[time-nuts] "The Penultimate HP5065 A15"

[Attila Kinali]

Hoi Poul-Henning,

On Fri, 31 Jul 2020 21:31:13 +0000
"Poul-Henning Kamp" <phk at phk.freebsd.dk> wrote:

> The main reason I dont just repair/replace the A15 is that 
> I want to find out how much instability the PSU contributes.

There are very few points where supply voltage (in)stability
contributes to long term frequency drift in a Rb vapor cell standard.
Noise is much more important than stability as it directly
contributes to the noise in the FLL. Especially in a circuit
like the 5065, where a lot of components don't have the high
PSRR that modern components give us.

I do not have studied the 5065 design in all detail (i.e. I might
be wrong), but the only points where I think it really matters
is the C-field and the Rb lamb supply.
 
> The outline idea currently is:
> 
> 	LM399 self-biased voltage reference
> 	    (See: Linear app-note 42, fig 72)
> 
> 	The ultimate board would use LTZ1000, but I have not quite
> 	convinced myself yet, even if it would be cute to have a
> 	HP5065 which delivered both precise frequency and voltage :-)

I would go for the LTZ1000 if you can. The LM399 has gone out of
fashion for a reason: Modern bandgap references have similar
noise at a fraction of the cost and if you need stability the
LTZ1000 is the better option. E.g., the LTC6655 is only about
a factor two worse in long term performance (8ppm/sqrt(k hour)
vs 20ppm/sqrt(k hour)) and about the same factor in noise voltage.

The LTZ1000 offers both better aging (including temp stability)
and noise performance. Of course an LTZ1000 costs 4 times as
much and that doesn't include the support circuit that you
need for temperature control.


> 	C-field polarity switch.  An experiment.  If nothing else
> 	I will be able to measure the residual magnetic field.

For this to be a valid test, you also need to switch the
polarity of the light, otherwise you excite different states
with slightly different magnetic dependence.

 
> 	Two +20V on-board LM317-style linear regulators, one for
> 	the lamp, one for the rest.  Split for noise reasons and
> 	to be able to play with the lamp voltage/power.

I would go for a lower noise regulator. I would recommend
the TPS7A45xx.. but unfortunately, they are 20V abs max.
A good choice might be the LT3081. Like the other members
of the LT3042 family, it offers quite low noise while still
being able to deliver 1.5A. It's quite a bit more expensive, though.

 
>         One or two pre-regulator current measurement shunts.
> 
> 	DC/DC-brick switchmode -20V supply with brutal filtering.

I would go for one of the inverting µModule regulators from ADI.
They offer quite low noise/EMI while being easy to use. Only
downside: They all come in BGA cases, which makes them hard to
solder by hand.
https://www.analog.com/en/products/power-management/umodule-regulators/umodule-inverting-regulators.html


> The downside of the two linear +20V regulators is that even with a
> heatsink, they will probably get hot-ish if the internal DC bus is
> too much over 24VDC.  For this reason, and because I may simply run
> out of PCB space, I may leave the bridge rectifier out, so it will
> only works with EXT-DC.

My recommendation would be to replace the whole power input
by a modern switched power supply. These are quite low noise
and, if used correctly, also low EMI. It also allows you
to reduce the voltage, due to better regulation, and thus save
on power. Just having an LT8640S as pre-regulator would reduce
the dissipated power quite considerably with having low to
negligible impact on EMI (Silent-Switcher 2 architecture).

But yes, power dissipation will be a problem, you have approximately
0.8W dissipated per difference in voltage. Which means it's already
3W for 24V. While not exporbitantly high, it's already quite
considerable if you have no forced air cooling. If you want
to go that way, I recommend using regulators in a TO-220 case
and something like an Aavid 7021, 7022, 7023, or 5510 heatsink.
The larger ones should allow you to even go up to 30V.

If you have the space to use a switched power supply and to
implement a feedback system such that the output of the
switched supply is kept at 2V above the output of the LDOs,
you'd reduce power dissipation quite a bit and should be able
to get away with D2PAK LDOs and a small heatsink.

			Attila Kinali

-- 
<JaberWorky>	The bad part of Zurich is where the degenerates
                throw DARK chocolate at you.