Shunt Reference Problems

[legg]

On Wed, 19 Feb 2020 10:24:36 +0000, Michael Kellett <mk@mkesc.co.uk>
wrote:

On 19/02/2020 00:59, legg wrote:
On Mon, 17 Feb 2020 11:19:43 +0000, Michael Kellett <mk@mkesc.co.uk
wrote:

On 17/02/2020 11:04, speff wrote:
On Sunday, 16 February 2020 09:51:24 UTC-5, Michael Kellett wrote:
I’m working on a power supply design. It has 4 independent and isolated
channels which use Microchip LM4040 2.5V references. It also has opto
isolated readback for each channel which uses Analog ADR530B references
and a tiny ST micro.

The ADR530B Temp Coefficient is specified as 15ppm typical and 40ppm
max. There is a note saying “Guaranteed by design, but not production
tested “ .

When I ran it though a temperature cycle, I was a bit disappointed to
get the results in psu_plot_1.jpg.

https://www.dropbox.com/sh/8f6b34u58gg400r/AABcRqFMQ-Q0JAVOAzFRcxJra?dl=0

For the temperature tests the power supply runs off the mains, its on
board processor reports results to a PC outside the test chamber. The 4
PSU outputs are also monitored by a Keysight 34972A outside the chamber.
I’m not bothered by the actual output voltages – these are dependent on
the cheap and cheerful LM4040s and within spec (although I don’t like
the steps on the blue 15V channel).

The 15 and 5 V measurement errors plots come from comparing the measured
(by 34972A) outputs with the self-reported output based on the ADR530B
references and the temperature drift is much higher than 40ppm.

My first thought was that external parts, probably voltage divider
resistors or maybe even the tiny processor ADC was causing the problem,
so I replaced the resistors with expensive ones and added another 4 x
34972A channels to monitor the ADR530Bs directly.

This got me psu_plot_refs_ltcrs_1.jpg.

There’s still some bad stuff going on with the blue 15V channel – but I
don’t care about that yet – concentrate on the last plot – that’s the 4
ADR530B chips monitored directly, the temperature coefficients over the
5- 60C range work out at:

65, 56, 53, 53 ppm/C.

Well, it could still be my ADC or some other thing I’ve missed in my
design, so I built a little test board with 2 x Microchip LM4040 and 2 x
Analog ADR530B, there’s a picture of it set up in the temperature cycler.

Each ref chip is connected to a common 10V supply by a 2k2 resistor.
Each ref chip is shunted with a 470nF cap.
The results are shown in ref_chips_1.jpg

The temperature coefficients over the 5- 60C range work out at:

-27, 46, 50, 54 ppm/C.

The Microchip parts are within spec but the one with the blue trace has
a seriously odd behaviour (this is NOT the same channel on the 34972A as
the previous “bumpy” LM4040).

The real problem is the ADR530B parts – perfectly linear and smooth
tracking of temperature, and out of spec.

I’ll be following this up with AD (of course) but I’d be interested in
any comments.

MK

In your tests to attempt to isolate the reference- did you have nothing else in the environmental chamber? If the power supply was there, as it appeared to be in the photo, was it operating during the test?

Best regards,
Spehro Pefhany

Good point, thanks - it was there but not plugged in - if you look very
carefully you can see the unnconected IEC mains connector at the bottom
left (and it was unconnected during the test - I checked the testfile
and all the PSU signals are at zero).


I've ordered some similar ADI parts (ADR525B) and I'll add those to the
test board, and maybe a few others - the 34972A has lots of channels !

MK

Watch the 34972A; You need slow/long measurement period and should
avoid large voltage differences between subsequent measurements.

It pumps current into the measurement nodes at the time of switching
and may also introduce CM loops, though 'isolated' from the internal
meter.

RL

It's set up sensibly enough and the data doesn't look very symptomatic
of the kind of problems you suggest. At some stage I can try using a DMM
directly connected but that will limit me to testing less than 1 part
per day !

MK

As a precaution, I always isolate the the 34972A measurement lines
with >=100R in series with each wire. You can add that into error,
if significant. It should be constant, if you're measuring deltas.

RL