Post reference filter -- implementation issues?

[essele]

I'm working on a parallel REF102 design, which is generally going well, but as part of my experiments I built three different output options into the board to ultimately figure out the best output method:

1. Straight 10V output direct from the parallel REF102's.
2. A simple buffered output (currently OPA189)
3. A filtered output (as per figure 14 on the REF102 data sheet)

Quite bizarrely I am getting the best results with the buffered output, then the 10V output and the filtered output is the worst with a huge amount of 50Hz noise. All of them show some 50Hz noise, but the filtered one is absolutely awful!

My assumption at this point is that this is a PCB design flaw on my part, either to do with the lengths of the tracks or the ground plane underneath this section (I've never really understood when you should and when you shouldn't!) ... I fear I have built the worlds best radio receiver!

Can anyone give me any pointers?

Attached is the section of the PCB (I can include the schematic, but it's actually really simple to see from the PCB and the component values match the REF102 datasheet (apart from it being OPA189's))

I've also attached a couple of the signal analyser outputs (apologies for the awful photos, I haven't got GPIB setup yet on this), the bad one if the "filtered" output, the other one is the unbuffered 10V output.

EDIT: Just having a quick look at the higher frequencies (2kHz+) I can see the the filter is having some positive effect (noise is around -111dBV on buffered, -114dBV on 10V, and about -120dbV on the filtered output) ... so this does seem like a low frequency problem.

EDIT2: I've just made a small circular "probe" which I connected to my sig-gen at 130Hz and I can induce that signal into the "filter" line really easily,  with some careful positioning almost as much signal as the 50Hz peak, but I can't get it to even show on the 10V and buffered lines. So I really have done something specific on that section.

[Kleinstein]

The filtering circuit is a little odd. For the most part it is a buffer that the tolerant to capacitive load and loaded with a capacitor. Some 2 K and 1 µF gives a cross over between the resistive and capacitive feedback at some 80 Hz. In the cross over region there can be some extra ringing and less filtering  - so the choice of cross over frequency is somewhat questionable  . The 1 µF at the output is already close to the maximum the circuit can tolerate.

The other possible point is the choice of the capacitor as small SMD form factor - which would likely be X7R ceramic. This can pic up mechanical hum.

For filtering I would suggest more like having a capacitor (e.g. 10 µF polyester) to ground at the input of the buffer.
The resistors for averaging could be a little larger than 2 K. Even some 10 K should be low noise compared to the reference own noise.

One could still use the more capacitance tolerant buffer, but with a much smaller resistor at the OPs and a higher cross over. So more like 10-100 nF (film or C0G) for C1,C2 and R1= 50-100 Ohms. This would provide some filtering against the really high frequency spikes from the AZ OP.

[Andreas]

Can anyone give me any pointers?

I can include the schematic, but it's actually really simple to see from the PCB and the component values match the REF102 datasheet

Hello,

I am doing noise measurements always within a metal cookies box (tinned steel) with battery powered DUT.
But even then I have to maintain minimum 0.5 m to the next transformer.
(E.g. the one which is shielded within the HP34401A).

If you don't include the schematic nor a link to the datasheet you do not want that (at least) I do further analyses.

with best regards

Andreas

[essele]

Thanks Kleinstein -- I think you are onto something.

Thanks Andreas -- I've attached the relevant part of my schematic (although it is just a copy of the circuit in the REF102 data sheet.)

Ref102 data sheet is here: https://www.ti.com/lit/ds/symlink/ref102.pdf?ts=1590423621326

So I've removed C20 (the output cap to ground) and the problem still exists.

If I remove C19 the problems goes away ... so this does look to be something to do with the interference being fed back, or potentially an issue with the choice of cap as you suggest, I have tried a different one, but still a X7R ceramic and it shows the same problem. I will try to find something of a different value to experiment.

Thanks,

Lee.

[Andreas]

If I remove C19 the problems goes away ...

Hello,

but if you remove C19 the output is likely to oscillate ...

I do not think that C19 is the root cause.
But R32 is unusually high for a buffer.
I am using usually something between 22R and 100R (depending on OP-Amp)
The actual filtering is then done on the positive input of the OP-Amp.

On the other side: After my experiences the OPA189 is not the best option for a output buffer.
Depending on input impedance there is a large increase of offset:

https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2805686/#msg2805686
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2819014/#msg2819014

So in the mean time I prefer the ADA4522-1

with best regards

Andreas

[kleiner Rainer]

Hi essele,

have a look at https://pearl-hifi.com/06_Lit_Archive/02_PEARL_Arch/Vol_16/Sec_53/Burr_Brown_App_Notes/1%20-%20Burr-Brown_App_Note_Compilation.pdf , in this case AB-003A. it looks like the resistor at the output is too big. I would retest the circuit with the recommended values. Maybe the OpAmp is saturating due to the voltage drop across the output resistor and then oscillating?

Rainer

[essele]

Thanks to all ... problem is solved.

I was actually aware of the app note, but decided to go with the REF102 datasheet values instead, however switching to the app-note values (50R and 0.05u) has solved the problem completely.

I have also ordered a couple of ADA4522-1's to see how they compare.

Thanks,

Lee.