[time-nuts] GPS Disciplined TCXO

[Nick Sayer]

> On Oct 23, 2015, at 9:10 AM, Attila Kinali <attila at kinali.ch> wrote:
> 
> On Fri, 23 Oct 2015 07:30:10 -0700
> Nick Sayer via time-nuts <time-nuts at febo.com> wrote:
> 
> 
>>> BTW: Tom van Baak mentioned in a private mail, that the DOC supply current
>>> is much higher than i thought it was. So it isn't exactly a drop in replacement
>>> either.
>> 
>> I’ve run them with the same LDO and they work. The LDO gets warm, of
>> course, but then, so does the oscillator.
> 
> The LM1117 is specced for 800mA output. The DOC is specced with max 2.5W
> at warmup, or 750mA @ 3.3V. Assuming the rest of the circuit does not use
> more than 50mA, that's barely within specs. I would guess that going up
> to 900mA shouldn't fry the LM1117 as long as you can deal with the produced
> heat. Even in the best case, it's too close for my engineers love of
> safety margins :-)

I agree. The only mollification I can offer is that the DOC pulls warm-up currently only for a few seconds before ramping down to ~330 mA steady-state.

But in any event, testing with the switcher showed that there was no benefit to an LDO for OCXOs. This was confirmed by Connor Winfield. I talked to them about ripple and noise tolerance and they said that their OCXOs have separate internal regulators for the oscillator, so the 3 mV P-P that I get out of the SC189Z is just fine. The extra capacitance on the DAC supply pins brings that down a bit more by the time the power gets over there, but the ripple is ~2.5 MHz, so it’s pretty much gone from the DAC output after the LPF. Any non-linearity that might result simply results in the discipline settling on a “different” value.

On the other hand, attempting to use the switcher with the DOT050V was a disaster. I had to add extra dummy load to bring the ripple down to 3 mV P-P, and even then the short term stability of the DOT was an order of magnitude worse. The extra load made the efficiency gains moot, and the actual load is low enough that the inefficiency of the LDO is hardly an issue. That was about the same time I gave up on trying to use the DOC part, so the need for a unified design went away. I was left with two boards - an LDO feeding the DOT050V and the SC189Z feeding an OH300.


> 
>> When I compare them to the same
>> circuit, but powered with the SC189Z switcher I use for the OH300 variant,
>> their performance is identical. Which is to say, a half order of magnitude
>> worse than the DOT050V at tau 1s.
> 
> Hmm... For a circuit like this, I would use a switched DC/DC converter
> to get down to 3.8V or even 3.5V, then use a modern LDO with high ripple
> rejection and low drop-out (<0.5V). This solution will be more expensive
> then a simple LM1117, but the heating of the circuit would be approximately
> independent of the input voltage and also much lower.
> 

I thought about that, but it turns out that 3 mV P-P is good enough. When I compare the DOC with an LDO to the SC189Z, the performance is identical. I haven’t tried an OH300 with an LDO, but its performance with the switcher is in the ballpark of spec (albeit with sample size 1).

> 
>> The data sheets of both claim a short term stability of 1e-9 at 1s. The DOC
>> is exceeding that, but the DOT is *slaughtering* it. I routinely see 9e-11
>> instead. I asked Connor Winfield about that and they said that they should
>> change the data sheet, as my observations are more in line with reality.
> 
> But no explanation why this might be the case?

Other than their TCXOs are really good? :)

> 
>> Meanwhile, the OH300 (I’ve only tried one of them so far, however) is at
>> around 3e-11 at 1s. So that’s what $100 buys you. :)
> 
> I'm taking notes here :-)
> 
>> The other thing I saw with the DOC020V that I have not seen with either the
>> DOT or OH300 variants is a weird oscillation on the VC pin. I posed the
>> question to Stack Exchange and they suggested that the VC pin internally goes
>> to a varactor which adds too much capacitance for the output of the OP amp. 
> 
> That sounds kind of strange. Yes, too much capacitance could lead to
> oscillations. But given the fact that the datasheet of the AD8538
> has graphs that go up to 1000pF load capacitance (although with huge
> overshots) I would expect it to be stable at least up to 100pF (where
> the overshot graph starts to go up). Most Varicaps are usually in the
> range of 2-20pF, there are a few that go up to 100pF, but it'd be
> surprised if they'd use one of these.
> 
> You can "easily" try whether the varicap theory is right, if you
> add an 2-10k resistor in the path between the AD8538 and the VCO/OCXO.

The oscillations stayed on the oscillator side and disappeared from the amp side of the resistor. It took adding capacitance to the VC pin (0.1µF) to damp the oscillations away. The final configuration was 0.1µF and 330Ω. With that, you could see very small little “ticks” where it was trying to start oscillating, but they were only a few hundred µV and decayed to average very quickly. I really don’t know for sure what the deal was, but it was/is unique to the DOC020V (sample size: two).

> 
>> I asked Connor Winfield about this and they didn’t have an immediate answer. 
>> The latest boards have the footprints for another RC LPF on the output to
>> counter this, but since I gave up on the DOC part, the cap is no-stuff and
>> the resistor is 0 ohms when I build them out (with the other parts).
> 
> ACK
> 
>>> 
>>> In the AD5061 datasheet (Rev C) on page 3, at the bottom,
>>> section "Reference Input/Output", item "V_REF Input Range".
>>> Yes, it will not fry the device, but it will not work up to spec either
>>> (unfortunately, manufacturers never mention what part of the spec gets
>>> degraded if you violate something).
>> 
>> I do see that now. If you look at the footnote, figure 27 which it references 
>> shows that the Vdd headroom with a VRef of 3.3 volts is more like 20 mV or 
>> so. I would posit that the headroom issue is more likely to result in non-
>> linearity at high DAC values, perhaps because the output amp may not be 
>> capable of full rail-to-rail operation. 
> 
> It is not clear where the limit comes from. It could be the ouptut buffer,
> which would then be a no-problem. Or it could be the input buffer for V_ref,
> which then would lead to an un-buffered, and hence unstable internal reference
> voltage. Or it could be a problem of one of the DAC internal stages, that
> needs the headroom….

I’ll try asking Analog Devices and see if I can get an answer to the question of what happens when Vdd=Vref=3.3V. All I know is that in practice, it’s working as designed, including meeting the gross tuning slope predictions within 10%.

> BTW: 20mV or even 50mV below VDD is considered close to rail
> (as in rail-to-rail). You need to compare this to the 1-2V (sometimes
> even 3V) that opamps needed as headroom (both at input and output) from
> the rails before rail-to-rail opams became en-vogue. IIRC any headroom
> around/below 100-200mV is considered close to rail.
> 
> 
>> In practice, the VC output of the analog section is quite stable (more stable > than the analog supply rail), however, because of the LPF on the DAC's 
>> output. Remember that the DAC was designed with a maximum clock speed of
>> 30 MHz and we’re using it effectively at 10 mHz. So the LPF trades bandwidth 
>> and slew rate (neither of which are interesting in this application) for 
>> stability. And to some extent you’d be throwing away the pains to which you 
>> went to come up with a super-accurate Vref because you then feed the DAC 
>> output into an op amp that’s powered by the analog supply rail, so its own 
>> noise and non-linearity would likely have a greater impact (at least, I would 
>> expect so).
> 
> Noise itself is not a problem. As long as it's "high" frequency (>10Hz or so)
> and not too large, it will get averaged out by the long integration times
> of the VCO. Yes, it decreases the short-term stability and phase noise 
> performance, but that's not the main selling point here, is it?

Well, I do hope to at least let the parts achieve as close the performance they’re designed to achieve as is reasonable. That’s always the thing, isn’t it?  There’s always ways to make the design better if price is no object…

My 52330A, unfortunately, can’t perform amplitude or noise measurements, so I can’t validate the design against that aspect of the spec. It’s also conceivable that the NB3N551 output buffer may degrade the performance from those specs as well. Performance below tau 1s wasn’t a design goal for me. My use case for this is as an external reference for lab equipment (like the aforementioned 52330A). In fact, the *original* use case was to feed 10 MHz into a home-brew specialized frequency counter to calibrate my Crazy Clock lavet stepper driver boards. For that, I need to measure a 16.384 kHz square wave for offset down to a granularity of 0.1 ppm, but I wound up moving the target out a bit further. :)


> On the
> other hand, long term stability (>1s) is important and does improve with a
> more stable DAC reference voltage. That the Opamp power supply is not as
> stable might be a little bit of a problem, but shouldn't be too big, as the
> AD8538 has a power supply rejection ratio of >100dB, ie power supply 
> noise/variations should be very small compared to the instabilities of
> the DAC and Vref.

Oh, I should say, I’m using the AD8691 now rather than the 8538. It’s a few pennies cheaper and seems to perform just as well in this application.

I haven’t focused on power supply stability beyond looking at noise and ripple with my scope, AC coupled (and I’ve been rather careful how I do that - I’m using the little “springy wire” grounding… thing… on my probe - set to 1x - to measure directly across filter caps and stuff like that). That necessarily implies a certain minimum timeframe below which I have not attempted to examine, so it’s conceivable that intermediate tau (1s < tau < 100s) is degraded and I’m none the wiser. The ADEV graphs I get do show a small rise out to ~1e3, so maybe that’s the explanation. After 1e2, I would expect to see the discipline take over, since 100 seconds is the PPS sample window the firmware’s using.


> 
> 
> 
> 				Attila Kinali
> 
> -- 
> It is upon moral qualities that a society is ultimately founded. All 
> the prosperity and technological sophistication in the world is of no 
> use without that foundation.
>                 -- Miss Matheson, The Diamond Age, Neil Stephenson