[time-nuts] Quartz crystal aging -- unlocked Z3801A

[Bob kb8tq]

Hi

Welcome !!!!

> On Apr 6, 2020, at 7:29 PM, Louis Taber <ltaber at gmail.com> wrote:
> 
> Hi All,  My first note to the mailing list.
> 
> I never thought I would be able to do see a quartz crystal age.  All
> earlier attempts were buried in the noise & temperature effects.  This data
> set results in a very straight line.
> 
> This study tracks the 10MHz output of a Z3801A GPS with a Agilent 53310A
> Modulation Domain Analyzer with its reference input connected to an
> FEI-Zyfer Nano Sync II GPSDO.
> The Z3801A GPS receiver has no antenna attached.  It is NOT locked to the
> GPS satellites.  My suspicion is the the receiver's physical location is
> locked.
> 
> I believe the  Z3801A GPS receiver has one of the best SC double oven
> quartz crystal based oscillator ever manufactured.  Perhaps some day it
> will also receive!

The OCXO in the X3801 is just. HP 10811 in a “blanket” to get it to function
better at -40C than the standard part. It’s aging / ADEV / Phase noise is not
really any different than other 10811s. If anything the 3801 parts are a bit 
worse than a typical 10811 due to the wider EFC range on the 3801 models. 

The 10811 was an amazing device in the late 1970’s. Over the many decades
since then, there have been a number of advances in OCXO design ….

> 
> For the majority of my data set I have been recording frequency averages
> once a day.  Each data point represents over 40,000 frequency
> measurements.  The total number of frequency measurements exceeds 2 million
> after February 22, 2020.  February 22 is when I started daily
> measurements.  There is almost a month of additional points not shown in
> the graphs.
> 
> From February 20 to April 6, 2020 the frequency of the unlocked Z3801A has
> aged from 10,000,000.006,069,6 to 10,000,000.008,649,7 or 2.5801mHz over
> 46.7 days or 56.4nHz/Day.  If my calculations are correct that is a
> Δf/f/day of 4.5e-12/day.  Which extrapolates to 20 mHz/year.

At least as I see the data, the unit has moved a total of 2.5 x10^-10 in a bit
over a month. That’s not unheard of on a part that has been on power for a 
long time. Sub 10 ppb per year numbers on well stabilized OCXO’s are 
seen in a lot of cases. 

> 
> A section of my spreadsheet that might not make it to the mailing list.
> The spreadsheet is attached exported from Google Docs to Microsoft .xlxs
> format.
> 
> Days μHz/Day mHz/year years/Hz Δf/f/day ppm/year n
> Averages 1.02 58.16 21.24 39.79 5.8 0.0021 45736
> x 10e-12
> Total 32.67 1463560
> My biggest conclusion is the HP did a GREAT job designing the double oven
> crystal oscillator!

Well, it’s actually a single oven with a booster heater that cuts out in normal 
operation. 

> 
> Also attached is a picture of the display of the Agilent 53310A Modulation
> Domain Analyzer after a day of creating a histogram.
> 
> John R. Vig has published several different versions of *QUARTZ CRYSTAL
> RESONATORS AND OSCILLATORS *over the years.  This is the latest one I am
> aware of:  http://beckelec.com/john-vig-crystal-tutorial.pdf  They are
> different.  Another one is at:
> https://www.am1.us/wp-content/uploads/Documents/U11625_VIG-TUTORIAL.pdf He
> poits out in one that SC crystals are "on the order of 1 million atomic
> layers thick".  Loose an atomic layer of quartz and there goes 1 ppm?
> There is often gold layers for the electrical connection, right where I
> believe the maximum acceleration is.  About 1 million Gs.  Another totally
> rhetorical question, "How many gold molecules need to be dislodged to shift
> the frequency 1 per part in 10e-12? <grin>".

Most aging involves lighter molecules rather than gold. 

> 
> Some more serious questions.
> 
> How many significant digits am I getting after measuring/averaging
> 40,000-60,000  sample?

You would need to do a proper noise floor test to work that out. 

> Why do my histograms have a long thin tail to the higher frequencies (in a
> 4 mHz window)?

Probably some systematic error in your setup. Cell phone signals are one 
common culprit these days. 

> What i causing the double peak in the histograms?

I suspect you are seeing the quantization error from some portion of the 
measurement device. Again, the only way to sort out the “is it the tester or the
tested?” is to do a run on the measurement instrument. 

One thought: by doing your very long data averages, you probably are smoothing 
over a lot of “interesting” features. It might be interesting to drop back to something 
like 120 second averages and plot that. 

Bob

> 
> An LSB in the FEI-Zyfer Nano Sync II DAC driving the VC XO?
> 
> An instability in the 53310A Modulation Analyzer's PLL?
> 
> Also attached is the graph of the oscillators drift with an offset of
> 10,000,000 Hz removed.  The vertical axis is in mHz.  You really need to
> watch the case of the "M"s!
> 
> Louis Taber, AF7XZ
> 
> PS: Another use of the Agilent 53310A can be seen on the Wikipedia WSPR
> page at
> https://en.wikipedia.org/wiki/WSPR_(amateur_radio_software)#/media/File:4-FSK_used_for_WSPR.jpg
> <Counter%20 Agilent 53310A.xlsx><IMG_2723a.JPG><freq.jpg>_______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.