Author Topic: DIY low frequency noise meter and some measurement result of voltage references  (Read 343810 times)

0 Members and 4 Guests are viewing this topic.

Offline Alex Nikitin

  • Super Contributor
  • ***
  • Posts: 1585
  • Country: gb
  • Femtoampnut and Tapehead.
    • A.N.T. Audio
I do this just for fun. Sometimes people want to see if things could be done better.

Or simpler  ;) .

Cheers

Alex
 

Offline aronake

  • Frequent Contributor
  • **
  • Posts: 614
  • Country: se
I have build one "Andreas-design" by Pmcouto

https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/msg1230192/#msg1230192

And one "LFLNA-80 inspired" build by ZhuraYuk

https://www.eevblog.com/forum/metrology/lflna-80/msg5995163/#msg5995163

Both have noice quite a bit lower than needed for measuring a ADR1000. Benefit with the LFLNA-80 inspired design is that it comes into regulation quite a bit quicker. But after my capacitor selection run finding a 2200uF + 1000uF combo with leakage below 2 nA, the "Andreas-design" is pretty quick too. Maybe around 30 seconds from idle vs 2 minutes from idle.

Where I see the biggest benefit with the "LFLNA-80 inspired" build is that it is much better at rejecting 50hz mains influence and see one third vs the Andreas-design/Pmcouto layout. I am not sure why. Both my builds are very well shielded in aluminum box, with extra copper film around the PCBs. So somehow trying to figure out good 50Hz rejection either by physical PCB layout or by earlier mentioned filter. So far I only tested on my desk with no "tin-can".

For the "LFLNA-80 inspired" build, I changed resistors so that the switch give 10.000 X or 100.000 X amplification. That is very useful. For the  Andreas-design/Pmcouto layout i modified stage 2 R9 resistor to 100 ohm to get 100.000 X amplification. I also tested to do amplfication boost in stage 1. Could not see much difference in outcome. Anybody have any view if it would be better to do the amplification boost in stage 1 (with a LT1037) or stage 2 with a LT1012?
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
My randomly chosen 1000uF/35V (105C) shows aprox 14mV DC offset at the output of the 4523, that leads to sub 1nA leakage (and the DC offset at the opamp input comparable with the opamp's DS offset). The capacitor is at 7V for "years", however (I've been using my noise meter for burn-in too).
Readers discretion is advised..
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 6265
  • Country: us
My randomly chosen 1000uF/35V (105C) shows aprox 14mV DC offset at the output of the 4523, that leads to sub 1nA leakage (and the DC offset at the opamp input comparable with the opamp's DS offset). The capacitor is at 7V for "years", however (I've been using my noise meter for burn-in too).

That's impressively low leakage for any electrolytic, at 1000uF even more so!! Do you see this across different brands?

Best
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 
The following users thanked this post: p-wave

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
If you ask me - I have here a bunch of capacitors in my junkbox (since ever perhaps), so I took the first one and soldered in. A small leakage current is not a problem, unless a) it fluctuates, b) it creates a drop on the input resistor which saturates the opamp. Frankly, I was expecting couple of volts at the opamp's output (gain 10000) there, but I see almost nothing. Perhaps I've been doing something wrong, or it is just a beginner's luck..  :D

PS: the 1000uF/35V radial capacitor has the name "LXV", "105degC 1 (2)" on it.
« Last Edit: August 20, 2025, 03:53:09 pm by iMo »
Readers discretion is advised..
 
The following users thanked this post: aronake

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 6265
  • Country: us
If you ask me - I have here a bunch of capacitors in my junkbox (since ever perhaps), so I took the first one and soldered in. A small leakage current is not a problem, unless a) it fluctuates, b) it creates a drop on the input resistor which saturates the opamp. Frankly, I was expecting couple of volts at the opamp's output (gain 10000) there, but I see almost nothing. Perhaps I've been doing something wrong, or it is just a beginner's luck..  :D

PS: the 1000uF/35V radial capacitor has the name "LXV", "105degC 1 (2)" on it.

Yes, that's what we would expect with a gain of 10,000 with less than a 1/2 ua of leakage!! What DC bias did you have across the capacitor?

Best
« Last Edit: August 21, 2025, 01:34:13 am by mawyatt »
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 6265
  • Country: us
Back in the days of Jim Williams legendary low noise work, electrolyics were not very good and about the only reasonable option size-wise for larger value capacitors were the wet-slug tantalums. We also used these in many applications and they weren't quite as expensive as today.

Now the electrolytics are much better all around and available from many sources, our own recent simple test also showed a surprisingly (to us) low leakage. Pre-stressing these electrolytics seems to improve them even more!!

Best
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
..Yes, that's what we would expect with a gain of 10,000 with less than a 1/2 ua of leakage!! What DC bias did you have across the capacitor?

With a 10Meg inp multimeter I see: 6.90V across the capacitor (currently the 1399#2 there), zero at 1k5 (in+), , zero at in-, +9.96V Vcc, -4.99V Vee, and the opamp's out is between -4mV and -20mV jumping..
So the output is actually with negative offset, on the scope like around that -14mV mid.

PS: I put there randomly chosen:

LF356 - output started at -4.65V, and the same after ~20minutes
OP27 - output started at +9.32V, and +0.88V after ~20minutes

So it seems the dominant input offset comes from the opamp and my capacitor's leakage current is causing an offset (on the 1k5) comparable with the 4523 datasheet offset (+/- 0.5uV typ, max +/-4uV).

PPS: and I invented a "Low Cost Capacitor's Leakage Current Indicator" as well..  :D :D
« Last Edit: August 21, 2025, 07:29:50 am by iMo »
Readers discretion is advised..
 

Offline Gerhard_dk4xp

  • Frequent Contributor
  • **
  • Posts: 402
  • Country: de
Back in the days of Jim Williams legendary low noise work, electrolyics were not very good and about the only reasonable option size-wise for larger value capacitors were the wet-slug tantalums. We also used these in many applications and they weren't quite as expensive as today.

Now the electrolytics are much better all around and available from many sources, our own recent simple test also showed a surprisingly (to us) low leakage. Pre-stressing these electrolytics seems to improve them even more!!

I once tried to build a low noise JFET amplifier but the usual circuits are not really stable
( CS gain stage, cascode, op amp, feedback into the sources). Not even the circuits by
the Gods, i.e. AOE ed3....

The input stage with its 1 Ohm source resistor looks like CS but it isn't because the
source follows the input b/c of the feedback. But the feedback is slow, so the JFET
feels like a capacitively loaded follower. That is like the typical uwave VCO.
Look at the negative real part of the input impedance (yellow line) between 100 KHz
and 1.5 MHz in the simulation and also in real life: the S11 trace in the smith diagram
leaves the unit circle.

Some TI CFB opamp was fast enough but it drowned everything in 1/f noise, even
after +40 dB of pre-amplification. (A follower with BFQ19s transistor seems to
do the job.)

I tried to get around this w/o feedback. I referenced the input to GND via the usual
1Meg and to get stable operating point I fed the sources by a CCS current mirror
but decoupled them with large electrolytics.

Now that was REALLY bad. The circuit seemed to show every single electron
that defected through the electrolytics. Oscons were by far the worst. They
produced a 1/f corner in the KHz region. Nippon Chemical Alu --> low 100s of Hz,
wet slug tantalum abt. 10 Hz. I write here 1/f, but it was worse, maybe 1/f**2.
My Agilent 89441A FFT analyzer went into clipping with the Oscons.

The AVX 4700u wet slug tantalums did cost €/$ 100 a piece, the same from
Vishay cost twice that ( all parts Digikey/Mouser).

Funny part values in the simulation (i.e. 1pF) are results of what/if tests.
« Last Edit: August 21, 2025, 07:48:25 am by Gerhard_dk4xp »
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
..Now that was REALLY bad. The circuit seemed to show every single electron
that defected through the electrolytics. Oscons were by far the worst. They
produced a 1/f corner in the KHz region. Nippon Chemical Alu --> low 100s of Hz,
wet slug tantalum abt. 10 Hz. I write here 1/f, but it was worse, maybe 1/f**2.
My Agilent 89441A FFT analyzer went into clipping with the Oscons...

Therefore a good noise meter needs (except the "wide band" LNA with A=1000 or 10000 for example) a variable gain amplifier (or better two VGAs wired somewhere where appropriate in the chain, each with say A=1 to 128, or A=1/10/100 at least) and sets of higher order (like 2xSK in series) HPFs (like 1mHz, 100mHz, 10Hz for example) and LPFs (10Hz, 1kHz, 10kHz for example) followed by a 16b+ ADC or a 12b+ scope..

Before each sampling the chain has to be set up such the signal level will be say half of the ADC/scope range (to have enough headroom for the outliers)..
« Last Edit: August 21, 2025, 08:48:02 am by iMo »
Readers discretion is advised..
 

Offline Gerhard_dk4xp

  • Frequent Contributor
  • **
  • Posts: 402
  • Country: de
You mean sth. like that?    :-)

Methinks my solder iron & the white relays have a love/hate relationship.
« Last Edit: August 21, 2025, 08:47:25 am by Gerhard_dk4xp »
 
The following users thanked this post: iMo

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
Yep, something like that but with much more relays..  :D

And you would need also a calibration input, with say 80-90dB attenuator (and an another relay :) )..
« Last Edit: August 21, 2025, 09:03:01 am by iMo »
Readers discretion is advised..
 

Offline Gerhard_dk4xp

  • Frequent Contributor
  • **
  • Posts: 402
  • Country: de
There is a calibrator included: A 61 Ohm resistor + relay. It delivers a flat noise
density of 1 nV/rtHz at room temperature, directly derived from 1st principles...
You don't even need to know the exact over all gain. Just plot the difference
from 1nV/rtHz.

In the flat region, the amplifiers own noise is nearly 20 dB below that.
(300 pv/rtHz for the 16*CPH3910, 200 pV/rtHz for 10 ADA4898-Duals,
70 pV/rtHz for 16* Zetex 851 bipolars, a single ended quarter of the 
ribbon preamp from AOE3.
« Last Edit: August 21, 2025, 09:37:38 am by Gerhard_dk4xp »
 
The following users thanked this post: iMo

Offline aronake

  • Frequent Contributor
  • **
  • Posts: 614
  • Country: se
In the Andreas/pmcouto build, what is the purpose if capacitor C21?

It seems to me that R11 (1K) and C21 (100nF)  would make a low pass filter at 1590 Hz. Very far away from the upper 10Hz. There ought to be very little signal left at 1.6 kHz after 10 Hz filtering. Or do R11 and C21 interact with other components for some other cutoff frequency or purpose?

When i use my amplifier in less optimal conditions I get quite some 50 Hz line noise. If I swap the 100nf for a 6.7uF capacitor this would give 23 Hz cutoff. So around -9 dB at 10 Hz, so quite limited impact on the measurement, but almost all of the 50 Hz line noise is gone. Alternatively add a capacitor as the one in the picture below for a steeper filter.

I would think this is a bad idea as it was not in the original design. Anyone with some insight who could help to explain?

2642691-0
 

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 6265
  • Country: us
We were/are concerned about large input capacitor leakage expecting microamp levels from typical large electrolytic capacitors. Overnight we measured some Rubycon ZLH 1500uF 16V capacitors at 10V bias, all 3 measured below 200na after settling down, which was much lower than we expected!!

We stressed these 3 Rubycon caps overnight again, this time at rated 16V. Just grabbed one and it's measuring ~11na under 9na leakage with 16V bias. So seems a very slow downward trend in leakage due to voltage bias.

Best
« Last Edit: August 21, 2025, 06:55:15 pm by mawyatt »
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 
The following users thanked this post: p-wave

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3593
  • Country: de
Or do R11 and C21 interact with other components for some other cutoff frequency or purpose?
Hello,
R12 + C21 are intended to keep RF from the output (Oscilloscope/Mains) away from being demodulated on the protection diodes of the OP-Amp. (giving a unintended Offset).
R11 isolates the OP-Amp against the capacitive load (and oscillations).
They have no intended influence on the frequency response.

Generally it would be better to have the mains frequency filtering at the input of the cirquit. (preventing clipping of the OP-Amps).

with best regards

Andreas
 
The following users thanked this post: aronake

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
..imho, the C21 100nF (most probably that MKS foil you are using often there in your schematics) is not good in suppressing of "RF", I am afraid.. (well, perhaps when RF is with small single MHz digits)..

For anything above say 20-50MHz (the strong RF fields today are from say 40-50MHz up) you would need small size and value ceramic capacitor (because of its small parasitic inductance) and a small inductance (like 1uH to 10uH in a toroid or ferrite bead) instead of the R12 resistor..

The most used combination with reputable manufacturers in the RF sector is a lowpass CLC wiring with C=1n to 4n7 and L=1 to 10uH (in form of a ferrite bead or toroid, not an axial choke), afaik.

For suppressing RF from, say, 0.5GHz-1GHz up (perhaps the strongest fields around you today - ie. mobile phones, wifi, xG networks, etc., etc.) you would need even smaller physical sizes and values (and often special construction and materials) for that CLC lowpass.

Btw., the oscilloscope itself generates a noise containing all the frequencies mentioned above (because of high clock and high speed devices inside).

..
R12 + C21 are intended to keep RF from the output (Oscilloscope/Mains) away from being demodulated..
« Last Edit: August 22, 2025, 05:54:31 am by iMo »
Readers discretion is advised..
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
I've started to experiment with TheoB's python tool "DHO Remote" he created while back for people without the Bode.
https://www.eevblog.com/forum/testgear/bode-plot-for-scope-owners-that-do-not-have-that-option/msg5770157/#msg5770157

The Bode analyzer includes a standalone FFT with some math inside, it can set channel, V/div and secs/div, FFT windowing funct., and download binary data off the DHO Rigols via LAN (there is a placeholder for Siglents without the download code - so perhaps Siglents are easily to add, Rigol template may help).
Output graphs are interactive - zoom etc. (with mouse).

Now, 10Mpts to download from DHO804 via LAN takes aprox 3 seconds, and the FFT with some noise math and graphs takes another 3 seconds (it depends on what you do).

So the working FFT part is there, after some minor improvements it could be easily used for spectral noise analysis - what to add for example:
a) an input for the LNA Gain
b) an input for the N (FFT averaging)
c) an automation for N FFTs collection (Nx single shot with averaging..)
d) an adjustments in the FFT and noise related math and graphs - what would be actually needed and what is the correct math (standardization)
e) etc.

I've played with it briefly and it works fine, imho. Below some random creations with new graphs and my attempt with some noise math (all with 10-100sec/div, no averaging, smoothing, 10Mpts).. It is an example only..

PS: AI started to mess into my experiments.. :)
« Last Edit: August 27, 2025, 04:04:14 pm by iMo »
Readers discretion is advised..
 
The following users thanked this post: Simmed

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
And here is an averaging of the FFT spectra.
MAC199#6 DUT in my simple noise meter (0.1-10Hz), Kaiser(14), 1sec/div, 10Mpts.
The averaging is done via powers of the complex bins, like the average over (abs(rfft)^2), and then sqrt(average) as the result..
Below NSDs for N=1, 10, 50.

PS: the reference lines there are for thermal noise of an ideal 50ohm resistor at 25C (an example only)..

PPS: the MAC199#6 has pretty large noise, with N=50 you may start to see the 1/f tendency inside the filter (not visible with N=1).
« Last Edit: August 28, 2025, 04:47:14 pm by iMo »
Readers discretion is advised..
 
The following users thanked this post: Mickle T., Simmed

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
Here is a comparison with different windows.
Still the same DUT, 10Mpts, 1sec/div, N=10, A=10000.
The windows have got their ENBW coef assigned.

PS: we need a simple noise generator/calibrator.. :)
Like XC9572XL (see the famous app note "XAPP 052 July 7,1996 Version 1.1" ) + MAX740x 8th order low pass at say 1-10kHz (its clock synced off the CPLD)..

Code: [Select]
        if window == "flat" or window == "none":
            win = np.ones(data["N"])
            enbw = 1.0
        elif window == "flattop":
            win = flattop(data["N"])
            enbw = 3.77
        elif window == "blackman":
            win = blackman(data["N"])
            enbw = 1.73
        elif window == "kaiser":
            win = kaiser(data["N"], beta=14)
            enbw = 2.23
        elif window == "hann":
            win = hann(data["N"])
            enbw = 1.5
        else:
            print("Unknown window")
« Last Edit: August 29, 2025, 07:56:28 am by iMo »
Readers discretion is advised..
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 3593
  • Country: de
Hello IMO,

if you look into your RIGOL handbook you will see that a "rectangular window" is recommended for wideband noise FFT.

with best regards

Andreas
 
The following users thanked this post: iMo

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
Rigol shall implement an user friendly and usable FFT first :) ..
I've been just running N=50 with 100sec/div, window None, hopefully the DUT batteries will last..

PS: I've stopped the sampling - using PC for that is BS, imho, now trying to port the dho-remote into my Armbian TVbox..
« Last Edit: August 29, 2025, 12:14:44 pm by iMo »
Readers discretion is advised..
 

Offline iMo

  • Super Contributor
  • ***
  • Posts: 6981
  • Country: li
Here it is on a small TVbox (Armbian linux on a 4core arm @1.5GHz, 2GB ram), it works fine as-is provided you will not set 10MPts. With 1Mpts the processing of 1 FFT till the ready graphs takes aprox 20 seconds, that is acceptable, imho. It runs Xfce over RealVNC viewer (picture from notebook), the box wired to scope via LAN..
Example below (mind the scope's AC coupling as the LNA was not settled enough during the experiment).
« Last Edit: August 31, 2025, 12:13:12 pm by iMo »
Readers discretion is advised..
 

Offline CurtisSeizert

  • Regular Contributor
  • *
  • Posts: 183
  • Country: us
I have finished layout for the PP cap/JFET input design I gave the schematic for a couple weeks ago. I prioritized making assembly reasonably easy (though you will still need a stencil). The BOM cost ended up at $75 (excluding resistors) with quantities sufficient to make one amplifier. The design files are in a Github repo (https://github.com/curtisseizert/SimpleLFLNA). I will include this board in my next JLCPCB order, so I will update the repo with testing results once they are available. I have attached a render of the PCB, and I will make some enclosure panel PCBs in the near future.
 
The following users thanked this post: 2N3055, iMo, mawyatt

Offline mawyatt

  • Super Contributor
  • ***
  • Posts: 6265
  • Country: us
I have finished layout for the PP cap/JFET input design I gave the schematic for a couple weeks ago. I prioritized making assembly reasonably easy (though you will still need a stencil). The BOM cost ended up at $75 (excluding resistors) with quantities sufficient to make one amplifier. The design files are in a Github repo (https://github.com/curtisseizert/SimpleLFLNA). I will include this board in my next JLCPCB order, so I will update the repo with testing results once they are available. I have attached a render of the PCB, and I will make some enclosure panel PCBs in the near future.

Like the idea of using the JFET Buffers to isolate the Op-amp input current noise away from the input capacitor, thus allowing a PP film input cap and higher overall input Z. The TI Dual JFET seems like a good choice, however with a +-4mv offset and +-6V supplies might require a DC Bias trim or JFET selection to keep the high gain 1000X preamp front end centered. Or maybe use the extra op-amp as a DC corrective feedback?

Anyway, agree that the electrolytics achilles heal (even with the unexpected low leakage, at least to us) is the DA, and interested to see how this moderate size PP Film Input Cap LNA behaves.

Best
Curiosity killed the cat, also depleted my wallet!
~Wyatt Labs by Mike~
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf

 

- - -
-->