On 6/6/2017 4:26 AM, Poul-Henning Kamp wrote:

My experience has been that, while ovens may not be perfect,
they are inherently linear.  So the exterior temperature is
a don't care.

I extensively characterized the E1938A oven.  By adjusting
the ratio of heat to the top/bottom vs edge, I was able to
get the thermal gain into the millions.  At that point,
finally a modicum of non-linearity showed up and the thermal
gain varied with ambient temperature.  It might be 2 million
at the ambient where I adjusted it, and drop to 1.5 million
when well away from that temperature.  Or it might change
sign at some ambient.  (Yes, you can have negative thermal
gain).  You shouldn't need to worry about this for any ordinary oven.

Rick N6RK

On Tue, 6 Jun 2017 11:00:57 -0700
Chris Albertson albertson.chris@gmail.com wrote:

You are looking at one minor issue here. Of course, if your control loop
is only of the bang-bang kind, then you will have a hard time to keep
the system parameter stable. But that is easy to deal with if one knows
the overall system.

I was talking about the control therotical "controllability":
https://en.wikipedia.org/wiki/Controllability

For a simple system like an OCXO that is mostly dependent on
thermal mass vs thermal flow (both in and out) which translates
into "delay" between the heating element and the temperature sensor.
A lot of people assume that adding more thermal mass is going to make
the control more stable. But in reality this might make it unstable
(aka cause oscillations) or make the control error larger.
What happens here is that the mass takes time to heat up. During heat
up you can describe it like a (very slow) transmission line. It will
take time until the signal (heat) reaches the other end (center of the mass).
When the heat reaches the sensor, the control electronics will dial the heater
down. But there is still a heatwave traveling inwards, ie the core will get
warmer and warmer. Thus the heater will be dialed down more and more until
it doesn't heat enough. This will cause a cold wave traveling inwards..

Having a larger thermal flow vs mass helps against this problem.
Good thermal conductivity between heater and sensor helps as well.
(that's why you will read often, that the sensor should be placed
close to the heater and not to the quartz)

The control loop does not need something slow to change. You need to
factor the termal mass, its insulation etc into PID parameters so you
get a stable loop. As I have shown above, if this is not correctly done
you will get oscilations. One way to avoid them, if your physical system
is fixed, is to lower the loop bandwith and thus make the system respond
much slower to changes than the time it takes to conduct the heat from
the heater to the sensor. But this means also that the control loop will
be slow to react to changes in the environment.

There are more sophisticated control loop designs that can handle this
better, eg by using two temperature sensors, one at the crystal and
one at the heater. But designing them correctly is more difficult
than the normal PID loop.

		Attila Kinali

--
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor

On Mon, 5 Jun 2017 20:21:10 -0400
Bob kb8tq kb8tq@n1k.org wrote:

That's the first time I hear of modes drifting respective to eachother.
Do you have any references I could read on this?

I always wondered why the MCXO approach was not used more often.
Or why none of the OCXOs used a dual mode approach to sense
the temperature of the crystal directly instead of using a
thermistor.

		Attila Kinali

--
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor

Hi

If you do the classic MCXO with two oscillator circuits and one resonator, the issue is
pretty simple. You have a load capacitance on the fundamental. You have a load capacitance
on the third overtone. Even if it is the exact same capacitor, the tuning sensitivity on
the fundamental is different than the sensitivity on the third overtone. As the load impedance
changes (parts do drift) the delta between the two modes will show up as an offset between
them. If you run through the math, it gives you a delta temperature. How much? How fast? Obviously
that depends. When I brought this up at the time with the authors of the paper, the reply was that
a recalibration of the MCXO was provided for for this reason.

Bob

Keeping with the thread topic, I think this is the key.    For the cost of
only one more cheap sensor you gain a lot.  Harder design as you say but
getting help on-line seems to be free.

I have gotten PID to work myself with linear systems (motor speed) and I
reading up on Kalman Filters as I need them for navigation using multiple
sensors.

I guess one could use the crystal frequency as a measure of its temperature
to tune the system.  Is there a name to Google to read up on using two
sensors and a pid-like algorithm?

Chris Albertson
Redondo Beach, California

Hi

In the case of a second sensor, “at the crystal” effectively means “inside the crystal package”.
That heads you into all sorts of “interesting” problems. Better to just read the papers and do
it the “old fashioned” way.

Bob

In the E1938A oscillator, we used a PIDI^2 loop.  IOW, a PID
plus a double integrator.  This was Len Cutler's idea.
Once the constants were dialed in, this worked phenomenally
well in terms of transient response.  Even dumping in liquid
nitrogen full throttle into the environmental test chamber
barely wiggled the crystal temperature/frequency.

Rick N6RK

On 6/6/2017 3:16 PM, Bob kb8tq wrote:

I don't understand what you are talking about here.  The tempco
difference between modes is unrelated to load capacitance.  The
dual mode idea would work just as well if the oscillators
operated at series resonance.

[I attended this talk in person ~25 years ago; it got a lot of
interest].

The reason why the SC cut mode C and mode B dual mode patent
from HP fell out of favor was the problem with activity dips
in mode B.  Otherwise, it was a great idea.  It would still
be fine for an OCXO, where you just avoid activity dips.
However, the circuit design is very complicated.

Rick N6RK

Hi

The circuit that Stan Shadowski presented is a fundamental / third overtone dual. The example
below is based on that circuit.

Let’s say both modes are running into a 32 pf load and it is a single capacitor.

The capacitor changes due to aging by 1 pf, you now are at 33 pf load.

The fundamental changes frequency ~ 3X as much (in ppm) as the third overtone.

The beat frequency shifts since the two modes do not tune identically.

Beat frequency shift = temperature error.

Yes the example is a little contrived. The real numbers would depend a bit on the design of
the crystal used.

Bob

On Tue, 6 Jun 2017 19:07:29 -0700
"Richard (Rick) Karlquist" richard@karlquist.com wrote:

The E1938 is kind of special in several ways, IMHO.
Beside having a nice zero-gradient design (I really love
this idea :-) it has also a quite large surface vs volume.
This means that there is a lot of heat flow out of the
can and at the same time the (face) heater is large, which
makes the PI loop better behaved. Also the thermal mass of the
crystal holder is quite small. Especially compared ot the 10811.

Due to the flat puck design, I assume that the majority of the
heat going to the crystal holder is due to radiation (unless I
missed some insulation around the crystal). Radiation, albeit being
a high "resistivity" transport mechanism, has very low inherent
"capacitance". And thus the delay associated with this transport
mechanism is low.

There are probably a lot more small design decisions in the E1938
that make it such a superb oven. More than I probably will ever
be able to figure out. And, I still am astonished how well it works.

		Attila Kinalid

--
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor