[time-nuts] yet another GPSDO design, or so

Sun Jun 27 23:18:41 UTC 2010

Stanley
Let me try one more time after that I give up maybe some one else can try.  
If the time between the GPS signal and the reference drifts by 100 nsec. 
the  counter in the Shera 24 MHz board will detect it and based on that make 
the  decision to change the D/A to compensate for it at a rate of 1.7 E-13. 
It is  part of the program. If in stead the count frequency is 100 MHZ the 
PIC will  think that the change is four times as large and will instruct the 
D/A to  correct accordingly. How ever I have attenuated the output of the D/A 
by four  making the correct change (4.3 E-14 steps) for the 100 nsec. 
change.  That is all, nothing else. 
Bert

In a message dated 6/27/2010 6:23:30 P.M. Eastern Daylight Time,  
stanley_reynolds at yahoo.com writes:

Jitter  may not be the correct word. I also don't know how accurate or 
repeatable the  averaging effect is as described in the article.  But I do 
believe the  amount of drift is important for this to work as stated before. 
Independent  is required, if locked then this is the extreme case of no drift 
and  average value will be the same as all the samples are the same  integer 
number. To see the fraction between the integer we need our data set  to have 
some number of samples above and below and hopefully in a ratio to  
indicate the real measurement. If the drift is too small then we will tend to  
calculate an average too small, as more samples will fall to the low side the  
high side is never reached. If the drift is too big we will over shoot to the 
 next higher integer number and this will bias the average up too  high.

Stanley

----- Original Message ----
From:  "EWKehren at aol.com" <EWKehren at aol.com>
To: time-nuts at febo.com
Sent:  Sun, June 27, 2010 4:27:39 PM
Subject: Re: [time-nuts] yet another GPSDO  design, or so

Stanley
I am not an expert but it is not only the  jitter it is the fact that since 

the two sources are not linked the  independent drift of the 100 MHz causes 
a  distribution of the count.  Again even with the same resolution the 
count 
error  will be four X  and the subsequent correction will be to scale, with 
out changing   the code.  Bert

In a message dated 6/27/2010 4:00:59 P.M.  Eastern Daylight Time,  
stanley_reynolds at yahoo.com  writes:

Without  knowing your 100Mhz clock would it not need less  jitter to 
average 
out +- one  count at 100Mhz vs one count at  24Mhz?

The GPS error has improved with  better hardware/software  in the receivers 
as well as turning off SA. So I'm  not so sure the  GPS error will wash out 
the counter's error. Even if it does,  as a  want to be time nut I need a 
better counter which is why more is  better,  a faster count with a 
recipical 
counter equals better unless  the faster  counter no longer averages out 
the +-   
count.

Stanley

----- Original Message  ----
From:  "EWKehren at aol.com" <EWKehren at aol.com>
To:  time-nuts at febo.com
Sent:  Sun, June 27, 2010 1:17:51 PM
Subject:  Re: [time-nuts] yet another GPSDO  design, or so

Going from 24 to  100 MHz only gives you smaller steps  (resolution) every  
thing  else stays the same.
If the he saw 2 to  3 nsec should be more like 8,  going to 100 MHz will  
improve it by a  factor of 4. In a  redesign of the total system I would 
have two 
sample  sizes maybe  stay with 30 or go to 50/60 and in the Rub. mode 200  
maybe 300 sec.  Let us not forget what we start out with the GPS signal  
does  
not  allow us to take advantage of the full resolution.
Do   not forget I did this to get smaller D/A steps and am not able to   
rewrite the code, basically fooling the controller that the error  should  
call  
for a 1.7 E-13 correction when in reality the  error is 4.3 E-14  and the  
resulting step is also 4.3  E-14  per D/A  bit.
Bert

In a message dated  6/27/2010 1:43:40 P.M. Eastern  Daylight Time,   
stanley_reynolds at yahoo.com writes:

Yes I  see your need for a  reduced range with smaller steps. But I was 
looking  for   smaller steps to improve the tracking accuracy without a 
loss 
of   
the benefit  of averaging. From the QST   article:

"Interestingly, it is  desirable to have the frequency  of  U7 drift 
slightly 
rather than being  synchronized with  the VCXO.  A
slight random drift averages out the count   ambiguity that is  inherent in 
any pulse-counting device. My   measurements
indicate  that the simple phase-measuring circuit I use  is  consistently  
accurate to 2 or 3 ns (for a 30-second  measurement),  while
without drift, the resolution would be limited to  42 ns. The $5  crystal 
oscillator module drifts adequately"

So  the drift should  just  cover the area of uncertainty that is one  
cycle, 
too much drift  would reduce  accuracy, not enough and  the average is of 
no 

benefit.
One extreme no  jitter, average  doesn't work as it  doesn't distribute the 
samples over the   range of uncertainty. The  other case too much jitter 
and 
the best to  expect is  an average  weighted to one side or the other (+-1  
count) with the extreme  producing multiple counts of error.  Something 
about this 
makes me  nervous  maybe the part about  "slight random drift" what is 
slight 
at  24 Mhz is it also   slight at 100 Mhz ? An average of 30 samples does  
have a limit to  what it will  correct. 

Stanley  

----- Original  Message ----
From:  "EWKehren at aol.com"   <EWKehren at aol.com>
To: time-nuts at febo.com
Sent:  Sun,  June  27, 2010 8:58:55 AM
Subject: Re: [time-nuts] yet another  GPSDO  design, or so

Stanley
the faster counter also has the  jitter,  no  change, as long as it is  not 
tied to the  input frequency.  The 24 MHz  is not unique, the 100 MHz is 
same  

technology just  four times faster  and thus gives me  smaller steps on the 
D/A  
and  since I use it on  Rub.  the full range of the 18 bit covers  the full 
tuning range   of  the Rub.
Bert

In a  message dated 6/27/2010  9:05:12 A.M. Eastern  Daylight Time,  
stanley_reynolds at yahoo.com  writes:

I  have  been thinking  about a faster counter  also but the Shera board 
was 

depending  on  the jitter in  the 24 Mhz clock to average out the +-  
count. 

The  
faster  clock would reduce the need for this but  without   the right 
amount 
of jitter we  lose the benefit of  this   average.

Stanley

<snip>

_______________________________________________
time-nuts   mailing list -- time-nuts at febo.com
To unsubscribe, go to   
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and  follow  the  instructions   there.
_______________________________________________
time-nuts  mailing  list -- time-nuts at febo.com
To unsubscribe, go to   
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow  the  instructions   there.

_______________________________________________
time-nuts   mailing list -- time-nuts at febo.com
To unsubscribe, go to   
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow  the  instructions  there.
_______________________________________________
time-nuts mailing  list -- time-nuts at febo.com
To unsubscribe, go to  
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the  instructions  there.

_______________________________________________
time-nuts  mailing list -- time-nuts at febo.com
To unsubscribe, go to  
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the  instructions there.