[time-nuts] Home made GPS disciplined atomic clock

Esa Heikkinen scifiscifi at sci.fi
Wed Jan 28 13:06:31 UTC 2009


> Esa,  the answer to your problem is, as I said before, to clean up the  
> signal, use a high quality oscillator locked to your system with a PLL that  has 
> the appropriate time constant.

Yes. My problem is to find time constants, with measurements and that's 
currently under work with GPS, later with LPRO.

I've done some HW and SW planning also. It could be like this:

- Tbolt is used to synchronize LPRO with own designed steering
   system.

- LPRO steering time constant will be long, let's say 24h or
   something, to cancel out any day/night variations (if any)
   on GPS reception.

- The goal for LPRO synchronization is to have constant frequency,
   not exact time. This will ease the LPRO steering algorithm and
   there's no need to catch up the exact time by changing the
   output frequency (like tbolt always does).

- Own steering electronics will constantly monitor the state
   of tbolt with serial port. If holdover is detected then
   the LRPO steering will also be stopped and averaging loop
   will be reset. DAC remains as it was.

- When the holdover is over the running averaging will be
   started again but C-field DAC will remain as it was until
   there's enough data for last 24h to do some C-field
   corrections.

- It might be a good idea to reset the steering also if the
   frequency error (ppt value) of Tbolt's 10 MHz output
   is detected to be too high.

- Steering MCU uses LPRO as it's clock because it will count
   it. MCU's HW peripherals like counters and capture unit
   is used to handle the 10 MHz count, the software just
   reads out the counter values occassionally.

- The counters are read only every n'th:s PPS so that
   mHz resolution can be achieved without using external
   frequency multipliers etc. Also the PPS drift will
   be averaged this way. The n could be at least 3600.

- These counted values are used as input data to
   running avg having a long time constant such as 24h
   or even more.

- So the period for DAC changes will depend how long
   counts (in seconds) is done at first stage. It's sure
   that the DAC won't be changed here in about every
   second like Tbolt does!

- Then LPRO's output is cleaned with Wenzel or some
   other OCXO, with suitable time constant which has to
   be find out. Phase detector and good reference OCXO
   is needed for that. May be it's wise to buy only one
   OCXO and use it for measurements frist and then as
   output oscillator for final setup.

- Finally there will be some kind of distribution amplifier
   for 10 MHz.

So it will be slow to settle but should be good enough for stable 10 MHz 
source. I think it will have good long term stability due GPS, good 
holdover performance and if the final OCXO cleanup works as excepted 
there will it should also have good short term stability.

Much work to do but for now this is only a hobby project.

> The time constant will smooth out the jumps  you 
> see right now. The question is only which oscillator. A Wenzel was suggested  
> but in my opinion it is not what you need.

The output phase noise may be not so bad issue as it sounds like. It 
seems that some RF instruments (like my spectrum analyzer) have their 
own clean up loops for external ref. Only problem is that their time 
constants are unknown. But it's in there - at least in spectrum 
analyzer; checked that from service manual yesterday.

> I have used a Wenzel because I wanted  the low phase noise when multiplied
 > to 10 Ghz.

In these days my maximum needed frequency in about 5.2 GHz so any error 
on 10 MHz ref will be multiplied by 520. For that reason I also want 
CONSTANT frequency, it's not so bad if it's off ±1 mHz at some day but 
it's very bad if it has different frequency error between measurements 
done on same day. So I'd like to drive that DAC as rarely as possible 
and as little steps as possible. May be even only one correction per 
day, done at nighttime when there are no measurements on going.. There 
will be LCD screen for status so it could also told the actual frequency 
  difference to averaged GPS reference but without any DAC changes if
they are not desired that time.

> plenty of  oscillators out there at a reasonable price starting with
 > the HP 10544. Yes, the  10544 will clean up your present setup and
 > will be at least > a 10X improvement  over the LPRO solution. But
 > the most widely available unit is the HP 10811 which  will do
 > a great job.

Haven't seen those available here in Finland. We have much too fanatic 
"recycling" on going on Finland and so even fully working electronics 
are doomed to be destroyed (shredded to get the metals out of them). One 
friend has seen his own eyes the destroying of only couple of years old 
servers etc. Madness! And any plastic material remained for this process 
is dumped as huge lumps for next generations. Saves nature indeed! All 
thanks to this goes to one company which just want to "recycle" metal 
and export it to china.

So there's no such "treasures" here which you guys have in US (according 
the stories on your websites). Also it's quite expensive to order things 
outside EU unless there are some friend who could handle the purchase 
locally and then send the items as a "gift". Some surplus stuff is of 
course available here but the sellers are asking quite high prices. For 
example, old used LPRO's will cost about 200 euros each, Tbolt will cost 
about same amount, and so on. 1 eur =  1.32 US dollar.

> Depending on how much you want to spend
> the FTS 1200, 1000  or 2000 are an option.

I think that the money is not so big problem that as the poor 
availibilty of this kind of stuff here in Finland.

-- 
73s!
Esa
OH4KJU




More information about the time-nuts mailing list