Finding and mitigating EMI in low ppm volt measurements

[lars]

As I follow the threads with calibration clubs – round robins around the world and also the KX and MX boards I see that problems with glitches and unintended offsets are evident. Probably a lot of this can be traced back to EMI in one way or another. Some of the problems are due to other environment problems and some may be due to oscillations due to loading. So a lot of reasons you can´t trust your result. But if we go back to EMI, how can I find it? Glitches are easy to see but an offset (or bias) due to EMI is more difficult.

I have tried to use some simple tests to see if I have an unintended bias.
- Clamp-on ferrite beads on in and output wires. By adding and removing I have tried to see a difference. Probably works best for high frequency EMI
- Adding a parallel capacitor on the output. Maybe mostly helps start an oscillation??
- Adding a series resistor of 1-10k on the output (needs a high impedance DMM). Probably helps if your DMM loads your (inferior) design so your reference oscillates?
- Moving in and output wires including separating the plus and minus on the output. Good to detect 50/60Hz??
- Battery power versus mains power. Breaks loops that can pick up whatever frequencies.
Of course, the effect very much depends on the frequency of the EMI. Is it 50/60Hz or GHz?

I really hope to get more tips and maybe examples.

Lars

[Andreas]

Hello Lars,

measures to reduce EMI:
- Ferrites or inductors help only for a relative narrow frequency band.
  So often you do not catch the frequency of interest.

- Adding a capacitor to the output helps usually for a wider frequency range
   But your cirquit has to be designed for capacitive loads.
   (see isolating cirquits for OP-Amps in the application notes).

- A 1-10k series resistor is usually a better idea than a inductor.
  At very high frequencies the ~0.25 pF parasitic capacity dominates.
  In DMMs you often see ladders of serial connected resistors (not only to improve max. working voltage)

- of course a Pi-Filter either with inductors or resistors and (different) capacitor values can be used to create a wideband filter.

Some methods to detect EMI:
- if your cirquit is "hand sensitive" (use your "RF-thumb" on different parts of the cirquit like battery, connection lines etc.)
- you get different results depending on how your measurement cord are placed
- you have changes in reading when you put all your gear on a metal sheet (oops where is that sidewall from my PC)
- track peak-peak noise of your readings. With a mobile phone in 2 m distance of my measurement setup the noise increased to 30uVpp.
  The averaged value changed only by 1uV but the peak-peak noise gave a indicator to discard this measurement.
- To detect the most sensitive pins in your cirquit it may also help to use a steel needle to contact every IC-pin with your "RF-thumb"
   while reading the output.
  (you may want to repeat the experiment under different conditions like battery power / mains power).

with best regards

Andreas

[dacman]

Some places use a screen room.  Coaxial cable or (shielded) twisted pair can be used to reduce EMI.  (Twisted pair works because if the same signal is induced into both conductors, it will cancel out due to common mode, in theory.)  Something that may help is to turn off nearby high current devices.  I've seen overhead lighting (non-shielded florescent) and computer monitors (CRTs) cause EMI interference.

[floobydust]

EMI is difficult to mitigate and blanket rules "put this there" "don't put that there" doesn't always guarantee quiet.

Mains is very noisy and EMI filters there can make it worse as they all (Y-caps) assume you have a good low impedance ground. This is not the case after many metres of 14/3 electrical cable. I use medical-grade mains EMI filters as they contain no Y-caps. One attenuates a good portion of mains garbage, leaving only the local instrument-generated EMI.

The quietest room I've used (for uV neural preamps) was all steel shielded, DC lighting, electrically isolated from HVAC ducts, power through with isolation transformer and dedicated (lifted neutral) earth-ground. Very quiet but turning on a PC in the room kind of defeated the purpose. A PC makes a fair bit of EMI. CCFL LCD monitors are terrible.

Very tough to track down blips and shifts:

Elevators have VFD's and can make a lot EMI when moving, as the long VFD cables run to the top floor where the motor is located. Took me a year to find that one.
Wi-fi routers, cell phones, tablets, any RF source- these are all polluters.

I connect an audio amp and listen with headphones to the noise. Any periodic stuff, demodulated beeps, clicks, hum - you can hear. With a scope trace it's harder to really know what just happened to cause an offset shift.
My main tool is a junker 900MHz cordless phone, which seems to get into every circuit/cable imaginable. I use that for a quick susceptibility test.
I also take a pocket AM radio and scan a room to find mid-band sources of EMI.

[Echo88]

Thanks floobydust for the suggestions. Could you tell us which medical grade EMI-filter u use? Ive also thought about installing LED-DC-lighting, instead of my normal IKEA-LED-lamps. Maybe a 19"-rack which houses the measurement/voltage-reference-stuff would attentuate a good amount of EMI?

[martinr33]

I live in an electrically noisy environment. Some of it is my doing - always-on modern electronics with SMPSUs; X10, Insteon, Zigbee, WiFi, and others; and some of it is because I can see big transmission towers on a hill not far away. Plus, my bench gets power through a tangled web of extension cords. I think I share power with a refrigerator; several power bricks; and sundry electrical and electronic devices.
I know that things are noisy because of my limited WiFi range, my garage door opener only works right outside the garage, for example.

In my noisy garage, I have an 8.5 digit Advantest meter hooked to a Fluke 732a reference using coax cable with grounded shields (and separate core conductor for measurement ground and thanks to Bill for showing me these tricks). I am seeing the meter report SD of 300nV across 100 readings. The best one-off I have seen is 130nV. The max-min is 1uV in both cases.

I m not twisting the cables to counter inductive noise, but I do have them tied together to minimize loop area.

It took a few days for the meter to settle down to this level of noise.

Key here is that everything is low noise to start with, with careful grounding and isolated, complete shielding. The shielding envelops almost the entire measurement chain, and has only one ground connection at the 732a. I did not try it with the 732a unplugged from wall power.  I am using copper banana jacks, with silver/gold plating.



 So I offer this as a demonstration that - with good shielding - environment is not so critical for precision DC measurements.

[floobydust]

With a spectrum analyzer, mains in an office building I see 66/132kHz (PC power supplies) and 100MHz (FM radio station) at high levels.
I use the "B" version of Schaffer EMI filters("medical" no Y-capacitors) in the FN2020. Filter attenuation is about 45dB at 1MHz and 20dB at 100kHz.
If on a real budget you can desolder a vanilla EMI filter and disconnect Y-caps. But you need a blow torch to heat up the can.

A piece of sheet metal on the bench top, 1-2mm thick steel gives shielding. I use 6mm thick aluminium slab in the EMC lab and it is ok, not perfect but still shields things on the bench.
I have an ESD mat overtop of it, so things don't always short dramatically to the grounded plate.

[Echo88]

Would a isolation transformer (230V prim, 230V sec) with a shielding layer, followed by a mains filter be even better than the mains filter alone, regarding mains noise? Of course earth would be connected on the secondary of the transformer. I cant find any documents about the achievable attenuation curve of a shielded tranformer as an example.

[Andreas]

Hello,

the normal mains filter is usually used to filter the conducted EMI out of the device.
(which is measured from 150 kHz to 30 MHz)
So those filters are optimized for this frequency range.

A shielding layer in a transformer is for the frequency range below the mains filter.
(because of the inductivity it cannot go into the MHz range).

Of course the mains filter has to be nearer to the mains (as feed through capacitors) than on the other side of the transformer.
(Otherwise the higher frequencies are already within the metal housing and are distributed within your cirquit).

with best regards

Andreas

[Andreas]

Hello,

by the way: the earth ground is mainly for security reasons. Not for EMI.
For EMI of higher frequencies it plays no role because it is a long inductive wire.

Ok for lower frequencies in the low kHz range there is some effect.

with best regards

Andreas

[Echo88]

So a mains-filter-transformer-DUT-configuration would attenuate mains conducted noise over a broader frequency range. But i guess i would have to measure the exact noise distribution (LED-lighting, PC, elevator, WLAN...) via a Spectrum Analyzer and maybe a Line Impedance Stabilisation Network. Since i dont own either i could simply test it with a very noise sensitive setup e.g. with my nV-measurement-box and flip the light switches in my flat and see if it still responds despite the filtering/shielding-effort.

[Moon Winx]

There is a great book on this subject: "Noise Reduction Techniques in Electronic Systems" by Henry Ott. There are also some books by Ralph Morrison on the subject that may be useful to some but I didn't like the writing style... I found it disjointed and overly dense but YMMV.

[Awesome14]

Use the shortest possible shielded leads. Ground the lead shield on one side only. These are the only ones I'm aware of: https://www.digikey.com/product-detail/en/pomona-electronics/1167-60/501-2099-ND/736140

Otherwise, twist the leads together. Also, when the meter is connected, read the ACV component.

[T3sl4co1l]

Use the shortest possible shielded leads. Ground the lead shield on one side only

Both sides.

But the meaning of "ground" matters.  "Ground" being the local circuit negative.  It's contextual, not absolute.  Don't go out of your way to tack a wire from galvanic ground onto the far end -- that would be silly!

Using a shielded cable, with the shield grounded (to the circuit's local ground and shield, whatever that may be), creates a Faraday cage around the circuits and signals.  Don't think in terms of galvanic grounding, think in terms of wrapping the entire circuit with a continuous shield.

Tim

[BradC]

But the meaning of "ground" matters.  "Ground" being the local circuit negative.  It's contextual, not absolute.  Don't go out of your way to tack a wire from galvanic ground onto the far end -- that would be silly!

Ok, so I have a meter. I want to measure a voltage reference so I put the reference inside a cake tin. I have a couple of feed throughs and I'm going to use coaxial leads to the meter.

- Do I connect the coax shield to the tin and the voltage reference/meter -ve terminal?
- In that case can I use a single coax with the shield acting serving double duty as the -ve and center conductor the +ve?

In the same vein, taking a recent mistake of mine I want to measure an 800 Meg resistor. I put the resistor in a cake tin with 4 feed throughs and kelvin connect it.

- Which meter terminal do I connect the shields and tin to?

Based on what I've read here I'd *assume* it's force voltage -ve.