[time-nuts] GPS: ADEV or MDEV?

Dr Bruce Griffiths bruce.griffiths at xtra.co.nz
Sun Apr 22 05:28:06 EDT 2007


SAIDJACK at aol.com wrote:
> Hi guys,
> Some comments:
> * The Vref output of most OCXO's is from a Zener diode inside the can.  These 
> typically have aging, thermal sensitivity and very poor voltage  accuracy, 
> and there are much better monolithic high-precision, low-tempco  voltage 
> reference available on the market now (Digikey etc). Depending on the  internal 
> design of the OCXO, it may help the noise and stability to add an  external 100uF 
> or larger tantalum cap to the Vref pin, since it's voltage is  likely used as 
> the power supply for the Oscillator etc.
If you want the lowest tempco and drift available try using an LTZ1000 
(used in HP 8 1/2 digit DVMs).
> * Dithering the DAC to get better resolution is not a good idea, since  it 
> will create spurs. If not through the low-pass filter into the EFC pin, then  
> through the power supply or ground rails into the PCB, or through all that  
> digital noise being generated by the constant updates of the DAC. It's  better to 
> cascade two dacs through a matching network (coarse/fine DAC) and have  as few 
> digital traces switch as possible. These DAC's as well as the matching  
> resistors, and especially the DAC voltage reference should be as low tempco as  
> possible. For the real time nut, the DAC's and Voltage reference will  be 
> selected to have canceling Temperature Tempco's :)
Surely there are problems with DNL when the output of the MSDAC changes?
Unless off course you use the 24 bit ADC to measure the composite output 
and correct this problem.
> One advantage of the cascaded DAC is that we can use 14, 16, or  even 20 bit 
> DACS and their DNL/INL only needs to have 12+ bits of accuracy to  make the 
> compound DAC linear enough for EFC control. For the dithered single DAC  
> approach the DNL/INL is totally critical (better than 1 LSB needed) to make  the 
> dithering work well.
Surely for the dithering approach only the DNL/monotonicity matters.
> * Putting Digital circuits (read DAC) into the OCXO can is not a good idea,  
> we really don't want high edge rate (<10ns rise/fall time) inside the analog  
> OCXO can. It's gonna put noise on the sine wave. Plus it requires a lot of  
> additional pins (well, at least one).
> * We investigated the effect of varying OCXO current on the  
> single-ground-pin Euro can (BTW: whoever designed that can without Kelvin  Sensing was just 
> not thinking right at the time). One way around the problem is  simply to solder 
> a thick ground wire to the can's case next to the VCC pin  through which the 
> oven current will pass, and use the main ground pin (placed  far away from the 
> VCC pin, another stupid idea) as the EFC Kelvin return. But  keep the 
> soldering time very short since the oven may overheat otherwise.
> As Bruce mentioned, measuring the oven current and compensating for it  
> electronically is another way to fix the problem. This is exactly what we do on  
> the Fury GPSDO, with a highly accurate 24 bit Sigma Delta instrumentation ADC.  
> unfortunately that part and the sensing element have their own very small  
> tempco's of course.
> With careful layout, ensuring that the oven current does not pass in the  
> direction of the DAC/voltage reference etc it is possible to mitigate the Kelvin  
> sensing problem of Euro-can OCXO's  to negligible levels.
> * Bruce forgot to mention the use of smart algorithms like Kallmann  
> filtering to measure, and compensate for aging and temperature effects, this is  
> especially important for hold-over performance and also to minimize errors due  to 
> diurnal temp changes etc. The algorithm should be as proactive as possible,  
> so that only small errors need to be corrected by GPS control (such as the rate 
>  of change of the aging rate).
Yes thanks for  that, I had such algorithms in mind but forgot to add 
them to the list as I had to leave and check a lower accuracy system 
(where millisecond accuracy is overkill for visual timing) intended for 
occultation timing at the local astronomical society.
> * maybe a bit above home-brew budgets, but being the lowest-cost  
> new-with-warranty GPSDO on the market as far as we know, the Jackson-Labs  Fury GPSDO 
> almost fulfill's or exceeds all of Bruce's requirements  :) I think our entry 
> level pricing is lower than most Z3801A's on Ebay  these days *
> bye,
> Said
Far better performance is possible when using carrier phase tracking 
techniques, however considerable software development is required before 
it's readily available at low cost (several orders of magnitude less 
than the Quartzlock carrier phase disciplined OCXO) 
implementation.Various tradeoffs such as using a non oven crystal 
oscillator when one only requires a short term stability somewhat better 
than 1E-9 may be then feasible.


More information about the time-nuts mailing list