EMI-Measurements of a Volt-Nut

[MiDi]

My estimation is: the isolation through the OP-Amp gets completely lost on the ADA when connecting a output capacitor across X4+X5. So the LTZ itself is affected.

The schematics shows the output buffer (R23 is not populated).
I think we would have a better behaviour when we would split R22 (10K) into two resistors and place one before the non-inverting input.

Thanks for the results for ADA-4522, very interesting 
Your estimation of loosing isolation would mean the input diodes of ADA start clamping?
Why do you think splitting R22 would improve the results?
If the LTZ1000 is loaded "down" this way, a split resistor would help for LTZ1000, but I do not see how the buffered output could be less affected?

[Andreas]

Hello,

my assumption is/was that the remaining pattern is that what remains from the LTZ.
But in the meantime I have found a Ground connection error in the layout.
The Buffer was added in a revision of the PCB.
So I thought it was a good idea to route the power supply for the buffer directly from the voltage regulator.
(instead first routing to the output buffer and from there routing it to the Star ground from the LTZ1000).

So when I add a capacitor across the outputs (output gnd) it has a very looong way to the ground pin of the output buffer (power gnd).
I fear I have to check what happens when placing the capacitor from the positive output to the GND pin of the output buffer.

The only remaining question is: why is the OPA189 not sensitive to this wiring where all other OP-Amps seem to have some problems.

with best regards

Andreas

[Andreas]

Hello,

first measurement of ADA4522-1 with the 4.7nF capacitor connected to Power Ground (PGnd) instead across the outputs.
-> much better result. -13 ppm @ 1.78Vpp.

with best regards

Andreas

[3roomlab]

i tried a few flavors of models from TDK and bourns
in short, looks like more pi filter stages is better than large C bank, even with small x ~ xx uH L value, large xxx uH values provides even more attenuation
C0G doesnt seem to help with EMI like X7R does, esp if we try to assume -140dB as the point where noise start to interfere with 1ppm of something?
if the simulation scales correctly, then 1 would assume all single C points of attenuations with large enough uF could only affect down to 100ppm to 10ppm scales
so my assumption is ~ very bad noise can be improved to at best 10ppm? by single LC?
then i try some theoreticals, capacitor with micro ohm esr which does not exist, it can attack down -140 to -160. but it cannot do broadband, only peaks

but multi stage LC pi seem to easily overcome -200 -300 db etc, this reminds me of the LISN thread
https://www.eevblog.com/forum/projects/5uh-lisn-for-spectrum-analyzer-emcemi-work/?all
and this
https://www.eevblog.com/forum/rf-microwave/50uh-and-250uh-inductor-design-for-lisn/?all

so in the 2nd pic i ran a rubbish mix of components and its under -160, i guess then 1LC is never enough, maybe 2pi?

[branadic]

Thanks Andreas,

could the ground issue also be a problem for LTC2057?

-branadic-

[Andreas]

but multi stage LC pi seem to easily overcome -200 -300 db etc, this reminds me of the LISN thread

Hello,

I fear only in theory. Practically every piece of a wire is a antenna.
So for higher dampenings than 20-40 dB you will need a shielded housing for every stage of the filter.

Of course you are right a X7R  or a lossy capacitor is more wide band than a high quality capacitor.
And also some ferrites act more as a lossy resistor for RF than as a inductor.

could the ground issue also be a problem for LTC2057?

When I look at the difference before and after it is mostly the 60MHz peak which is gone.
If this is right then the LTC2057 should also profit from correct connection of the output capacitor.

Even the OPA189 should have a little improvement near 60 MHz. But I guess the 20 MHz peak will not change much.

We will see.

with best regards

Andreas

[Andreas]

Hello,

one side effect that I mentioned with my experiments.

The OPA189 always gives a 12-18uV higher output voltage than the other devices. (LTC2057/ADA4522)

This is a bit strange since the offset is in the 3-4uV range typical and also the bias current by the 10K resistor should give a lower offset than 2uV additional.

with best regards

Andreas

[TiN]

Andreas
I have ordered ADA4522 (-2, but PCN affects all flavors anyway) from DigiKey few days ago. Will let you know what arrives once I get package next week.
It was intentional choice why I selected ADA4522 for output stages on my FX and QVR reference designs. 

[Andreas]

Hello,

I also ordered some parts from DigiKey (ADA4522-1ARZ-ND) packaged in tube but they ares still the old version.
(perhaps I should have ordered the cut tape version to get newer parts?)

But I fear the stock of DigiKey has to flat down before they (hopefully) get the new version.

But obviously this will take some time: (CT = cut tape, ST = tube)

Digikey ADA4522-1

13.11.2019
CT: 1395
ST: 449

24.11.2019
CT: 1075
ST: 277

29.11.2019
CT: 1062
ST: 277

with best regards

Andreas

[TiN]

How you decipher AD's/LTC datecodes?

[Andreas]

Hello,

if you look at the picture above
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2780364/#msg2780364

the middle line A#643 looks like a date code = 2016 week 43

The recently bought ADA4522-1 are datecode 751 = 2017 week 51

with best regards

Andreas

[MiDi]

The OPA189 always gives a 12-18uV higher output voltage than the other devices. (LTC2057/ADA4522)

This is a bit strange since the offset is in the 3-4uV range typical and also the bias current by the 10K resistor should give a lower offset than 2uV additional.

A bit strange seems understated.
In such a configuration it is way more than expected maximum for all obvious contributors to offset in sum.

Any idea what could be the cause?
Is the offset independent from value of output cap?

I assume there were 2 different OPA189 in the 2 LTZ?

[rhb]

I quit reading unread posts for a while because EEVblog was eating too much time.  I've only skimmed this thread, so please forgive me if I restate something already said.

I have 2 LED lamps in a 4 ft fluorescent fixture.  This produced a tremendous amount if EMI until I added a large Corcom filter and 1/8th" hardware cloth screen.

I have also put HW cloth over the openings of my Z400 workstations.  But still more to do.

Because of the staggering amount of EMI produced by my Instek MSO2024EA SMPS, I made a custom power cable for all my instruments which is fully shielded with MX and EMT all the way to the back of the instruments.   I posted about the project some time ago with photos.  I also bought, but have not installed a 2.5 KVA isolation transformer.

The Instek GDS-2000E line can be hacked to enable the spectrum analyzer app for the MDO-2000E variant.  This is a very useful tool and far more sensitive than my 8560A.  Above the 200 MHz maximum BW of the GDS-2000E, an SDR is the cheapest solution.  A SDRplay RSP2 will give a lot more dynamic range than and RTL-SDR, but also at 8x the price.

Most of the noise of concern is below 1 MHz.  There is a lot of it,  from a variety of sources.

Even without a good FFT on a scope, a small wire loop from tip to ground on a scope probe is very useful for locating nearby EMI sources.  A proper H field probe of large dimension is needed to locate more distant sources, but easy to make from solid shield RG402.

I went on a rather extreme TEA binge and now have three  5ft stacks of T&M kit on mover's dollies in my dining room for lack of space to set it up.  My current space is part of a 7' x 10' closet.   In order to have the displays  for SA, VNA, scopes, etc at a comfortable level I need 16' of bench, so I am building a 12' x 16' room in my shop which will be lined on all faces with 28-30 gauge galvanized steel sheet soldered at the seams.  I am told by a retired Tempest engineer that should provide about 120 dB of shielding.  The room will have its own HVAC zone on a 3 zone mini-split system.  Counting the cost of the extra HVAC zone, I expect that will run about $3K.  The steel sheet alone is about $800 for a 16' x 12' x 8' room.

A less costly approach and more easily moved solution would be to build desktop size steel box to hold equipment and DUT and extract the data via an optoisolated link.  Galvanized wire screen soldered to the steel sheet will allow airflow while blocking EMI.

To get from the 60 Mhz limit of an F***Tech to 120 MHz try one of these:

https://www.ebay.com/itm/NEW-1-200MHz-radio-frequency-multiplier-module/112082979199

I just saw them a couple of days ago and bought a couple.  Above that an ADF4351 module from ebay will serve to 4.4 GHz with harmonics much higher.

This book:

https://www.amazon.com/Electromagnetic-Compatibility-Engineering-Henry-Ott/dp/0470189304

has a great deal of useful information. however, the author writes poorly in passive voice, so it is difficult reading.

Finally, if you live near a high power broadcast tower, move.  I lived about a mile from a 250 KW FM tower and it drove me nuts.  Even a few inches of wire picked up strong EMI. I built a 120 dB DC audio amplifier.  With a speaker hooked to the output I could listen to the FM station merely by holding my finger 1/2" from the input. Driving a 3" speaker it was loud enough to hear in the next room.

That experience left me with a very strong desire for a quiet room in which to work.

Have Fun!
Reg

[Andreas]

A bit strange seems understated.
In such a configuration it is way more than expected maximum for all obvious contributors to offset in sum.

The values are "typical".

Any idea what could be the cause?
Is the offset independent from value of output cap?

No real idea up to now.
The value seems not to influence the offset. But the capacitor position (OGnd, PGnd) seems to have a influence.
(have to sum up the values).
I have checked for oscillations: but found none on my floating scope.

I assume there were 2 different OPA189 in the 2 LTZ?

yes. 2 different OPA189

The Instek GDS-2000E line can be hacked to enable the spectrum analyzer app for the MDO-2000E variant.  This is a very useful tool and far more sensitive than my 8560A.

Interesting.
But I cannot find any noise specifications for the scope nor any jitter specs.
Or is the 50 ppm/ms the jitter spec = 50 ns which would be rather high for a FFT.

Finally, if you live near a high power broadcast tower, move.
That experience left me with a very strong desire for a quiet room in which to work.

Yes, distance is the easiest way to avoid influence from EMI.
(and at low frequencies like 50 Hz the cheapest option.)

with best regards

Andreas

[Andreas]

Hello,

although I am doing much measurements,  I see not much progress in the "right" direction. (see attached image).

After having reached a good result when connecting a capacitor to PGnd on LTZ9 with the ADA4522 and even 100nF + 100nF (one to Output Gnd = OGnd, and one to Power Gnd = PGnd) did not show much difference.
Column D is a normalized value which makes the drift independant of the actual EMI-Level (with the square root of voltage formula).

My Target would be to get the influence down on a 5Vss level to below 1 ppm.

after reaching  ~100 ppm in the LTZ9 and OPA189 was the "winner" in the OGnd blocking scheme.
I thought that it would be a good idea to exchange the LTC2057 in the LTZ8 with a OPA189 to see what happens when connecting to PGnd.
But obviously in LTZ8 configuration (where I have all EMI capacitors populated so that the Output buffer has a 100nF at the positive input against OGnd) the EMI hardened device does not give any improvement.

So my next planned steps are to add the EMI capacitors on LTZ9 step by step
(moving away from datasheet cirquit) to see where something goes wrong.
I fear I still have a long journey until the final optimized cirquit can be drawn.

with best regards

Andreas

[Andreas]

Hello,

Is the offset independent from value of output cap?

The value seems not to influence the offset. But the capacitor position (OGnd, PGnd) seems to have a influence.
(have to sum up the values).

After reviewing my measurements the statement for the OPA189 here is:
the value of the capacitor (4.7 nF or 100nF) makes no change. (at least in Output Ground position).

For the position of the cap.
Output Ground gives the lowest Offset drift.
Followed by capacitors in both position.
Highest drift is performed with capacitor from output to power ground.

with best regards

Andreas

[TiN]

I've got ADA4522-2 today from DK.
Attached photo.

[Andreas]

Hmm,

840 -> week 40 in 2018 from the date code.

so still the old version before the PCN

with best regards

Andreas

[notfaded1]

What do you think the change might be with the product change?  Has no one gotten a new one yet?  You are talking about the die revision at end of 2018 correct?

Bill

[Andreas]

https://www.analog.com/media/en/PCN/ADI_PCN_18_0171_Rev_A_Form.pdf

According to the PCN the "new" devices will be from datecode 22-Aug-2019 on.

The change is described within the PCN.
two metal layers of the design are changed.
the new die will not be delivered before 22-Aug this year.

with best regards

Andreas
 

[3roomlab]

My Target would be to get the influence down on a 5Vss level to below 1 ppm.

http://www.giangrandi.ch/electronics/anttool/tx-field.shtml
on this website, it has a conversion which spits out V/m field strength

this is another site with V/m numbers
https://www.emf-portal.org/en/emf-source/110
somewhere in the middle is a line stating some measurement of 241V/m when 10cm from a tube

using the giangrandi "tool"
a TX of 1W @ 0.0001km (10cm) 0dBi is approx 54V/m
from the 732A pdf, it states it can go out of spec @ more than 0.18V/m, which means any WIFI 25mW within 10m
0.18V/m according to the giangrandi "tool" is 10uW from 10cm.

edit : im curious as to how much power is injected by the 5V, since the voltage is fixed, the way to know how much EMC enters is by current? do EMC testing also measure current?
by treating the LTZ circuit like an antenna, the amount of EMC (RF power) it could impede going in is purely by the right RF chokes? by having caps it allows shorts, but i think in RF where a cap "shorts", it could be creating a resonance. this scenario could be similar to the input front end of the 34401a vs K2000. i think the crucial part are the L, a number of chokes which makes nodes high in impedance to RF like the string of chokes in LISN
maybe some RF nut can shed some more light? as im not very RF inclined. i assumed many many many things

[Andreas]

The OPA189 always gives a 12-18uV higher output voltage than the other devices. (LTC2057/ADA4522)

Hello,

today I reduced the 10K feedback resistor (R22) by paralleling a 1K resistor. (giving 909 Ohms total feedback).
Cirquit see here:
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2784834/#msg2784834

Result of the 9.1K resistor change was a 21.1 uV reduction of the offset voltage.

So the OPA189 seems to have a effective input bias current of 2.3 nA
which is much higher than the datasheet value (with a weird measurement condition of 500pF and 100k).
For me the OPA189 seems to be only suitable for source impedances below 1K at the inputs.

with best regards

Andreas

[Andreas]

edit : im curious as to how much power is injected by the 5V, since the voltage is fixed, the way to know how much EMC enters is by current? do EMC testing also measure current?

The amount of current which enters is dependant on "parasitic capacities".
So it also depends on the distance to the next metal plane.
That makes these measurements a bit difficult for the repeatability.

The day before yesterday I forgot to remove the charger (switchmode supply) before starting the measurement.
The result was much worse than only with the battery powered voltage reference.
(more parasitic EMI current entering through the device).

with best regards

Andreas

[Andreas]

Just a little summary what I did in the mean time.

As preparation for further hardening measures I added C8 + R19 from my cirquit diagram to the data sheet population of LTZ9.
Without these components I cannot add C9 to the cirquit (as this would lead to oscillations).

The comparison of the measurement from 17.10.2019 and 02.12.2019 shows that adding C8+R19 has no visible impact on EMI behaviour at the unbuffered output.

So I increased power level and added C9. (03.12.2019 + 04.12.2019)

Interestingly the deviation at 60 MHz remained at -88 ppm when adding C9. Whereas the 20 MHz deviation is now suppressed.
So I assume that the 20MHz deviation belongs to the LTZ part of the cirquit and the 60 MHz deviation to the output buffer.

But a comparison before and after adding C9 (01.12.2019 + 06.12.2019) on the buffered output shows that for the output buffer C9 increases the EMI sensitivity.
The sensitivity is even worse (and more wideband) when the switchmode charger is attached to the LTZ9. (05.12.2019)

with best regards

Andreas

[MiDi]

The OPA189 always gives a 12-18uV higher output voltage than the other devices. (LTC2057/ADA4522)

today I reduced the 10K feedback resistor (R22) by paralleling a 1K resistor. (giving 909 Ohms total feedback).
Cirquit see here:
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2784834/#msg2784834

Result of the 9.1K resistor change was a 21.1 uV reduction of the offset voltage.

So the OPA189 seems to have a effective input bias current of 2.3 nA
which is much higher than the datasheet value (with a weird measurement condition of 500pF and 100k).
For me the OPA189 seems to be only suitable for source impedances below 1K at the inputs.

Would not have expected that the offset voltage of a modern device suffers significant from this typical input impedance 
Wondering if an input capacitance at the inverting input to gnd would improve this - I know this is going a bit offtopic...
Andreas, you could populate R23 with ~100pF-1nF (leakage <100pA@10V) with R22=10k, C22=10nF to verify.

These measurements (incl. charge injection with circuit by TI) are on my TODO-list, but for now it had low priority as LTC2057 and ADA4522 are used for this kind of application and no problems documented so far.
Odd that these effects are not even mentioned in datasheets and there are only few AN/documents mentioning it, afaik none does describe or gives details on it.

Edit:
There is another report of this effect in the TI Forum