Randy wrote:

I have used LTC1043s in the voltage-multiplier configuration, and
based on that experience I believe you will find there are too many
surprises hiding there to reach your error budget.  First, there are
losses in the conversion -- IME, even when driving nothing but the
non-inverting input of an LTC1050 chopper-stabilized opamp used as a
follower, the output voltage of an LTC1043 doubler is quite a bit
(50uV or more) less than [Vin * 2].  Second, the output of the 1043 +
buffer is about 100x noisier than the output of the same reference
followed by a non-inverting gain-of-two LTC1050 (on the order of
100uVp-p for the 1043 + buffer, about 1uVp-p for the non-inverting
gain-of-two amplifier in a 10Hz bandwidth, IME).  Also, the 1043
noise is not symmetrical, so different DC meters may give readings
that differ by 50uV from one another.

If you try the LTC1043, I'll be interested to see what you find.

Best regards,

Charles

Charles,

For the design I am contemplating, the accuracy of the X2 in not important,
only the stability.  For example, if the output of the X2 is 50uV low, I
don't care so long as its always 50 uV low +/- a few ppm over time and
temperature.  The noise could be filtered but its must also be stable.
Based on what I am hearing, the switched capacitor multiplier doesn't look
promising.

Randy

On Tue, Jul 22, 2014 at 5:40 PM, Charles Steinmetz csteinmetz@yandex.com
wrote:

Hello,

the 50uV loss is plausible to me.

But for the 100uVpp I have to ask for the measurement conditions.
Is the source (reference) connected via (long cable) and supplied with
another mains line
or on the same pcb with the same power supply (or battery supplied).
Was the pcb cleaned before measurement to keep leakage low?
How much noise did the reference generate? 1uV output measurement is
quite low with
typical 0.6ppm/V of a buried zener multiplied by 2 and a amplifier with
1.6uVpp (referred to input).

As every CMOS input is susceptible against noise (getting rectified over
the input protection diodes)
you can easily shift the input voltage by an offset due to RF noise.
For LTC1043 inputs which are connected via external line I use a 1nF filter
capacitor against ground to keep the influence of the RF noise low.

With best regards

Andreas

Am 23.07.2014 02:40, schrieb Charles Steinmetz:

Andreas wrote:

No

Yes.  (And I have extensive design experience with ultra-low noise
power supplies, decoupling, RFI prevention, and low-noise grounding.)

Yes, it was freshly built and cleaned upon completion (vapor-phase),
although that really has no bearing on commutation noise (charge
injection) internal to the switched capacitor circuit, which is where
all of the cyclical noise is generated.  Can it be filtered
out?  Perhaps, but not easily without raising some other noise or
drift problem.

I was using the ADR4520 (1uVp-p noise in the 0.1-10Hz band), with
post-filtering.  When I said "about 1uVpp" output noise, I was
thinking order of magnitude for purposes of comparison to 100uV, not
precise value (which was more like 2-3uVpp).

Best regards,

Charles