[GPS_Standard] vcxo control averaging?

Sun Feb 18 08:23:53 EST 2007

Hi again Arie,

The question you are bringing up is the following:
What is better, one 5000 second sample, or 313 16-second samples used to
calculate an average offset? You say that the former is better. I am not a
statistics expert so other members' opinion are welcome here. Of course, the
technique used to calculate the average of the frequency samples will have
an impact on the final verdict...

The way the PIC works would not allow to easily do the single 5000 second
sample. We deal with a 16 bit counter that already rolls over 2441 times
during a 16-second interval. With this PIC micro, there is a built-in
divide-by-16 prescaler that counts 16 1PPS clocks before latching the 10MHz
counter value. Simple and effective. Since the final 16-bit counter value
falls somewhere in the middle (26624), I don't even keep track of the number
of counter rollovers. Keeping track of the absolute count for thousands of
seconds would be much more complex. Not impossible, of course, but a totally
different approach...And a lot of re-work!

Cheers,

Bert, VE2ZAZ

----- Original Message -----
From: "Arie Dogterom" <pa0ez at amsat.org>
To: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>; <gps_standard at mailman.qth.net>
Sent: Sunday, February 18, 2007 7:53 AM
Subject: Re: [GPS_Standard] vcxo control averaging?

Hallo Bert,

The integration still is an interesting point. As I wrote, using the HP
oscillator a sample time of 5000 seconds is about optimum. But in the
current program there still is "recent jitter" present which in my
experience should become less if the sample basic unit can be extended to
16^x seconds.

The strange exponents I encounter are in the average calculation.

Regards

Arie

----- Original Message -----
From: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>
To: "Arie Dogterom" <pa0ez at amsat.org>; <gps_standard at mailman.qth.net>
Sent: Sunday 18 February 2007 1:04
Subject: Re: [GPS_Standard] vcxo control averaging?

> Hi Arie,
>
> I agree that the longer you average out, the better the filtering will be.
> The S parameter does exactly that: it averages out 16-second frequency
> samples for as many as 65535 samples. This should remove most GPS
> disturbances. Of course I wish we could sample forever....but there is a
> tradeoff between accuracy and integration time.
>
> As for the statistics window, you are the first one to report a potential
> problem with the maths. Which field are you reporting as problematic,
> Standard Deviation or Average Frequency Offset? When I get a chance, I'll
> have a look at the formulas.
>
> Regarding the graph, there is one bug I am aware of, and it is that you
> must
> change the graph type back and forth to update it. So load a Log file, and
> click on a graph type other than the default one, then click back to the
> graph type you want. The graph should now display all the points. Note
> that
> the tool only displays the points for which there was a DAC update. For
> plotting graphs, you may want to use MS Excel or OpenOffice Calc. You'll
> get
> more flexibility...
>
> Regards,
>
> Bert, VE2ZAZ
>
> ----- Original Message -----
> From: "Arie Dogterom" <pa0ez at amsat.org>
> To: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>; <gps_standard at mailman.qth.net>
> Sent: Saturday, February 17, 2007 3:19 PM
> Subject: Re: [GPS_Standard] vcxo control averaging?
>
>
> Hallo Bert,
> I do fully agree with you. Did not realize that some aspects of the PIC
> limit possibilities. But probably a good solution would be to have a
> sampling time larger than 16 seconds, say 16^x seconds. It would require
> some adaptation of the decision level, but the GPS receiver jitter would
> normally be averaged out. According to DF6JB ( an expert in the time
> science) GPS only has a better allen variation than the HP10811 for t>
> than
> about 5000 seconds, so a value of x in the order of  3 would be
> sufficient.
>
> Another matter, but that in fact is another thread, is the monitor
> programme. I like it very much but the statistics part puzzles me. The
> value
> of the average deviation in general has an exponent e-10, sometimes e-11
> en
> very seldom e-12 but then the next values have exponents e-25 or e-26.
> Nothing in between. Is there a slight error in the calculations?
> I have not yet been able to use the programme to arrive at a graphic with
> the control voltage as a function of time. It only shows with me a
> southwest
> to notheast straight line.
> 73
> Arie
>
>
> ----- Original Message -----
> From: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>
> To: "Arie Dogterom" <pa0ez at amsat.org>; <gps_standard at mailman.qth.net>
> Sent: Friday 26 January 2007 1:24
> Subject: Re: [GPS_Standard] vcxo control averaging?
>
>
>> Hi again Arie,
>>
>> This is the reaction we have when we see it happening. We think that one
>> sample or two out of 500 screwed up everything. Statistically, if after
>> 498
>> samples, the accumulated frequency difference is 0, we can say that the
>> system is quite close to nominal frequency, BUT we cannot claim it is
>> right
>> on frequency. I'll bet you that the 497th sample did not give an
>> accumulated
>> difference of 0...Right? So we are interested in the "general trend" over
>> several samples. Ideally, it would have been better to perform a true
>> average calculation (sum of all samples divided by the number of
>> samples).
>> Unfortunately, this requires floating point calculations and the PIC
>> micro
>> is not designed to easily do this. Note that the Montrol software
>> performs
>> a
>> true average in the statistics window. Have a look for fun...
>>
>> Now, in order to make sure that a couple of samples do not impact a long
>> averaging cycle, I have added parameter N, the Frequency Change Negate
>> threshold. Please read the definition carefully in the user manual. With
>> this, you will deny a frequency change if the accumulated frequency
>> difference is too close to nominal frequency. By the way, I recommend
>> using
>> the voting mode on long averaging cycles.
>>
>> One more comment, the Holdover Limit (H) parameter is used on each and
>> every
>> sample, not on the accumulated frequency difference.
>>
>> I hope I have answered your question(s).
>>
>> Best Regards,
>>
>> Bert, VE2ZAZ
>>
>>
>> ----- Original Message -----
>> From: "Arie Dogterom" <pa0ez at amsat.org>
>> To: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>; <gps_standard at mailman.qth.net>
>> Sent: Thursday, January 25, 2007 8:04 AM
>> Subject: Re: [GPS_Standard] vcxo control averaging?
>>
>>
>> Dr OM Bert,
>> I think I did not formulate the point correctly.
>> In fact I did all you suggested.
>> But my point is different: After a certain number of samples, albeit 50
>> or
>> 500 the difference counter can give any value between 0 and the holdover
>> threshold, although the long term average is nearly or exactly zero. For
>> example I see in a random case ( and this is not much different in the
>> two
>> modes) with a 500 sample period that after 498 samples the cumulative sum
>> is
>> 0. After 499 it is -1 and at 500 it is -3. Well that results in a change
>> of
>> the dac output. But the real average deviation is very near to zero.
>> Or do I see this incorrectly??
>> 73
>> Arie
>>
>> ----- Original Message -----
>> From: "Bert, VE2ZAZ" <ve2zaz at sympatico.ca>
>> To: "Arie Dogterom" <pa0ez at amsat.org>; <gps_standard at mailman.qth.net>
>> Sent: Thursday, January 25, 2007 1:20 PM
>> Subject: Re: [GPS_Standard] vcxo control averaging?
>>
>>
>>> Hello Arie,
>>>
>>> You seem to have left your parameters to the initialisation mode. Once
>>> the
>>> FLL has reached equilibrium, you must set your parameters for a much
>>> longer
>>> averaging period. As a minimum, perform two things:
>>>
>>> Your Ageraging Sample size (S) is set too small. 10 samples are not
>>> enough.
>>> To really try to measure to a better accuracy, you must set your S
>>> parameter
>>> to something, say, more than 100. A one hour long period is a good
>>> start.
>>> Set S to 225.
>>>
>>> Also, to decrease the impact of one "way off" sample, set your FLL
>>> Sampling
>>> mode (M) to Voting.
>>>
>>> Please read my website and the user manual. I explain all of this in
>>> more
>>> detail in there.
>>>
>>> Regards,
>>>
>>> Bert, VE2ZAZ
>>>
>>> ----- Original Message -----
>>> From: "Arie Dogterom" <pa0ez at amsat.org>
>>> To: <gps_standard at mailman.qth.net>
>>> Sent: Thursday, January 25, 2007 6:55 AM
>>> Subject: [GPS_Standard] vcxo control averaging?
>>>
>>>
>>> Hello,
>>> I have built the gps controlled standard and am very impressed by the
>>> intelligent control and monitoring software.
>>> But one question remains:
>>> Say one sets a period of 10*16 seconds to average out the "noise"on the
>>> one
>>> sec impulses. If the vcxo is about correct on frequency the result of
>>> the
>>> counter jumps around 0 but I see deviations of +-4 at times. The PC
>>> programme statistics picture indicates gradually an average which goes
>>> down
>>> and reaches values around 10^-11. But at the end of the averaging period
>>> it
>>> is possible that just at that moment the result of the 16 sec
>>> measurement
>>> indicates not 0 but + or - 3. In that case, although the frequency is
>>> almost
>>> correct, the control voltage suddenly is changed and the measuring
>>> period
>>> starts again. But now it is possible that the vcxo is slightly set away
>>> from
>>> the correct figure.
>>> I would think a better solution would be that not the momentary
>>> deviation
>>> but the long time average is used to set the control voltage.
>>> Or do I see this incorrectly?.
>>> Arie
>>>
>>> ______________________________________________________________
>>> GPS_Standard mailing list
>>> Home: http://mailman.qth.net/mailman/listinfo/gps_standard
>>> Help: http://mailman.qth.net/mmmain.htm
>>> Post: mailto:GPS_Standard at mailman.qth.net
>>>
>>>
>>>
>>
>>
>>
>
>
>