Nanovoltmeters Performance

[chuckb]

I have several Nanovoltmeters (K181, K182, K1801(A10), K2182A, and HP34220A) and the results reported here are similar to what I have measured. If I had a better filing system on my PC I would post the results, but I don't so I can't.
 
Many years ago I contacted Magnicon about their fine JFET chopper preamp (mentioned in a post above) but the cost was $10,000 (if I remember correctly). That's a little out of my budget.

One of the lowest cost Nanovoltmeter configurations is an EM Electronics A10 Preamplifier with a 5 1/2 digit or 6 1/2 digit meter. This combination of equipment can be put together for less than $1000 USD.

Actually the A23 Preamp is about half the price of the A10 preamp and it has similar performance. Several years ago the cost of an A10 was under 700 Pounds Sterling, the A23 was under 400 Pounds Sterling. I am developing a support PCB for these two amplifiers. I have been using it for the last year. The details will be published soon.

Attached are some results of a 23 hour test of an A10 preamp with the input shorted. I used a 10 Megohm feedback resistor to set a gain of 10,000,000. I would have had this same performance with a gain of 10,000. I used an HP3458A to digitized the amplifier output. An 8 1/2 digit meter is not needed for this but it was already connected to the data logging program. I set it all up to record and left for a long weekend. I hoped to collect several days of data but the batteries for the preamp ran out after 24 hours. Oh well.

The preamp has a white noise of 0.52nV / rt Hz. That's pretty good! This is about ten time less than the HP34420A. The lower white noise allows you to reach the same test equipment noise level 100 times faster.

There is a 23pV stability floor at 1000 seconds. I believe this an artifact of the 1nV / deg C temp co and the local temperature instability. I have installed the preamp in a temperature controlled metal box to reduce this instability for previous projects.

Stable 32 was used to process the raw data. I attached the Allan Dev results if someone wants to combine the results of all the meters in one graph. Or post your Allan Deviation data and I will gladly do it.

[TiN]

Several years ago the cost of an A10 was under 700 Pounds Sterling, the A23 was under 400 Pounds Sterling.

About right what I paid few weeks ago.

The preamp has a white noise of 0.52nV / rt Hz. That's pretty good! This is about ten time less than the HP34420A.

I'll have to get busy to make some more X1801 boards to test my amps.

[maxwell3e10]

This is impressively low 1/f noise! I think they use a transformer to provide modulation and amplification at the input.  Thanks for all the info. I am going to get one of these.

[maxwell3e10]

I am wondering what is the dynamic range of these  EM Electronics amplifiers. The spec sheet says the maximum input voltage range is +/- 2mV and the maximum output voltage is +/-3 V. So it seems that running them with x1000 gain would use most of the dynamic range. It would also amplify the noise above the level of a general purpose multimeter.
So, the question is what would be the noise level with x1000 gain amplification? Is it limited by output stage noise?

[chuckb]

The amplifier noise will be the same as long as the meter does not add any instabilities. For simple null tests I run it at absurd gains (10,000,000) so I don't have to think about the noise from the digitizer (DVM).

I started a test of 6 nanovoltmeters this afternoon. I'm testing the noise of a A10, A22, A23, K181, K182 and a HP34420A. The EM preamps are running at a gain of 1000. A K2000 is scanning and digitizing the outputs. If the noise is an issue I'll change the gain to 10,000 and try it again.

For the A10 and A23 the factory recommends a minimum gain of 1500 (1500 ohm feedback resistor). This minimizes issues with the output voltage swing limit of +-3V and the +-2ma output limit. With the Keithly 1801 preamp (A10) they could use a gain of 1000 because they were on a +-2V meter range. The output current limitation caused me issues when I was running at a gain of 1000 with a 3mv offset. I did not realize I had exceeded the 2ma limit. The amplifier does not clip it seems to do a gain compression.

The A22 module uses a 10 ohm internal resistor so it can work with a 20mv input differential voltage.

In the Keithley 1801 datasheet or manual there is a block diagram of the EM preamp operation. It chops the signal at 488Hz and amplifies it with a transformer. There has been at least one disassembly of the EM Electronics N1 or N2 Nanovoltmeters on the forms.

Many years ago I got a badly damaged A10 amplifier and did some reverse engineering. Including the potted mu metal box, which was a real pain in the ass to open. I hope I can still find the documents. I don't promise that they are complete or correct, but it shouldn't be that difficult to fill in the gaps.

A10 circuit:

This had a partial schematic attached. I did a little Photoshop work on it and have attached it below.

Hopefully this is not to far off topic...

[maxwell3e10]

Eagerly awaiting your data to decide which one of the EM preamps I should get.

[branadic]

Eagerly awaiting your data to decide which one of the EM preamps I should get.

Dito, that will be kind of interesting, a comparison with comparable ambient conditions. 

-branadic-

[Kleinstein]

The noise seen for amplifiers can still depend on the DMM behind it, even if the DMMs own noise is low. Quite a some DMMs are often used in an auto-zero mode, that is about equivalent to sampling the input for some time (e.g. 1-10 PLC) and than a zero for some 1-10 PLC.  This sampling the input only for about halt the time and this increases (e.g. doubles) the noise bandwidth seen.  However a few meters have the option to use an filter or/and use an chopper amplifier input so that the signal is sampled all the time and thus no higher bandwidth.  So the meter and the settings used can make a slight difference.

[chuckb]

I started the testing last week but there was to much temperature variation for good results. It's probably all the inadvertent thermocouples in my wiring. I reduced the temperature variation around the EM amplifiers and they are more stable now. When Keithley sold the K1801 preamp based on the A10 preamp it included a foam enclosure to provide some isolation from air drafts and short term temperature variation. I have the preamplifiers inside cast Hammond enclosures but a layer of 2" foam insulation would help also.

I'm using a Keithley 2000 with a scanner card to sample the output voltage of the EM amplifiers. After the temperature variation was taken care of I still did not like the noise level. I increased the A10 and A23 gains up to 100k. The A22 now has a gain of 10k. I like to give these tests a few days settle and get steady numbers. I will take a look at the data tomorrow.

The K2000 scanner card relays have a 1uV offset spec. I have verified that but I also typically see a 0.1uV repeatability when the scanner comes back to the same channel. I should short one channel to get the meter noise. I will see what I can do.

[chuckb]

Here's some AV results for an HP34420A and a Keithley 181. I looks like the K181 has an internal 10 second filter that you cannot shut off.  My results are similar to what Alex posted earlier.

Each meter had a shorting plug on the input. The AV bump at 30,000 seconds (8 hours) is probably temperature regulation in the lab. The temperature stays within +-1 deg C in the lab.

I tested the K182 also but it had 5x more white noise than the K181 so I will have to look into that.

[chuckb]

During the testing of the EM Electronics A10, A22 and A23 preamps I learned a few things.
The preamps want a stable temperature and a dedicated meter to monitor the output.

Attached is an AV plot for 5 preamps. The top 4 traces are all in the same thermal area on the same shelf in the lab. They are scanned by the same K2000 meter. The A10 and A23 are set up with a gain of 100,000. The A22 is setup with a gain of 10,000. They are sampled once every 8 seconds by the K2000. The preamp support PCB has a 4.7ufd cap across the feedback for roughly a 1 second filter. Each preamp has individual +-8v power supplies.

All of these 4 preamps are in cast metal boxes with screwed on lids. Performance noticeably improved when I simply wrapped the metal boxes with a shop rag. This improved short term temperature regulation.

NOTE: The support PCB for these preamps includes a reversing relay to help isolate external wiring thermal EMFs. In this situation it may be adding to the apparent TC of the top 4 preamps.

The lower trace is a battery powered A10 in a metal box installed inside an insulated enclosure made of 2" thick foam (R-13). A dedicated HP3458A monitors it's output voltage. The gain was 10,000,000. This was my first setup and it does not have any support PCB. It just has point to point wiring. This preamp had no filtering in the feedback loop.

General conclusions -
These preamps need to be installed inside a metal box with a close fitting 2" foam insulation for best performance. A simple 0.1uV / deg C thermocouple somewhere can really impact the performance of a preamp with a 0.00002 uV noise floor.

I think the heavy metal case of the A10 preamp helps with the ultimate noise floor and general stability.
With this setup, it looks like the A23 preamp has twice the white noise of the A10 preamp.

Scanning the voltage out of the preamps leads to an apparent increase in AV. I believe if I scan the top 4 preamps at a 1-2 sec rate the curves will look the same they will just be shifted to the left. Then the two A10 preamp lines will match the lower A10 preamp line.


Next tests (when I get time to do them) -
Bypass the reversing relay on the support PCBs.
Sample the output of the preamps at a faster rate
Test the preamps inside a thermally insulated enclosure.

[Kleinstein]

Sampling only every 8 seconds with a filter for 1 second increases the noise bandwidth. So it is natural to have a higher noise in the white noise region. So ideally one should have a dedicated meter (this might be even a lower grade one if a high gain is used) and a filter to cover that gaps, e.g. from AZ mode of the meter. With many older meters using a 10 PLC auto-zero time the filter should be good to average over some 0.1 s at least.

The filtering might introduce an artifact (lower apparent noise) at the very short times of the AV curve, similar to the effects seen with the K181.

The better thermal insulation should mainly effect the very low frequencies and may shift the minimum more to the right by extending the white noise region to lower frequencies. So it should be import for the ultimate resolution, but not so much in the white noise region, as the thermal induced noise is usually not white. The A23 curve if more what is expected from thermal effects.

[egonotto]

Hello,

chuckb wrote:
"I looks like the K181 has an internal 10 second filter that you cannot shut off"

You can the filter only shut off with an GPIB command (is P0).

Best regards
egonotto

[branadic]

It's essential to compare all amplifiers under same condition, such as bandwidth, sample rate, filter etc. otherwise it's not a fair comparison.

-branadic-

[Echo88]

Absolutely remarkable low noise results from the A10! I would love to see one cracked open one day with a detailed reverse-engineered schematic.

[chuckb]

I am repeating the preamp testing at a higher sampling rate. Everything else that can be easily controlled is kept identical.
I started another K182 noise test with a shorting plug.

Echo88
This link is still working.
ftp://picovolt.com/Lepaisant_preamp-xfmrs_RSI.pdf
They describe an amplifier architecture that is probably similar to the EM Electronics design. There has been a teardown of the EM electronics N1 style nanovoltmeter in the last few years on the forum. There are lots of large copper plates in the input stage to keep components at exactly the same temperature.

Branadic
The A22 has a gain 10x less than the others. At these averaging times the K2000 noise would influence the A22 amplifier if the A22 noise floor approached 10pV. I will monitor that.

Bandwidth of all the amplifiers is limited by the 4.7ufd feedback capacitor. Again the A22 is different but the difference will be seen as reduced AV at very fast sampling intervals which will be apparent in the AV curve. It will not affect the white noise level or the flicker noise floor.

All four preamps are in the same style metal housing with the same support pcb supplied by identical power supplies. The inputs are shorted by a 0.1 ohm resistor.

We will see how this next round of testing goes.
BR
chuckb

[chuckb]

Background information.
The A10 and A23 amplifiers have an internal 1 ohm shunt resistor for feedback. My support PCB has a 100k feedback resistor. This yields a gain of 100,000 +-0.1%.
The A22 amplifier has an internal 10 ohm shunt resistor for feedback. My support pcb has a 100k feedback resistor. This yields a gain of 10,000 +-0.1%. So we have this basic difference between the amplifiers that we have to work with and or analyze.

[chuckb]

I was able to speed up the sampling rate for the 4 amplifiers from 8 seconds per sample to 2.3 seconds. That allowed a more accurate result.
 
Now the two A10 amplifiers monitored by a scanning K2000 have a white noise very similar to the reference A10 monitored with an HP3458A.
The reference A10 has much, much better thermal isolation so the flicker noise floor is lower.

I included the curve for the HP34420A as a reference.

My next testing will attempt to reduce the flicker noise floor of the already great thermally insulated A10 amplifier. I rearranged the parts inside the insulated enclosure. The A10 preamp is powered from two sealed 12V, 10AH batteries. To improve thermal stability I made a Lead-Acid sandwich with one battery laying on it's side, the amplifier in it's enclosure, then the other battery on top. The temperature probe (HP-2804) is taped to the side of the amplifier case. Then insulation fills the rest of the space. So far the amplifier has changed temperature by 0.01 deg C during the last 14 hours. The amplifier is still reaching the final temperature.

[TiN]

Once you have a chance I'd be interested also to see 2182A in the comparison.
Meanwhile I built second X1801 supply board and received rest of parts for more, so will join you in testing next month.

[maxwell3e10]

Thanks for all the testing! It would be interesting to test the amplifiers at lower gain. I am planning to use them in order to measure the difference between two voltage references (after heavy filtering and signal conditioning). Since its hard to make the voltage references exactly the same and there maybe some long-term differential drift, it would be nice to allow for up to 1 mV of differential voltage.

Maybe you can use the HP34420A to read the output of A10.

[chuckb]

Thanks for all the testing! It would be interesting to test the amplifiers at lower gain. I am planning to use them in order to measure the difference between two voltage references (after heavy filtering and signal conditioning). Since its hard to make the voltage references exactly the same and there maybe some long-term differential drift, it would be nice to allow for up to 1 mV of differential voltage.

Maybe you can use the HP34420A to read the output of A10.

Another option is to use an AD8429 IA amplifier at a gain of 1000. It's a lot smaller and cheaper than the A10 and it has 1nV / rt Hz of noise. If you add a CMOS sw to chop (reverse) the input stage you could stabilize the DC offset and flicker noise. With 4 of these in parallel you could approach the 0.5nv / rt Hz noise of the A10.

Now I'm way off the topic!

[Kleinstein]

There should be no real advantage of using something like the 34420 to read the amplifier output. The output can be somewhere in the +-10 or at least +-1 V range. So the important feature of the DMM would be more in having a well working low PLC AZ mode (e.g. using less than 10 PLC for the slow modes). So the DMM6500 or 3446x would be likely the better choice, as they would need less filtering for best performance.

Even the good references are more like in the >10 nV/Sqrt(Hz) range, especially in the low frequency range. So no need for a super low noise amplifier.  These amplifiers are more something if you need to read an PT10, sub-mOhms shunts or other really low impedance and low noise sources.
Much of the A10 performance is the low flicker noise, not so much the low white noise. Even AZ OPs or self made chopper amps tend to have some thermal flicker like noise - that is often higher than with the A10.

[Gerhard_dk4xp]

Is there anything known about the 1/f noise of analog switches used as choppers?
Anything more than the thermal noise of the channels? Downconverted phase noise
of the clock? Side effects of the charge injection? The web is quite silent on that.

regards, Gerhard

[chuckb]

I have not heard of any EXCESS noise from the on resistance of CMOS switches. EXCESS noise in resistors is typically proportional to the voltage drop across the resistor. When a properly sized CMOS switch (<<100 ohms) is used in a chopper there will not be any voltage drop across it so it will not have the opportunity to add noise. The CMOS switch will of course need very good power line filtering to prevent capacitively coupled noise getting to the amplifier input. You would also have slow edge rates on your control signals.

[chuckb]

I tried for a week to get the A10 AV below 22pV. I have found the limits of my local passive (insulated box) temperature stability. I also discovered EMI from two strong local broadcast stations.

One station operates at 100 MHz and supplies 100kW less than a mile away. The scope probe picks it up at several mVpp. The other AM Broadcast station 2 miles away changes power from 10kW to 1kW from 9 pm to 9 am every day. I can see that power level change every day on my nV noise plots.

To prevent local environmental disturbances, for several days I did not get within 3 meters of the A10 preamp or the HP3458A monitoring it. I also minimized any local electrical noise, no WiFI, no soldering irons, no switching power supplies, etc. I also electrically isolated the temperature probe from the AL enclosure with Kapton tape. I wanted to break up the loop antenna formed with the temperature sensor wiring and the DVM wiring. 

So out of a 213k sample record (1 sec per sample) I had 9000 samples of pretty stable data. However it had the same AV as I had several weeks ago when I left town for the weekend.

Attached is a plot of the A10 Sheetmetal enclosure temperature over several days. The temperature was basically 25.4 deg C +-0.15C.

Many more things to investigate for EMI reduction but I need to get back to other testing.

So the A10 stability is 22pV at 1000 seconds in my lab. That's good enough for now.


TiN - I checked the K2182A many years ago and it had slightly more noise (instability) than the HP34420A.