[volt-nuts] Making a Reference IC

[Charles Steinmetz]

I wrote (regarding replacing the op-amp in a Fluke 731A or B voltage standard):

> > If you replace the op-amp with a modern precision part (I suggest the
> > OPA277 -- others may try to talk you into an auto-zero op-amp, but I
> > think that would be a mistake in this application),

Attila wrote:

>Why would this be a mistake?

Let me count the ways....

First, the ground rules:  My suggestion was to modify the 731 (we now 
know the OP has a 731B) with as little invasion as possible.  I 
suggested (i) replacing the reference with an LM399 and changing the 
gain-setting voltage divider (R2 or R3) to obtain 10v with the new 
reference (only a small change would be necessary), (ii) if desired, 
replacing the op-amp, and (iii) replacing all of the divider and 
zener bias resistors with lower-tempco units (moot, now that we know 
the unit is a 731B, which has low-tempco WW resistors).  I do not 
recall how much current the selected resistors, R5 and R6, deliver to 
the existing reference; the LM399 seems to be most stable over the 
long term when operated at 1.5 to 1.8mA, so R5/R6 should arguably be 
changed (if necessary) to get the current into that range (about 
2k).  Any replacement resistors should have tempcos at least as good 
as the original WW resistors, but anything better would be wasted on 
a circuit that sits out in the open like the innards of the 731B.

My notion was to use an op-amp that will drop directly into the LM308 
position, with only a few component changes on the circuit board.  If 
someone wanted to do something more elaborate, there may be other 
possibilities, but even then I do not believe an AZ op-amp would be a 
good choice for the reasons given below.   [since Fluke used an LM308 
in the TO-5 package, which is not a current package for either of the 
op-amps I recommend, in reality a small adapter board would need to 
be made (or the op-amp would need to be mounted on stilts -- ugh, PC 
meets dead-bug)].

I have now recommended two candidate op-amps -- the OPA277 and the 
LT1012.  I have not run the numbers to see which is better in this 
application, but I suspect the LT1012 is.  The LT1012 was once made 
in the TO-5 package -- if one can be found today, that would decide 
the matter for me, for two reasons: (1) it fits with no surgery, and 
(2) the hermetic package resists drift due to atmospheric humidity 
and pressure.  The LT1012 also has an overcompensation pin, like the 
LM308, which is handy for making the particular circuit used in the 
731B stable.  So, I'm recommending the LT1012 in preference to the 
OPA277, even if it is not in the hermetic package.

Finally, note that the LM308 is a respectable performer even today, 
and with everything out in the open on the PC card it is quite 
possible that other errors may completely swamp any improvement the 
LT1012 might promise.

So let's start counting:

1.   Maximum power supply voltage.  The 731B uses an arrangement in 
which it is powered by the regulated 10v supply.  Precious few AZ 
op-amps can handle a power supply of more than 5 or 6 volts, and the 
ones that will are not the best choices for other reasons.  Even the 
reference output (not quite 7v, typically), which is connected to the 
op-amp's noninverting input, is higher than the vast majority of AZ 
op-amps' maximum supply voltages.  So, even changing the topology of 
the amplifier would not help.

2.   Switching noise.  All AZ amps suffer to a greater or 
not-so-lesser degree from switching noise.  Some of them allow the 
designer do a halfway decent job of filtering the output -- but all 
of them put all sorts of hash and garbage onto their supply 
lines.  As noted above, the supply is connected directly to the 
regulated output in the 731B, so there is no good way to clean it 
up.  (Same for the AZ amp's output noise in the 731B circuit.)

3.   Input noise current.  AZ amplifiers have little or no 1/f noise, 
because it is cancelled by the AZ action.  However, they have 
substantially more input bias (leakage) current than regular CMOS 
op-amps and, consequently, a higher input noise current 
density.  This is a dirty little secret of AZ amplifiers -- most 
manufacturers calculate the input noise current density as the shot 
noise of the input current (if they even list a noise current spec -- 
many do not).  However, the actual measured input noise current 
density is frequently as much as 100x higher than this calculated 
value.  So, even ignoring switching noise, many AZ amplifiers are not 
even as quiet in real-world circuits as run-of-the-mill JFET 
op-amps.  Since both of the op-amp input nodes are moderately 
high-impedance, input current noise will be a factor in the noise analysis.

I'm sure I've forgotten a few, but those are some of the high points.

Best regards,

Charles