if I understand you correctly, this is NOT the case. For example think of this experiment (which has been done).

Atomic Clock A and Clock B are exactly synchronized.
Clock A remained in New Delhi.
Clock B went on a plane to launching pad, went on an orbit, remained there for some time, came back to earth and went back to New Delhi.
Now question is will clock A and clock B will still show the same time?
Answer is NO. Clock B will be behind. (And this has been verified experimentally).

(And actually an astronaut who went to a star (turned back and came back) will be YOUNGER than the one who stayed on earth)

[disha]

One way to always remain young is to travel at speed of light., time stops for you.

Or go and stay near a black hole., gravity also slows down time, one will age slowly.
---

I think the focus should shift back to

1. How India's NAVIC works
2. What is the impact of the broken clock on India's NAVIC
3. How can we create our own atomic clocks and send it up space (should be a national project IMHO)

[Hari Seldon]

I just hope so many clocks breaking like this is not enemy action. Only. /Duckin' for cover.

[vina]

Excellent posts Amber G and SudeepJ . Very illuminating and it did answer the fundamental questions I had , which are.

1. Why does GPS need clocks to such accuracy ?
2. Is that an inherent "need" for such a clock or is that a result of "architecture" / "engineering design" of GPS and similar systems
3. Why should the GPS time need to sync with "ground based true time" ?

AmberG's question of "How does GPS on a cell phone can get your position even though it does not have an atomic clock" got me thinking and I did answer it . I had absolutely NO idea of how the GPS/other similar systems were implemented beyond knowing that they worked on Time Delay of Arrival and my "solution" didn't need ANY CLOCK at all in the receiver, and indeed, the satellites themselves need to be internally consistent, but don't need the ground based time at all !

SudeepJ made an excellent post on how GPS does what AmberG asked about (how it gets the time, and why GPS was built the way it is and indeed, why it needs such high accuracy clocks).

It turns out that there are multiple ways of solving the 'problem' and the difference is between how GPS and other systems are implemented. GPS uses something called Trilateration , while the other kind of systems use something called Multilateration . Turns out that in the "solution" , I had posted to AmberG's question (using phase difference of TDOA signals) , I had inadvertently stumbled upon Multilateration, which as per wiki is used by a whole host of applications including Loran-C and even cell phone tracking.

But notice, the important point with Multilateration kind of scheme and quoting.

Multilateration is a common technique in radio navigation systems, where it is known as hyperbolic navigation. These systems are relatively easy to construct as there is no need for a common clock, and the difference in the signal timing can be measured visibly using an oscilloscope

So, let me summarise what I understand from AmberG and Sudeepj's posts and my own digging around and thinking to answer the questions I had in reverse order .

3. Why should the GPS time need to sync with "ground based true time" ?
It doesn't . As Sudeepj pointed out, what is needed is for the GPS cluster to have an internally consistent time. That means as long as the clocks in the GPS cluster are synched, you can forget about synching with earth (where time is adjusted every now and then ) , as long as you are interested only in location . However if you want GPS also to provide "time service" (i.e. you are in the poles in winter where it is dark 6 months in a yearand lost and you as wtf is the time, GPS time will be a great help). In that case, yeah, you will synch with earth time. Otherwise, you can forget about it and all the relativity and other effects on the satellite clocks.

2. Is that an inherent "need" for such a clock or is that a result of "architecture" / "engineering design" of GPS and similar systems
No. The Multilateration (MLAT) schemes don't need a common time at all! So, it is probably architecture /engg design requirement based on the way the system is designed and built.

1. Why does GPS need clocks to such accuracy ?

The accuracy of the atomic clocks is much much higher than what can be achieved with a controlled XO. They shift on very short time scales, because of temperature, acceleration, and sometimes, just because they want to. You will never be able to get an XO frequency that stable. There may also be unknown physics involved here. Who knows what unknown behavior the XO will present when in orbit.

To me, this explained everything perfectly. You theoretically COULD have just frequency signal generators in each of the satellites and go with the MLAT scheme , but it turns out that the atomic clocks (more accurately signal generator, the clock is just a adding a start date highly stable periodic signal) is the best there is and gives the best results and accuracy with generating those signals.

So, now that the atomic signal generators perform better and are far more stable , you can actually get a stable "GPS" time and you can go with a Trilateration scheme like in the GPS ,and if the GPS time is synched with "earth time" , one more useful service in addition to location and speed, provided.

(As an example given by Vinaji - In Network protocol the S0 server standards are atomic clocks.. even S1 servers are accurate only up to a microsecond or so.. (your local computer may be as far away as a fraction of a second to a millisecond).. cost of faster cpu adds up (much cheaper to get another atomic clock than a slave).)

AmberG, a nit , and a bit "OT" ,but I do think relevant here in light of the discussions we had.

I am not sure if the 'local computer' is as inaccurate as you calculate it to be. For instance, Timestamp Based Ordering is a very well known (efficient) and widely used method for database concurrency control. The time queried from a machine is upto microseconds and the time stored in the database is precise to microseconds as well.

Now in distributed database and systems (think of a database split between Tokyo,Sydney, Bangalore, London, NY and SF), if you need to ensure concurrency (i.e.,consistent reads and writes), with a timestamp based ordering protocol, you do need the timestamp on each of the distributed database clusters (in the far flung locations) to synch to microseconds and that is done with the NTP protocols. And there are huge latencies everywhere and distributed databases have been around for a long time and for IBM systems the communication was via VTAM access methods (and even older methods) which hark back to days of dialup over copper!

[disha]

This doesn't belong here, but just look at the Proton M launched from Baikanor Cosmodrome, launching a sat to Geostationary orbit. See the orbit and the changes they have to do from that latitude !

Just to note., Gagan'ji - the above orbit Proton-M does is to put other nations' sats into GTO/GEO. Russia itself uses Molniya orbits.

[Gagan]

Yes they do, damn the polar region

[Singha]

My nokia 720 with offline maps worked fine inside a plane.

[Gagan]

How good is the map resolution? How big is the map area?

[Rishi Verma]

The accuracy required (in both stable frequency and latency (wrt to "standard" TAI) is about MILLION times more than what is typical needed (or achieved) on a typical computer or a cell phone.

Yes. What I am trying to understand is WHY is such accuracy required .

Saar, not the accuracy but stability is of prime importance. If Sat A has 1.0000000Mhz clock and Sat B has 10.00000000 Mhz clock, the ground receiver can handle it... But if a sat has drift of say 1ps per day (+ or minus in random direction).. That's where the prob lies.

We know that PLL are a good filter of such noise (or drift), so my original question was why satellite clock cant be phase locked to an atomic clock located on earth.

PS added later

I think i have the answer, the atomic clocks (purchased from UK) are phase locked to each other on the ground, and are always hot standby (powered on). And when launched in space separately.. Months or years apart, the phase relationship is maintained. And signal from different satellites will have phase delay based on ground position of the receiver and simple math can then tell where the ground object is.. Then the velocity calculation is easy

[vina]

Saar, not the accuracy but stability is of prime importance. If Sat A has 1.0000000Mhz clock and Sat B has 10.00000000 Mhz clock, the ground receiver can handle it... But if a sat has drift of say 1ps per day (+ or minus in random direction).. That's where the prob lies.

We know that PLL are a good filter of such noise (or drift), so my original question was why satellite clock cant be phase locked to an atomic clock located on earth.

I am NOT an electrical engineer, I barely managed to scrape through the mandatory courses in that and even today, the word "H Parameters" brings a cold sweat in my back and makes me sit up in the middle of the night! I got through the courses with an "Honorable" "D", I must have been borderline "E".
(Back in my Madrassa, "E" was worse than failing the course ,repeating the courseand scoring top grade . Your CGPA ended up being lower if you got an E, even though if you flunk a course, the failing grade stayed in your records. In unmentionable places (I think Dilli, and Kanpur), you flunk a course and repeat it, the records are scrubbed clean and get replaced with the good grade.. Their inherent Pakiness I assure you, but I digress)

However, that said, I do understand the signal processing math quite well . I use it in a different context in everyday life and do understand your question , especially the random drift which as you describe is Brownian motion (and which characteristically is not mean reverting and is ultimately diverging) and why the stability of the signal is so important for the entire thing to work.

PS added later

I think i have the answer, the atomic clocks (purchased from UK) are phase locked to each other on the ground, and are always hot standby (powered on). And when launched in space separately.. Months or years apart, the phase relationship is maintained. And signal from different satellites will have phase delay based on ground position of the receiver and simple math can then tell where the ground object is.. Then the velocity calculation is easy

. YOU are the kind of guy who made ordinary mortals in the Madrassa feel inadequate and indeed "diminished" , like the guy who lazes on the bed and looks at the tutorial sheet with the circuit diagrams with which 4 classmates sitting on the floor are struggling and trying to do it ,and guy on the bed yawns and just tells the answer, looking at it upside down ! (this is from experienced reality)

But what you are saying is fantastic , even to imagine !

Just how will you maintain the phase lock between a satellite in the sky and on the ground (which was what your original question was), and in the absence of a common driving signal, why will the different clocks remain synchronised and how will you propagate the signal with all the latency delay to a satellite in orbit and the same signal with incredible fidelity to a clock a few wire lengths away ? How can you maintain phase, while taking it from say the building where the clock is , to the satellite, where it is being integrated, to the launch pad? The sheer mechanics of that seems unbelievable .

[SaiK]

Amber ji, if the two clocks are assumed to be "exactly synchronized", then they should be verified as "showing the same time" unless the synchronization failed when it went outta reach to synchronize [assuming signal loss]. no?

[shiv]

Is it necessary at all for every satellite of the GPS constellation to be exactly synchronized? I am not sure if that is possible, or if possible, whether it is trivial. It may be just unnecessary. Doesn't "internal consistency" mean that the clock in each individual satellite should be consistent and not that it should be exactly synchronised with other satellite clocks.

As I understand it as long as the distance of a point on the earth's surface from each of 3 satellites can be measured accurately the location of that point can be fixed. That does not require all satellites to have exactly synchronized clocks. It only needs internally consistent clocks on each of the satelllites.

But if the clocks on each satellite keep changing over time - then the coordinates of my house on earth as measured on June 15, 2016 will be different from one measured on June 15, 2017. So a periodic correction needs to be applied to satellite clocks so that their changed times do not lead to coordinates on earth creeping and changing position. Is this correct?

[negi]

Distributed databases is one thing but to support transactions across distributed a different problem it was not solved until recently . Google has done exactly what AmberG has explained they are using atomic clocks and GPS to support transaction properties on their cloud spanner offering . However even after ensuring perfect sync in terms of timing I still wonder how do write locks get synced but that's a topic for different thread

[negi]

On a related topic do all satellites need an atmoic clock or it is only needed based on application and requirements of an orbit for instance a satellite in GTO will need it to maintain position but what about those in LEO which are placed for applications like imaging ?

[vina]

Distributed databases is one thing but to support transactions across distributed a different problem it was not solved until recently . Google has done exactly what AmberG has explained they are using atomic clocks and GPS to support transaction properties on their cloud spanner offering .

Pah! I have personally been hands on with such systems back in the late 80s to mid 90s ! I have always maintained that anything that needs to be "invented" in Computer Science has already been done by IBM circa 1979 and the rest is all just packaging and rehashing of the same by "wannabes". Old man RBose though wants to move the entire universe to HTML5 (which is a "meh"), and Anujan will contest that statement of mine, but it is largely true.

Why, all kinds of distributed architectures were available. Centralised TP Monitor (CICS/Tuxedo/Encina) with Distributed DB, Distributed TP Monitor and Centralised DB, both TP Monitor and DB distributed. You name it. Clustered ? Check. What is called "Grid/Cloud" , check. All this on 31 bit address machine (1 bit was for backward compatibility) .

As for "GPS" and Atomic Clock, it is that GPS time IS an S0 Atomic Clock! Get GPS receiver and query it for time and sync with it, you have an S1 Server. Since GPS is available worldwide, it is much easier to do so. Earlier, they probably got their timing from the closest atomic clock in some Govt/Research lab.

Anyways. Deeply OT. Take it to Nukkad and ask RBose.

However even after ensuring perfect sync in terms of timing I still wonder how do write locks get synced but that's a topic for different thread

The transaction "ACID" properties are managed by a 2 phase commit protocol by the Transaction managers (the individual "resources") are also on the same commit protocol. Long story short, the locks don't need to be synced . The TP managers wait and hear from individual resources on status and manage this. Cacha probably uses a customised version of a TP manager spread over the cloud. Haven't looked into it , just my guess.
Anyways. Deeply OT. Take it to Nukkad and ask RBose.

[LokeshC]

I may be reading you wrong but if I understand you correctly, this is NOT the case. For example think of this experiment (which has been done).

Atomic Clock A and Clock B are exactly synchronized.
Clock A remained in New Delhi.
Clock B went on a plane to launching pad, went on an orbit, remained there for some time, came back to earth and went back to New Delhi.
Now question is will clock A and clock B will still show the same time?
Answer is NO. Clock B will be behind. (And this has been verified experimentally).


(And actually an astronaut who went to a star (turned back and came back) will be YOUNGER than the one who stayed on earth)

Sorry for the OT (will take it to another thread if we continue).

Amber G,
You are reading me right and you are right that I am wrong (I spent sometime trying to figure out where I went wrong).

I assumed (due to the illusion of symmetry) if we "fix" the reference frame as the one of the traveling Clock B, then we would see Clock A travel the opposite direction with the same velocity. So clock A would be behind, by the same amount as Clock B. Then we have a Age(A)>Age(B) and Age(B)>Age(A) problem.

So here is where I went wrong:
While I grasped the fact that relativity means the order of occurrence and spacing (in time) of events occurring will change depending on reference frame. I completely missed the length contraction part

My assumption was when B moves away from A it will appear redshifted (i.e. clocks appear to run slow) and then when it moves towards A it will appear blueshifted (i.e. clocks appear running fast). Ignorning relativistic doppler effects (and length contraction). I did some elementary high school math and assumed that the blueshift and redshift would cancel each other out and we have consistency.

With a little more diligence, it now appears to me that the travelling Clock B will see a different distance of travel than Clock A, which means the time it travels that distance will also be different. And hence Clock B would indeed be behind Clock A.

The asymmetry is quite fundamental and it is something that should have been very obvious to me if I had grasped what length contraction *really* does. I feel so dumb now

[Amber G.]

Good questions.. to be clear one needs clear answer.. so I will take a stab at it.

Is it necessary at all for every satellite of the GPS constellation to be exactly synchronized? I am not sure if that is possible, or if possible, whether it is trivial. It may be just unnecessary. Doesn't "internal consistency" mean that the clock in each individual satellite should be consistent and not that it should be exactly synchronised with other satellite clocks.

The atomic clocks are and needed to be synchronized as exactly as possible... and technically it is a VERY big deal. Seeing a clock from a foot away and you are already 1000 ps behind! If you connect it with a cable, even an extra millimeter of wire will delay it and has to be taken into account. In practice the standard time is TAI.. check out wiki (or good reference) for details on how one synchronizes different clocks over distances etc.
(google for "International Atomic Time" or "reference second/clocks"). *all* individual sats need to be synchronized with each other (or one must know the time diff from TAI).

In addition accurate elements of ephemeris (so accurate position of the satellite can be calculated at any given time (when the time-signal pulse is sent).
In other word, receiving GPS unit knows (or can calculate) the exact position (say longitude, latitude, and distance from the center of earth) and time of the origin of signal, at the time the pulse was sent.
If GPS has this data from 4 sats, it can solve and find the location of the receiver.

***
Side remark - related to India's sats..

Though we use the term "atomic clocks" in practice any precise clock can be used.. there are some which are little cheaper (say Rb clocks) .. there are masers... or optical clocks which are even more accurate/stable than Cs clocks.
So this technology can change in future... something our scientists and bright engineers must keep an eye on.

[prasannasimha]

Deleted

[prasannasimha]

We have hydrogen maser clocks at various ISRO centres like at Byalalu. The thing is they are very expensive. In fact we sell the data to various countries like US.

[Amber G.]

Thanks for some interesting posts. Glad that people found it useful. Some quick comments as I think they are important for basic understanding. Sorry for brevity as I may comment on a few critical parts only, sorry if it seems nit-picking. (I hope it brings some clarity and adds to discussion).

Excellent posts Amber G and SudeepJ . ...
...
So, let me summarise what I understand from AmberG and Sudeepj's posts and my own digging around and thinking to answer the questions I had in reverse order .
..
3. Why should the GPS time need to sync with "ground based true time" ?
It doesn't . As Sudeepj pointed out, what is needed is for the GPS cluster to have an internally consistent time. That means as long as the clocks in the GPS cluster are synched, you can forget about synching with earth (where time is adjusted every now and then ) , as long as you are interested only in location . However if you want GPS also to provide "time service" (i.e. you are in the poles in winter where it is dark 6 months in a yearand lost and you as wtf is the time, GPS time will be a great help). In that case, yeah, you will synch with earth time. Otherwise, you can forget about it and all the relativity and other effects on the satellite clocks.

This is not correct, that is you need GPS cluster to have to sync with TAI (earth's standard time). To locate your position, one needs exact location of the satellite at the time the signal (pulse) was sent. The data from GPS sat contains, among other things, ephemeris of the orbit so that one can calculate the position of sat (wrt to a fixed reference -- say center of the earth ) at the moment when the pulse was sent.

(I other words, the orbit elements are known, so one can calculate (X,Y,Z) coordinates at time T
(Another way to deduct that you need 4 sats)!
(** if your coordinates are (x,y,z ) and your (car's gps clock is offset by t, while the pulse originate at X,Y,Z at time T,
the time delay (T-t) = sqrt(X-x)^2+(Y-y)^2+(Z-z)^2)/c .. 4 unknowns, you need 4 equations! **)

***

(As an example given by Vinaji - In Network protocol the S0 server standards are atomic clocks.. even S1 servers are accurate only up to a microsecond or so.. (your local computer may be as far away as a fraction of a second to a millisecond).. cost of faster cpu adds up (much cheaper to get another atomic clock than a slave).)

AmberG, a nit , and a bit "OT" ,but I do think relevant here in light of the discussions we had.

I am not sure if the 'local computer' is as inaccurate as you calculate it to be. For instance, Timestamp Based Ordering is a very well known (efficient) and widely used method for database concurrency control. The time queried from a machine is upto microseconds and the time stored in the database is precise to microseconds as well.

I know what you mean but I think I am correct...(speaking about ordinary networks)

Again for concurrency or time-stamp time problems you need that all clocks in the cluster are in phase with each other.. you do not need it synched with TAI .. (unless you are doing astronomical or GPS type work).

Many unix servers (in US schools) , (and time.windows.com server for PC) have granularity of microsecond in time-stamps but can be off by even a second or so from TAI. ..Now a days typical accuracy IMO is about 1 ms .. (of course if one needs better accuracy up to 1 microsecond it is easy to do it technically).

More - this may give more information .. Windows Server 2016 Accurate Time..
Lot of good details, I may just quote a few lines.. (For complete picture see the article)

Government Regulations like:
50 ms accuracy for FINRA in the US
1 ms ESMA (MiFID II) in the EU. (These are for apps like Distributed systems like Cluster/SQL/Exchange and Document DBs)..
Additionally, the source we use for accurate time references an improved API which gives us better resolution. With these improvements we are able to achieve 1 ms accuracy with regards to UTC across a domain.

Of course, Rb clocks are not that expensive these days (~$1K) and it can attach to your server with truly great accuracy if you need one... And as you already mentioned in another post GPS unit is easily available which can give similar accuracy.

[vina]

We have hydrogen maser clocks at various ISRO centres like at Byalalu. The thing is they are very expensive. In fact we sell the data to various countries like US.

Don't know about ISRO, but the place in India which always had such clocks was NPL in Delhi. In fact , today's Hindu had an article exactly on the topic we have been talking about.

Time Is Money- ISRO to pay Official Clock Keeper

The National Physical Laboratory (NPL), part of the Council of Scientific & Industrial Research, is host of the most accurate clocks in the country, and the only agency in India authorised to maintain Indian Standard Time (IST). The NPL maintains accuracy of ±20 nanoseconds through the Primary Time Scale, an ensemble of five caesium clocks and one hydrogen maser. The rest of the world connects to these via tele-clocks, satellite links, and Network Time Protocol services (which, for example, reflect in laptops’ clocks).

[Amber G.]

My nokia 720 with offline maps worked fine inside a plane.

Newer Garmin with faster CPU, iPhone 7 works too..older ones not that well.

[Amber G.]

Distributed databases is one thing but to support transactions across distributed a different problem it was not solved until recently . Google has done exactly what AmberG has explained they are using atomic clocks and GPS to support transaction properties on their cloud spanner offering . However even after ensuring perfect sync in terms of timing I still wonder how do write locks get synced but that's a topic for different thread

I heard Google has even used Sr - optical clocks (which are much more accurate than Cs clocks)..I think in few years we may even have a new standard for time - from Cs to optical clocks.

[Amber G.]

If the times are off b/w receiver and sender, won't the distance be miscalculated? If time diff b/w satellite and cell phone is say 2 secs, that means the distance that is calculated will be wrong, and so is subsequent triangulation point? And so the triangulation spot will move off the actual location??

I think many have answered this. Basically we do not calculate time diff between receiver and sender -- but rather difference between (<time diff b/w sat A and receiver and tie diff b/w sat B and receiver> so inaccuracy of receiver clock is canceled out).. Check out Sundeepj's excellent post for more details.

SundeepJ - excellent posts.

[JayS]

Is it necessary at all for every satellite of the GPS constellation to be exactly synchronized? I am not sure if that is possible, or if possible, whether it is trivial. It may be just unnecessary. Doesn't "internal consistency" mean that the clock in each individual satellite should be consistent and not that it should be exactly synchronised with other satellite clocks.

As I understand it as long as the distance of a point on the earth's surface from each of 3 satellites can be measured accurately the location of that point can be fixed. That does not require all satellites to have exactly synchronized clocks. It only needs internally consistent clocks on each of the satelllites.

But if the clocks on each satellite keep changing over time - then the coordinates of my house on earth as measured on June 15, 2016 will be different from one measured on June 15, 2017. So a periodic correction needs to be applied to satellite clocks so that their changed times do not lead to coordinates on earth creeping and changing position. Is this correct?

Since the ranging between satellite and GPS receive is based on one way communication, its imperative that the satellite clocks are synch-ed. Lets say Sat A sends time stamp saying its 1:00:00 PM and a GPS Receiver R who starts listening receives it at 1:00:01 PM. Now it knows its 1light sec away from Sat A. Receiver R also receives a time stamp from Sat B saying 1:00:00 PM at 1:00:01 PM. Then it will calculate that R is 1 light sec away from Sat B as well. But if the Sat A and Sat B clocks are not sync-ed, how will R be sure that the time stamp of 1:00:00 PM were generated by A and B at the exact same instance..? Say there was time difference in A and B of 1sec, and B actually sent time stamp 1 sec earlier than A. But how will R know this..? Had It been a two way communication this could be dealt with. But one way communication means R needs to assume when Sat A and Sat B both say 1:00:00 PM, they are both sending that signal exactly at the same instance. So you have to sync the satellite clocks as well.
Google search tells me they do it every 24hrs using the ground station network.

[negi]

How does GPS on a cell phone can get your position even though it does not have an atomic clock??

Consider four satellites and a receiver. The local clock has an error term e. As the ranging measurements are made from the four satellites, they are time stamped with the local time at the receiver, with the error term e. All four measurements are made at different time instances, but they all have the common error term e. (We ignore the local oscillator drift at the moment).

This error term is common to all the ranging measurements made from the satellites. This common error term gets eliminated as we solve the non linear system of four equations.

Excellent so can you elaborate when someone gives gyan about military grade versus civilian grade GPS signal is it correct to assume that the only difference is between the 'e' of the local clock in question i.e. in case of civilian use case our mobile phone might be getting a feed via NTP which might be only accurate to milliseconds while the Military chaps have receivers that refer to a local time source which has a far greater resolution ? How the GPS satellite interacts or to be be more specific serves a mobile phone versus say a GPS guided munition is exactly the same , it is the low value of e in case of a military GPS receiver kit that allows it to get more accurate coordinates .