Peter wrote: > On Wed, 11 Jan 2006, Robert Rolf wrote: >>>Maybe that is locked to a rubidium standard somewhere. >> >>Nope. Except perhaps for SCPC (single channel per carrier). > > > That has nothing to do with it. Think about it. A studio somewhere sends > TV (say analog). This goes up, then down, is converted, demodulated etc. > The frequency stability may not be so famous. But the *count* of say > frames or lines sent, per day, depends on the initial standard, not on > the link. And how many frames get dropped as sources get changed, machines burp, etc. I used to work in a TV station. We tried very hard to use station sync, for everything but there were always hiccups over a day. Today the is much less likely since the frame synchronize would fill in the holes. > F.ex. for PAL a studio will send 1.35x10^9 line pulses per day. Nobody > cares as long as they all get to the other side on the same day. If one > pulse falls 'tomorrow' instead of today, who cares. By counting the > pulses over 24 hours one should achieve 1msec/month. Maybe. I have the advantage of being able to look at several TV stations at once (both off air, satellite and cable). Frequency accuracy is all over the map. If one beats one station against another (two inputs to an old analog switcher set to 10% mix between them), one can easily see the 'spin' of one stations frame rate compared to another. Some stations were offset by as much as 10 lines /second (the fastest I could count). 10x64uSx86400s/day =5.4 seconds a day DIFFERENCE! I quickly realized that you can't use a TV station as a time reference. > Same for V rate. 4320000 V pulses are sent in 24 hours (50Hz) with the > *studio*'s timebase accuracy. A normal mains operated clock can be > operated almost directly off this signal (just tap to the V yoke's stray > radiation with a coil). I do not care how they get to the V yoke as long > as they come from the studio. Yes, there is enough stray field that it is easily picked up. > Imho this is too good for not to be tried with an old TV and a LED clock > mated to it temporarily and observed over a week or so. It's probably easier and less power hungry to use an old VCR. Easy enough to make an R/C sync separator to pull out the vertical or horizontal sync. (LM1881 if you want a chip to do it) > studio and it should have the accuracy of their output TBC, which is > locked to the studio timebase. Maybe this last TBC is not so accurate > but it should be within 2ppm or less even if it is not locked to > anything afaik. 2ppm is quite could compared to what I've seen for REAL. PAL has much more stringent timing specs that NTSC. http://www.microsoft.com/whdc/archive/VidSynch.mspx "Most decoder crystals in mass production have a rated accuracy of more or less 100 parts per million (ppm), which on a 27-MHz crystal equals more or less 2700 parts per second." > But you are integrating pulses over 24 hours (or more)!. Even if the > phase noise is lousy (say 50dB) the integration effect will improve this > by 66dB or more for V pulses and even more for H. > > More to the point: assuming you have a PIC using an ordinary Xtal and > also receiving 15625Hz picked off a TV that somehow receives a studio's > signal. It will see 1.35x10^9 H pulses per day. Assuming it can detect > that it has received the pulses without interruption (that many pulses), > at the end of the day (or whenever convenient), the count from the > internal oscillator is compared. Say it runs at 4MHz, then there will be > 8.64x10^10 pulses in the clock counter when the H counter reaches > 1.35x10^9 (=24 hours). If so, then the quartz clock is exact. Suppose > the quartz clock counter shows 8.64x10^10 - 10 pulses when the H counter > reaches 1.35x10^9. Then the quartz is slow by ~1.1x10^-10. This > information can be used to operate a pulse 'stealing' constant that > effectively disciplines the quartz. The constant will be written to > EEPROM and used all the time until overwritten (usually this is done > with two constants, to achieve fractional approach but for such a huge > accumulator a signed integer should work as well). This ASSUMES that the television signal is broadcast with NO dropped frames and it that has been referenced to a good source. That is not always the case. So you will then argue that one could look for dropped frames and adjust the count accordingly. Sure. In the 'old' days, networks did use Rubidium standards. Today, with cheap frame synchronizers, most stations are closer to 50ppm than 2ppm accuracy. At least thats what my comparison of off air sources in Canada shows. > Further, assuming the user switches channels about 20 times per day and > every time he does that, the TV jumps about 100 H pulses until resyncing > then the uncertainty will be +0/-2000 pulses per day. So it still works > out to better than 2ppm (as long as he does not watch DVD or video or > Nintendo or Playstation). Right. Which is why a VCR or Cable convertor with baseband out is a better starting point. > Wiring a LED clock into an old TV for timing should not be so hard. > Maybe someone can try this and see what happens after a week, by > comparing to NTP (Internet) or similar ? The LED clock should show > seconds for best results. It's certainly doable. And if you used the power line AND the TV signal as interrupts to a PC, (RI and RX on a serial port) you could see the jumps in phase very readily. And just for interest's sake NTSC vertical rate is 59.940052333 14318180/455/525 Robert -- http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist