[time-nuts] FE-5680A GPS discipline board on sale now

[Nick Sayer]

> On May 20, 2016, at 2:29 AM, Attila Kinali <attila at kinali.ch> wrote:
> 
> Hoi Nick!
> 
> On Thu, 19 May 2016 10:08:10 -0700
> Nick Sayer via time-nuts <time-nuts at febo.com> wrote:
> 
> 
>> This is a mash-up of my breakout board and GPSDO. You give it 30+W 18-24
>> VDC in (hack up a surplus laptop power supply) and it supplies up to 2A @ 15
>> VDC and 500 mA @ 5 VDC. In my testing I see around 25 mV p-p of ripple on the
>> 15V rail. The 5V rail is a bit noisier at around 35 mV. Functionally, the
>> discipline system is the same as the latest OH300 units - it has the phase
>> discriminator that Jim Harman recommended, but now there’s a JFET that acts
>> to make the phase ramp more linear (useful if you want to gather stats).
> 
> Nice! Thanks for sharing!
> 
> Three comments:
> 
> The discharge resistor of the integrator (R13) is with 10M way too high.
> The PCB resistance is usually in the same order of magnitude. Ie with
> a resistance that high, the actual resistance highly depends on how
> clean the board is (finger prints, dust) and the amount of humidity
> it absorbed (both the epoxy and the solder resist are quite hygroscopic).
> A resistance of 1M usally is the maximum recommended value. For systems
> that should run reliably for a long time, i would even try to keep it lower.
> Of course, you have to increase the capacitor accordingly, which also makes
> the capacitor less dependent on parasitics. 

The discharge time constant is not critical to the design. It needs to be long enough to not have a significant impact own the cap over the course of the first few microseconds, but it needs to clear out the cap before the next second. A larger cap requires a smaller resistor, which increases the current from the PLL output pin. I joked at the time that I could just clip a fingernail and glue that onto the pads instead of using an 0805, but the standard deviation would be quite high.

> 
> The FET Q2 makes the phase detector more linear, but also makes it more
> dependent on temperature. Also the V_GSth has a variation of a factor of 3
> between different transistors. But I have to admit that I don't know a
> good solution for this. 

The linearity is optional. It’s only useful if you want to try and make sense of the phase numbers logged. As alternative, you could attempt to curve-fit them instead. It’s a happy coincidence that an 0805 footprint is very similar to the D-S pads of the transistor, so the schematic notes that you can use a 0Ω across D-S if you like. In that case, the code needs to change to select the 4.096V ADC reference instead of the 2.2V one.

For the actual PLL, I believe the impact of either choice is merely to lock onto a different phase center point. And if that center point drifts with temperature, then I think it’ll be dwarfed by the rest of the temperature impact on the system.

As with my TCXO/OCXO based designs, there are a lot of choices that just wind up being “good enough,” given that the target ADEV is easily an order of magnitude higher (10^-11 ish for the OH300) than many Time Nuts(tm) aspire to.


> 
> The output of the FE-5680 is exteremely noisy. A small VCXO locked
> to the 10MHz output using a simple PLL (an XOR gate with RC filter
> should be enough) is recommended to get at least rid of the spurs
> above 100Hz-1kHz

I have a design in mind for an OH300 based clean-up oscillator as a separate module. It would use an OH300 VCOCXO and a 4046 to attempt to lock against an input reference. It would be useful for those who want to use the 5680 in microwave applications (whether or not GPS disciplined) where the low-tau stability is an issue. For myself - at least at the moment - my use case is a reference for my 53220A, so I’m not sure stability below 1s is much of a factor. I do admit, however, to being curious as to whether I can do significantly better than my OH300 GPSDO with such a configuration. There’s not a lot of difference between the two when I measure them (that said, I am measuring them both against a Thunderbolt and have not yet attempted the three-cornered hat to tease out how much impact it’s having on the results). A comparison ADEV graph is on the Tindie store page.

Another use case for this GPS discipline module is to use it to figure out the correct tuning value for a particular oscillator, then unplug it and use the DOS software to commit that value to EEPROM (multiplying the logged value by 4 to make up for the resolution reduction). Having done that, you now have a portable calibrated reference. From what I can tell the holdover performance of the 5680A is particularly good - exactly why they made them, in fact. If I had just one more input on the controller, I might have added a “commit” button that would send the EEPROM burn command on demand with the current EFC value (the controller is almost out of flash space, though, so that’s an issue).