Hi Bob,

Some numbers. Maybe you can double check my math, just to be sure I'm
not getting something completely wrong. That is very possible, since
I'm new here... ;) 
The DAC is moving up and down about 7.5 counts with room temp
swings. 20 bit resolution at 6.6volts full scale output.
6.6 volts * 7.5counts / (2^20) =  +/- 47uV.  (This is verified as
reasonable with a 24bit data logger, as it's seeing about +/-50 uV temp
swings on EFC. Resolution of about 1uV.)
Tuning sensitivity of the oscillator is 1Hz/10Volts. Or 47uV *
1Hz/10V = +/- 4.7uHz. 
The temp swing is +/- 2 degF with ~45 minute period. So, in the
ballpark of your +/- 1 Deg C guess. 
+/-4.7uHz / 10e6 Hz oscillator = +/-4.7e-13, or near a 1e-12 full
swing over 2.2 Deg C. (Or, am I completely out to lunch here???) 
I should qualify, there is aging/retrace here. It's in the range of
3e-11 per day right now, and I took the +/- 7.5counts off of what was
left after removing the slope of the aging drift. The aging looks huge
over a day compared to the thermal cycling. 
Currently the system has ~2ppm/C reference, and .04nV/C opamps. So,
Yeah a little overkill. But these things are getting cheap nowadays, so
why not? Before the 'good' reference and opamps, there was about 10
times as much swing in the PWM DAC over temp cycles. As you have
suggested there is probably some room to 'relax the spec', and still be
fine...
 
Thanks,
Dan
 
 
 
 
 
 

picked to make math easy):

as much as it does to absolute temperature.

over 20C. That’s a cheap regulator for 2x10^-12.

probably good enough. Knocking it down to that level is just a couple
of resistors. Way less money than fancy references.

Hi

or 1x10^-8 per volt. If it’s a 10 MHz OCXO.
That would be 1x10^-14 per uV
4.7 x 10^-13 for 47 uV

That’s pretty normal for a modern HVAC system.

1x10^-12 for full swing is about right.

That’s a very common (and legit) thing to do.

What makes you believe that the OCXO’s temperature performance it not the issue?

1x10^-12 per 2 C -> 1x10^-10 over 100 C (say -30 to +70C). That’s a very good spec on an OCXO. Also consider that gradients could easily amplify the impact by 2X or more.

I’d guess that the analog stuff is much better than it needs to be.

Bob

Hi

Separate from the analysis of the voltage on the OCXO, there is another part to this:

Ok, so why am I harping on the “need” for all this from a system standpoint ?

1. Adding stuff to a design that does not make it measurably better is simply a waste of money. That’s ok if it’s your money.

2. Others read these threads and decide “maybe I need to do that..”. Now it’s a waste of somebody else’s money.

3. Still others look at this and decide “If I need to do that, I’m not even going to start”. That’s not good either.

4. Analysis is part of the design process. It should very much be part of this.

5. Focusing on a design aspect “because I can” is a very common thing. I do it all the time :) Because I fall into the trap often, I recognize just how much time gets soaked up on dead ends this way.

6. There are very real issues when doing this. Sorting out what’s real and what isn’t is far from easy. The more “noise” in with the signal, the less likely others are to figure out a good approach.

I’m quite sure this thread will keep going for quite a while on references. At some point we will be talking about 1 pV / C 1KV designs with 0.1 nV/Hz noise levels. It worries me that people may believe that in some way that applies to a time or frequency standard ….

Bob

Bob wrote:

We've been around this track a time or two before, me frustrated with
your "make it just good enough" philosophy and you with my "always do
the best you can" philosophy.  We're not likely to persuade each
other, or even influence anybody else, but I think it is worth going
around at least one more time.

Preliminary nit:  I agree that any "improvement" that does not make
something measurably better is of no value.  Indeed, it is no
improvement at all.  But you didn't mean literally "not measurably
better" -- you meant "not better for the task at hand."  A digital
caliper reading to 0.0001" is "measurably better" than a ruler
graduated in 1/32 inch, although the difference is not important if
one is measuring the thickness of a 2x4 for framing a house.  But
some day you may want to measure something besides a 2x4....

On to the substance:

"Do the best you can" isn't necessarily about adding anything to a
design.  It's about carefully determining an error budget and
developing a design that meets the budget.  Of course, you can set
the design goals for each subsystem so that the overall system should
juuuust work if everything else is perfect, or so that the system
should work under most conditions, or so you'll never have to
consider whether that subsystem might be contributing anything
significant to the system errors.  If the latter is no more difficult
and no more expensive than either of the former, why WOULDN'T you
design it that way?  I was taught many years ago that "good thinking
doesn't cost any more than bad thinking," and I have generally found
that to be true.  Meaning, it is frequently the case that "the best
you can do" is no more difficult and no more expensive than doing
something less, it just takes better thinking and a more accurate
analysis.  Whenever that is the case, which IME is very often, doing
less is, IMO, a design fault.

Most often, it's a matter of, "Why ground that capacitor there?  Over
here would be better," or "Why use a noninverting amplifier?  If you
use an inverting amplifier, the HF rolloff can continue beyond unity
gain," or something similar.

Note, also, that many of the people asking questions on the list do
not seem to have a thorough design specification for their project,
and may not even know what all they will use a gizmo for.  Settling
for what a list pundit might think is "good enough" for the person's
needs (e.g., residual phase noise floor ~ -150dB and reverse
isolation of ~ 40dB for a buffer amplifier) may turn out to be
inadequate when the person acquires some better oscillators and a
DMTD setup and needs -175dB and 90dB.  If they do the best they can
the first time, they may not have to re-do it later.

Again, neither one is a problem if doing the best one can is no more
difficult and no more expensive than doing something less.  If
someone has already done the good thinking and suggests a workable
approach, and all you have to do is a sanity check to implement the
idea (perhaps even improving on the design), again -- why WOULDN'T
you?  There is always someone handy who is quick to point out all of
the other ways to do things, so the person contemplating the project
can evaluate the different approaches for himself.

Sometimes, of course, going the next step up the "best you can"
ladder involves an expensive part (e.g., silicon-on-sapphire
semiconductors), or a much more complex design, or some use
restriction (must be submerged in liquid nitrogen).  In that case,
one must think very carefully about the error budget and determine if
that step is really necessary.  But the vast majority of the time, we
do not face that situation IME.

The bottom line is:  There is no virtue in doing "just enough,"
certainly not in the case of amateur projects that will not be
manufactured in large numbers for slim profit (where every millipence
must be saved, if the accountants are to be believed -- often, they
shouldn't be, but that's another topic entirely).  Never apologize
for doing better than "just enough," as long as doing so does not
cause collateral problems.

To me, that is the art of design -- knowing that the finished gizmo
is the best I could make at the time and with the resources available.

In philosophy-of-design circles, one sometimes hears that "a race car
should be designed so that everything is totally spent as it crosses
the finish line -- the engine should explode, the transmission should
break, and all four tires should blow out simultaneously.  Anything
that is still working was, by definition, overdesigned."  Aside from
the obvious hyperbole, I think the underlying point is plain
wrong.  I know I admire the designers, whoever they were, when
someone pulls a submarine off the ocean floor after 70 years and the
batteries still have a charge and the gauges and radios still work.

Finally, one not-so-obvious point about amateur designs.  Sometimes,
something is a true one-off -- there will never be another made to
that design.  In that case, some design requirements can be
relaxed.  You can use trimmer caps or resistors where you would
prefer not to in a commercial design, for example, and you may use
disfavored logic kludges to work around timing problems.  But then
there are designs that you will publish or otherwise share -- and
these, I suggest, need to be even more bulletproof than commercial
designs, since you are not in control of the construction, parts
choices, etc. that others who follow your lead will make.  Yes, you
can make disclaimers and suggest where the sensitive bits are, but
for the design to be truly useful to others, you need to pay
attention to all that and design as many of the traps out as you
possibly can -- which can be much harder than designing something to
work properly when it is made in a factory under your supervision.

Best regards,

Charles

Hi Charles,
I hope you don't mind if I throw my two cents in, as this began as a question about my GPSDO project.  We had a thermal drift problem that Dan traced to the PWM to EFC interface and resolved.  The question to the list was whether there was a regulator package that had a built-in reference with good thermal performance.  Somehow the thread went off on all the tangents, which can be good.  In the process it became clear that there weren't any regulators that would fit our needs, so we would have to go with a reference and op-amp etc.  So, now we just need to decide whether to use a pass transistor or a controllable regulator.  The budget will probably result in a 25 or 50 cent pass transistor, a good(ish) op-amp and a reference that's multi-purposed for the board's other needs (ADC, RC integrator, etc).  But it's certainly good to see what the other options are just in case.  And if the project stays a two-off, then there's plenty of leeway to use better parts here and there if the pinouts are the same.

Bob - AE6RV

  From: Charles Steinmetz <csteinmetz@yandex.com>

To: Discussion of precise time and frequency measurement time-nuts@febo.com
Sent: Thursday, December 11, 2014 11:28 PM
Subject: Re: [time-nuts] Linear voltage regulator hints... --> WHY?

Bob wrote:

We've been around this track a time or two before, me frustrated with
your "make it just good enough" philosophy and you with my "always do
the best you can" philosophy.  We're not likely to persuade each
other, or even influence anybody else, but I think it is worth going
around at least one more time.

Preliminary nit:  I agree that any "improvement" that does not make
something measurably better is of no value.  Indeed, it is no
improvement at all.  But you didn't mean literally "not measurably
better" -- you meant "not better for the task at hand."  A digital
caliper reading to 0.0001" is "measurably better" than a ruler
graduated in 1/32 inch, although the difference is not important if
one is measuring the thickness of a 2x4 for framing a house.  But
some day you may want to measure something besides a 2x4....

On to the substance:

"Do the best you can" isn't necessarily about adding anything to a
design.  It's about carefully determining an error budget and
developing a design that meets the budget.  Of course, you can set
the design goals for each subsystem so that the overall system should
juuuust work if everything else is perfect, or so that the system
should work under most conditions, or so you'll never have to
consider whether that subsystem might be contributing anything
significant to the system errors.  If the latter is no more difficult
and no more expensive than either of the former, why WOULDN'T you
design it that way?  I was taught many years ago that "good thinking
doesn't cost any more than bad thinking," and I have generally found
that to be true.  Meaning, it is frequently the case that "the best
you can do" is no more difficult and no more expensive than doing
something less, it just takes better thinking and a more accurate
analysis.  Whenever that is the case, which IME is very often, doing
less is, IMO, a design fault.

Most often, it's a matter of, "Why ground that capacitor there?  Over
here would be better," or "Why use a noninverting amplifier?  If you
use an inverting amplifier, the HF rolloff can continue beyond unity
gain," or something similar.

Note, also, that many of the people asking questions on the list do
not seem to have a thorough design specification for their project,
and may not even know what all they will use a gizmo for.  Settling
for what a list pundit might think is "good enough" for the person's
needs (e.g., residual phase noise floor ~ -150dB and reverse
isolation of ~ 40dB for a buffer amplifier) may turn out to be
inadequate when the person acquires some better oscillators and a
DMTD setup and needs -175dB and 90dB.  If they do the best they can
the first time, they may not have to re-do it later.

Again, neither one is a problem if doing the best one can is no more
difficult and no more expensive than doing something less.  If
someone has already done the good thinking and suggests a workable
approach, and all you have to do is a sanity check to implement the
idea (perhaps even improving on the design), again -- why WOULDN'T
you?  There is always someone handy who is quick to point out all of
the other ways to do things, so the person contemplating the project
can evaluate the different approaches for himself.

Sometimes, of course, going the next step up the "best you can"
ladder involves an expensive part (e.g., silicon-on-sapphire
semiconductors), or a much more complex design, or some use
restriction (must be submerged in liquid nitrogen).  In that case,
one must think very carefully about the error budget and determine if
that step is really necessary.  But the vast majority of the time, we
do not face that situation IME.

The bottom line is:  There is no virtue in doing "just enough,"
certainly not in the case of amateur projects that will not be
manufactured in large numbers for slim profit (where every millipence
must be saved, if the accountants are to be believed -- often, they
shouldn't be, but that's another topic entirely).  Never apologize
for doing better than "just enough," as long as doing so does not
cause collateral problems.

To me, that is the art of design -- knowing that the finished gizmo
is the best I could make at the time and with the resources available.

In philosophy-of-design circles, one sometimes hears that "a race car
should be designed so that everything is totally spent as it crosses
the finish line -- the engine should explode, the transmission should
break, and all four tires should blow out simultaneously.  Anything
that is still working was, by definition, overdesigned."  Aside from
the obvious hyperbole, I think the underlying point is plain
wrong.  I know I admire the designers, whoever they were, when
someone pulls a submarine off the ocean floor after 70 years and the
batteries still have a charge and the gauges and radios still work.

Finally, one not-so-obvious point about amateur designs.  Sometimes,
something is a true one-off -- there will never be another made to
that design.  In that case, some design requirements can be
relaxed.  You can use trimmer caps or resistors where you would
prefer not to in a commercial design, for example, and you may use
disfavored logic kludges to work around timing problems.  But then
there are designs that you will publish or otherwise share -- and
these, I suggest, need to be even more bulletproof than commercial
designs, since you are not in control of the construction, parts
choices, etc. that others who follow your lead will make.  Yes, you
can make disclaimers and suggest where the sensitive bits are, but
for the design to be truly useful to others, you need to pay
attention to all that and design as many of the traps out as you
possibly can -- which can be much harder than designing something to
work properly when it is made in a factory under your supervision.

Best regards,

Charles

time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

In the case at hand, the task is a GPSDO with a frequency vs temperature issue. The issue is coming from the OCXO and not the reference. Improving the reference will (in this case) have no impact on the problem.

It is rare that an improvement does not impact cost or complexity. It most certainly is not the case in this situation.

But - rather than looking at the system and it’s needs, we spin off to “improvements”. Inevitably the result is a -175 db solution to a -145 db problem.

Except that in the actual example case at hand it very much is more expensive and more difficult.

That’s not what’s being done here. The example case is not following the course you are talking about at all.

For a very few people the limit may indeed be liquid helium. There is a much larger group who are turned off at a far lower cost or complexity point.

The issue is not “do people go overboard”, of course they do. Everybody does. Turning that by it’s self into a virtue is a mistake. In 99.99% of the real world cases, cost and complexity will go up and reliability will thus go down. The result will not be better, it will be worse. The best design that achieves the goal is simplest design. That’s every bit as true in the basement as it is in volume production.

Your comments do not address the other side of what I’ve been trying to convey:

Going overboard with no analysis is not a good way to do any of this. Even after getting the system specs and design constraints for this example, we do not bring that into the discussion. We get into long (and very well written) defenses of complexity for it’s own sake. We don’t get into analysis. The takeaway to the casual reader is that complexity is the goal and that analysis is an un-important part of the process that only optional comes into it.  There are a lot of people reading the list who could execute a simple design. There are far fewer who can properly execute a very complex one. The focus on complexity for it’s own sake does have impact.

————————————

Are we really that far apart - not really. We each are talking about two sides of the same coin. The real world is a messy place. Analysis often takes a back seat to the “fun of doing something”. That’s not to say it should though …

Have Fun

Bob

I grew up in an industry where we called everything that was way
overspec'ed, "platinum-iridium xxx".

I think there is a broad interest in e.g. low-tempco engineering or
low-noise regulators and having some in the pocket designs that start with
jellybean discrete parts and occasional hi-spec parts where they actually
matter, is a great idea.

Tim N3QE

On Fri, Dec 12, 2014 at 8:09 AM, Bob Camp kb8tq@n1k.org wrote:

On 12/12/14, 5:09 AM, Bob Camp wrote:
———

And sometimes, the analysis is the fun part.

These days, with awesome computational horsepower at your fingertips,
anywhere, you can do analysis in places and times that are infeasible
for bench work.

I get a lot of useful analysis and modeling done on airplane flights
across the country. No phone calls, no meetings to go to, no emails to
respond to, no self induced distractions. 4-5 glorious mostly
uninterrupted hours with forced seat time. (especially with more planes
having 110V outlets at the seat)

Hi

Which is the reason I split this part of it off from the main thread. There are most certainly reasons why you would use low noise / low temp chef. / low aging references. Digging into that is not a bad thing and I’m in no way knocking that part of it.

Bob

Hi

Indeed

In some cases, the analysis is pretty simple ...

As long as you don’t have some guy named Bob in the seat next to you doing all sorts of distracting things …:)

Bob