[time-nuts] Fury Interface Board: How about TI OPA277?

[kevin-usenet at horizon.com]

> Even better use a device with built in trimmed resistors.

But they usually only come in decade gains.  1:2 and 1:4 limit
you to only a few devices which may not be suitable on other
specs.

> Try actually reading the OCXO datasheet rather than speculating.

> The simple zener regulator built into the OCXO has finite rejection of
> input supply fluctuations.
> The output buffer stage is supplied directly from the voltage at the
> OCXO pins, it has no internal voltage regulator.

Well, I was busy and trying to respond quickly.  And, more importantly,
I though we were talking about OCXOs in general and not one particular
make or model.

> Stop speculating and look at the actual circuit schematics for the
> 10811A and the 10544A.
> neither the oven controller nor the oven heater are powered from an
> internal regulator.

Okay, so I did.  Well, the former, at least.
It sure looks to me like the 10811A manual section 8-35 and associated
Figure 8.10 say that the oven controller is powered from precision 10.0V
regulator U2.

The text explains that this is required to maintain consistent
self-heating in thermistor RT1.

However, the heater transistors aren't regulated, so frequencies higher
than the heater bandwidth will get in.  Of course, much above that
and the oven's thermal mass filters it out.  (But the two regions
can't overlap or the controller would be unstable.)

> There's also the voltage drop across the OCXO socket contacts with the
> 10811A and the 10544A.
> This will be modulated by the oven current variations.

That's why both of those have separate oven power supply pins.
I'm a little confused how someone can berate me for not reading
schematics and not be aware that OCXOs usually have separate oven power
supply connections for exactly this reason.  (More generally, separate
the high-current loads fromn the precision circuitry.)  In particular,
the 10811A and 10544A have completely isolated overn control circuitry.
They don't share a ground or anything.

(Or do you imagine that the oven controller is hurt by a bit of
series resistance?  If it really is tuned for maximum stable gain,
then adding enough negative resistance in series to cancel the socket
wiring resistance could make it unstable.)

>> Good point; I just mentioned the TL431 because it was already there.
>> FWIW, the LT1019 is excellent.

> If you don't mind its noise.

Noise?  I use it, and many others use it, for 24-bit ADC applications
specifically because of its extremely low noise.  The manufacturer
boasts that they are 100% noise tested.

Maxim make some parts with claimed lower noise, but they are infamous
for parts availability problems, so I only recommend them to my enemies.

Are you talking about some frequency range that I never use?

>> Personally, I like the idea of feeding the EFC to a high impedance
>> amplifier.  However, I am afraid that some would object to having the
>> EFC go through an op-amp. It will add noise and distortion.
>>     
>
> It's going through an op-amp on the Fury board anyway.  If we have any
> specs on that, we can try to match it rather than overdesigning.

   
> Yes and distortion is virtually irrelevant as the EFC frequency versus
> voltage characteristic is unlikely to be that linear.
  
It's important to reassure people that there are some specs that
can be ignored!

> The INA114 has much lower bias currents than the INA103 but again has
> high noise and drift at low gains.

Whoops!  I was looking at the 0.25 uV/C input offset drift spec, and
missed the 5/G uV/C output offset spec.

Since this is basically a DC application, maybe it's time to dig
out the zero-drift op-amps?

Why are we worried about noise, per se?  Especially at high frequencies?
The oscillator is a gret big integrator that averages it out.

>> H'm... and if you really want the full +/-10V range, fitting a regulator
>> and op-amp into the 2V of available headroom requires an LDO and a
>> rail-to-rail output op-amp.
  
> Most rail to rail output amplifiers tend to have poor PSRR when the
> output approaches the rails.

>> Outside the pi filters, take the EFC supply, drop it
>> through an LDO, and watch the PSRR specs on the EFC components.

> LDO's tend to be noisy and have poor high frequency line rejection.

I apologize for not explicitly stating that.  The reason I drew
attention to the necessity was precisely because PSRR is difficult
with lower dropout.

If we had a bit more headroom (+/-15V supply is traditional for +/-10V
signals), it would be a lot easier.

> However with an EFC voltage swing of 20V the absolute noise and
> stability requirements may be relaxed somewhat but it depends on the OCXO.
> Having specifications for candidate OCXOs having -10V to +10V EFC swing
> would help in making a rational selection of components possible.

Good point; more data would be appreicated.

>> The brute force approach is to get 3 separate supplies. However, I
>> don't think I need to go that far.
>>
>> Any ideas here are most welcome.
>>     
>
> Do you have any idea what the external power supply will be?
> It could be taken into account.
>
> It's noise coming OUT of the oven supply that matters, so subregulating
> it doesn't help; muting the current spikes requires energy storage
> (L and C).  Outside the pi filters, take the EFC supply, drop it
> through an LDO, and watch the PSRR specs on the EFC components.
>
>   
LDO's tend to be noisy and have poor high frequency line rejection.
> (Remember, the OCXO is not *that* sensitive to its own oven hash,
> or it wouldn't work in the real world.)
>

> I often use the little things called "wound beads" or wideband
> chokes", e.g. http://www.ctparts.com/widebandchokes.asp, for generic
> supply filtering, but you may want a little more L in your pi filter.

A much higher inductance than can be achieved with a ferrite wideband
choke is required as the 10544A oven switching frequency is relatively
low being determined by a UJT oscillator.

The appropriate filters and supply regulators for the 10544A are
detailed on the datasheets which are available on Didier's site.

> The recommended circuit has much lower noise than any 3 terminal LDO
> regulator.

The 10 uF/10 mH LC filter in Figure 3?  I'm afraid that I don't
understand what you mean by "noise".  Could we clarify this?

As an ideal circuit made of lossless components, obviously it has
zero noise.

Real regulators have several parameters that describe the output
variation.  It varies with temperature, load current, supply ("line")
voltage, and there's a little bit of intrinsic noise associated with
the voltage reference and amplifiers.

The latter is quite low, and as real circuits are only affected
minimally by power supply variations (see a typical op-amp's PSRR
specification), I've been assuming that any cheap regulator
has an adequately low intrinsic noise.

The OCXO circuitry is a constant load, so load regulation is not a
significant concern.

But we are worried about power supply noise getting into the EFC signal.
This is not noise intrinsic to the regulator, but something that we
want it to filter out.  In a regulator, it's generally called the
"ripple rejection" spec, and tends to be good at low frequencies,
and get worse at higher ones.

So, for example, that 10mH/10uF circuit has a nice gain spike at 500 Hz.
Supply noise at that frequency will be amplified at the output.

And, in fact, its ripple rejection only crosses that of an LT1763
(for example) at 10 kHz and 60 dB.  At frequencies below that, the
regulator filters better than HP's recommended LC circuit.

As I mentioned, for timekeeping applications, high-frequency noise
on the EFC pin has only a second-order effect; the linear effects
are filtered out by the inherent integration done by timekeeping.

So I'm worried about contaminating signals on the order of 1 Hz or
below, what is often called "drift" rather than "noise".  (They're
really just the same thing, unwanted variation, but drift is
often specified as a function of temperature.)

And at those frequencies, passive filtering doesn't work.
(Assuming passives of reasonable size.)


So when you say that it "has much lower noise than any 3 terminal
LDO", I'm confused.  Obviously, lacking any source of Johnson noise,
its intrinsic noise is lower than any device containing a resistor or a
transistor.  But that's so trivial that it's hard to imagine that's what
you mean, and it's unclear why you'd limit yourself to LDO regulators.

The ripple rejection (how good it is at keeping external "noise" signals
out) is demonstrably NOT consistently better than a 3-terminal LDO.
It is better for high enough frequencies (which is preceisely why
people add passive low-pass filters to regulator outputs), but does no
attenuation at all for frequencies below 500 Hz.


"Noise" is very much NOT a simple parameter that cen be described by a
single number, so there is not a total ordering.  Especially when
used informally to refer to things like interfering signals rather than
various forms of (amplified) quantum noise.

Of all the things called "noise", which are you referring to?