Author Topic: DIY low frequency noise meter and some measurement result of voltage references  (Read 70451 times)

cat87 and 1 Guest are viewing this topic.

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
@Kleinstain

Actually I cannot even take a photo of the scope with my phone. The last photo I published was taken from 2 meters away or the all setup was disturbed.

I'll try to play with some of this stuff in next days.

Inviato dal mio Nexus 6P utilizzando Tapatalk

 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
When I look at your photo and compare:

On my PCB I have intentionally used no socket for the pre-amplifier stage.
(On the 2nd stage I also have only a socket because I had to exchange the OP-Amp against another type).

I also use some cloth (cotton pads) to keep air currents away from the pre-amplifier on both sides of the PCB.

And finally I keep the input capacitor under bias to keep the leakage current (and according noise) low.
(otherwise you will have to wait 2 days before the noise level settles down each time you want to measure).

with best regards

Andreas
 

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
Hi Adreas,

I found that main source of all my noise problems was the flaky BNC cable I used in first place.
As soon as I replaced with another cheap cable of the same type noise and susceptibility went down and
all measurements now seems quite consistent and reproducible. I can now even use it with "mains" powered
references running out of the panettone box. In this configuration it still pick up some 50Hz common mode
noise, but it's well down to the noise level and really acceptable for "quick" and handy tests setup.

Today, after I swapped the BNC cable I take the chance also to measure noise floor again by connecting a 9V
alkaline battery. After waiting many hours it went down to the equivalent of 150nV (p2p in 12 seconds) that
should be what we want here. If I'm not wrong this should add up a +1% circa of noise when measuring our
LTZ1000 circuits.

>> I also use some cloth (cotton pads) to keep air currents away from the pre-amplifier on both sides of the PCB.

I have also insulated the PCB from the the rest of the box by putting it inside a tight cartoon box.
I didn't put any cotton around the cap, this single 6mF cap is huge so it wan't be effected too much by the small air
draft that can be generated inside the small cartoon box.

I still have some "deflections" now and than, but I think this is mainly because of the cap leakage than because of
thermal drafts. I designed my PCB with slots so that you can thermally insulate the first amplification stage and solder
on some shielding too. In next days (or when the final PCB come) I will try to check what improvement I will have with
this other level of shielding.

>> On my PCB I have intentionally used no socket for the pre-amplifier stage.

I knew this could be a source of problems, but I considered that for a low frequency AC application this should
be less problematic (or not?). I used socket also because I could never imagine that this first prototype will end up
working so well. I will have the PCB fabricated very soon with dual layer layout and my intentions are to not to use
sockets there.

>> And finally I keep the input capacitor under bias to keep the leakage current (and according noise) low.
>> (otherwise you will have to wait 2 days before the noise level settles down each time you want to measure).

Well yes I understood this very soon. My settling time seems to be 8 to 12 hours if I make the cap discharge
for a short period of time, or up to 24 hours if disharge time is longer. More than 24h seems not to improve the
noise.

I also figured out that you can use the open 3.6K ohm resistor jumper pins as shunt to check when leakage
and dielectric absorption has calmed down (be careful if you use it with unbuffered/uncompensated LTZ1000
circuits as it can start oscillating).

So, even if maybe by selecting a less leaky cap and by putting some more thermal/EMI shield overall performance
can be improved I'm now totally satisfied with it. This also confirm how good is your design.

Time to dedicate my time back to the 4 way LTZ1000 averaged circuit. I still have to start building the mini
Arduino programmable thermal chamber too.
« Last Edit: March 19, 2017, 12:36:22 am by mimmus78 »
 
The following users thanked this post: Andreas

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
Good results :-+

by the way: for me it was the 5th design within several months
until I got to my LTZ1000-target  (< 0.3uVpp). I.E. to have
less than 10% influence by the noise floor.

with best regards

Andreas
 

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
Andreas it was the third PCB I routed and got fabricated. This was the main part I didn't trust. 😁

Inviato dal mio Nexus 6P utilizzando Tapatalk
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Hi Andreas and mimmus78,

I’m about to join the club and build this LF-LN amplifier.

Looking at the pictures, I can see you both used WIMA film capacitors.
What type did you use, PET or PP (MKS or MKP)?

Now I’m starting to look for a suitable input capacitor.
Like you, I’m not willing to spend thousands of Euros buying wet slug tantalum capacitors…
Searching my own stock, I could only find some top-brand (Panasonic, Rubycon, CDE) capacitors – all low ESR and rated for 105 C…

Just out of curiosity, I picked a 2,200uF 35V Panasonic (FR series) and checked it for leakage.
As expected, leakage is not as low as required for this project but surprisingly good – The screenshot below shows the results after approximately 10 hours at 10V.

I’ll order some 85 C capacitors from several sources and check them. Hopefully, I’ll find a few meeting the requirements.

Pedro Couto
 

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
Pedro,

I used MKS but I don't know if MKP would be better for this application. I leave the reply to the ones with more knowledge.

I tested all my capacitor stock, and the best one was a single cap rated 6.XmF 16V 105°C, so don't buy more capacitors unless you are sure you tested all your stock. You can find a good capacitor also among 105°C and lower voltage rated ones.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 6219
  • Country: de
For the film caps MKS should be good enough. The later stages don't contribute significant to noise anyway. MKP might be better in some aspects, but much larger form factor and more expensive at essentially no effect.

For the electrolytic caps it might be worth looking at classical (not low ESR) caps. It could also help (is faster) to do forming at a higher voltage.
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de

What type did you use, PET or PP (MKS or MKP)?

Now I’m starting to look for a suitable input capacitor.


Hello,

I used WIMA MKS-2. For MKP you would need a monster large cookies box.  ;D

As Kleinstein already mentioned: the most critical is the input capacitor.
Branadic examined that 85 deg C are usually lower leakage than 105 deg C types.
I took what I had in the drawer. (also 85 deg C types).
From 10 good quality capacitors you should be able to get 2-4 suitable ones.
(after 2 days forming).
Solder with minimum heating of the capacitor. Otherwise leakage will rise.

with best regards

Andreas

 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Thanks for your very informative replies.

Considering this specific application, I also thought that using MKP vs. MKS capacitors would not result in any measurable difference.
But it’s always better to have other opinions, to make sure one is not missing any important point.  :)

Good to know about your experience selecting the input electrolytic capacitor – You gave me some very useful tips.
I’ve ordered some different capacitors, both 85 C and non “low ESR”. I’ll also measure the leakage current of the ones I currently have in my stock.
Unfortunately I‘m not (yet) equipped to measure several capacitors simultaneously, so this will take a lot of time…

P.S.
I’ll be using MKS caps but that won’t stop me from getting a large cookies box.  >:D


Pedro Couto
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Gerber files sent to PCB house.
PCB was designed to fit inside an aluminum Hammond enclosure (1590T).
Searching for suitable input capacitors is work in progress…  :-DMM

Pedro Couto
 

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
Nice!

Hope to send gerbers of "my" design tomorrow too along with other two others PCBs.

What eda you used?
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
Gerber files sent to PCB house.
PCB was designed to fit inside an aluminum Hammond enclosure (1590T).

nice design  :-+

with best regards

Andreas
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Thanks  :)

I use Altium Designer.


Pedro Couto
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
After some delay in customs, PCBs have finally arrived.
Not perfect, but good enough and acceptable:



PCBs populated with all components, except the input capacitors.
A few words about components used:
-All resistors are 1% or 0.1% metal film
-0.1% resistors are Vishay MPR24 or Welwyn RC55
-All film capacitors are WIMA MKS2
-ICs soldered directly to PCB; no sockets used



Input capacitors were selected from a lot of 70 pieces from different manufacturers (Rubycon PK series, Wurth Elek. WCAP-ATG8 series, ELNA RE3 and RA3 series) and sources such as Farnell, Mouser and eBay (yes, I know that I probably bought some fake capacitors from eBay sellers…).
Capacitors were connected to a PSU set to 80% of nominal voltage and left “soaking” for approximately 3 days.
Then, they were individually measured for leakage using a Keithley 2450 SMU set to source 10V and measure current. After 12 hours, the leakage current was noted.
From tested lot, Rubycon PK series yielded the best results.

Selected capacitors for PCB #1:
Rubycon 35PK2200MEFC16X25 - 2,220uF 35V (Measured 1,859uF@120Hz) – Leakage 5.7nA@10V after 12h
Rubycon 35PK1000MEFCT810X20 - 1,000uF 35V (Measured 833uF@120Hz) – Leakage 3.8nA@10V after 12h

PCB #2:
Rubycon 35PK2200MEFC16X25 - 2,220uF 35V (Measured 1,840uF@120Hz) – Leakage 5.5nA@10V after 12h
Rubycon 35PK1000MEFCT810X20 - 1,000uF 35V (Measured 839uF@120Hz) – Leakage 3.9nA@10V after 12h

I decided to shield the capacitors to reduce EMI. This is probably not needed as the whole PCB is already shielded inside a grounded metal enclosure.
Capacitors were insulated with a couple of layers of Kapton tape and then wrapped in copper tape.
When assembled on the PCB, a wire loop “presses” the capacitor against the large exposed copper area in the PCB, connecting the shield to circuit ground.



Finally, capacitor pins were soldered to the PCB while clamped by an hemostat, to prevent heat from reaching capacitor’s body.
This is fully assembled PCB, ready to go inside the enclosure:



A fully assembled amplifier, ready for testing.
The very crude front panel is just a laminated piece of plain paper, printed on a laser printer.




Next steps:
-Electrical test
-Performance verification


Pedro Couto
« Last Edit: May 21, 2017, 11:34:33 am by pmcouto »
 
The following users thanked this post: SeanB, 3roomlab, dr.diesel, cellularmitosis, doktor pyta, CalMachine

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
Hello,

thanks for the tip with the Rubycon capacitors.
seeing forward for your measurements of noise floor.
(Preferably with 8* NiMH AA cells).

with best regards

Andreas
 

Offline SeanB

  • Super Contributor
  • ***
  • Posts: 15089
  • Country: za
Like the label, I did that many years ago, using a copier that had colour cartridges, so I could do a 3 colour print, though I was limited to the available colours of black, blue and red. Then ran it through a hot laminator and cut it out. Nice work with the borders there though, cutting the edges to fit the case so nicely.

For the capacitors, you probably could use 63V capacitors instead of 35V ones, the thicker oxide has lower leakage current at 9V. 3 days of forming probably was excessive, you probably could get away with overnight and then place in a desiccant filled box for a further day, and then measure to find those with highest open circuit voltage, as a proxy for leakage. Tip for reforming is to use the surge voltage rating, via a series resistor, will heal any imperfections and reduce current in normal operation. Slightly reduced lifetime, but thicker oxide all over, as after all they are formed at higher than this voltage.
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Hello,

thanks for the tip with the Rubycon capacitors.
seeing forward for your measurements of noise floor.

These Rubycon capacitors were bought from Farnell, order codes 2346523 (2,200uF) and 2346522 (1,000uF).
Hopefully, I’ll have some time to complete the tests during this week and publish the results.


For the capacitors, you probably could use 63V capacitors instead of 35V ones, the thicker oxide has lower leakage current at 9V. 3 days of forming probably was excessive, you probably could get away with overnight and then place in a desiccant filled box for a further day, and then measure to find those with highest open circuit voltage, as a proxy for leakage. Tip for reforming is to use the surge voltage rating, via a series resistor, will heal any imperfections and reduce current in normal operation. Slightly reduced lifetime, but thicker oxide all over, as after all they are formed at higher than this voltage.

Thanks for the tip regarding capacitor plate (re)forming.
Always learning something new!  :)


Pedro Couto
 
The following users thanked this post: Muxr

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
I’ve assembled and tested two units. Both units produced almost the same results, so there’s no point in posting duplicate information.
Unit #2 test results below.

Electrical test

-DC voltages checked at OpAmp supply, input and output pins – All OK.

-Current draw is approximately 3mA at each power rail – Within expected value range for this circuit.

-Each amplifier stage checked for correct operation and DC gain accuracy by injecting a DC signal at stage’s input and measuring level at output – Both amplifier stages behaved as expected and measured DC gain was within expected tolerance.


Performance verification

Performance verification tests focused the following characteristics:
-Noise floor
-Filter performance and accuracy
-Gain accuracy

Equipment used:
-Tektronix MDO3054 oscilloscope
-Siglent SDG2122X Arbitrary Waveform Generator
-Mini-Circuits VAT-15 attenuator (x2)
-Huber+Suhner 6630_SMA-50-2 attenuator
-50 Ohm BNC feed through terminator
-BNC-SMA adapter (x2)
-BNC-F to BNC-F adapter
-Non-shorting BNC shielding cap
-BNC-BNC cable (x2)


Noise floor

MDO3054 noise floor, including BNC cable.
One end of the cable connected to scope’s input and the other end shielded by a non-sorting BNC cap.


MDO3054+Amplifier noise floor.
One end of the cable connected to scope’s input and the other end connected to amplifier’s output. Amplifier’s input shielded by a non-shorting BNC cap.



Filter performance and accuracy and Gain accuracy

A SDG2122X AWG was used to generate a sine wave with an amplitude of 10mVpp. The signal was then attenuated by 60db (/1,000) and fed to amplifier’s input, via a feed through 50 Ohm terminator.
Amplifier’s output was connected to scope’s input, set for DC coupling, 1 MOhm and 20 MHz BW limit. Scope acquisition set to “sample mode” (no signal averaging).

Test setup simplified diagram:



Considering expected amplifier’s 80db gain over the BW, oscilloscope should display a signal between 70.7mVpp (-3db point) and 100mVpp for input signals in the 0.1-10Hz frequency range; Outside this frequency range output signal’s amplitude should be below 70.7mVpp.


Lower corner (90mHz – 100mHz)





1Hz – 4Hz – 7Hz







Upper corner (10Hz – 11Hz)







Next steps:
-Post project documentation
-Post some “real world” noise measurements of a voltage reference

Pedro Couto
« Last Edit: May 28, 2017, 08:59:21 am by pmcouto »
 
The following users thanked this post: vindoline

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
MDO3054+Amplifier noise floor.
One end of the cable connected to scope’s input and the other end connected to amplifier’s output. Amplifier’s input shielded by a non-shorting BNC cap.

Hello,

since the input capacitor has leakage current at nominal input voltage this contributes also to the noise floor.
So the more interesting measurement would be with a ultra low noise 10 V source.
(I use 8 NiMh AA cells e.g. ENELOOP @room temperature for it).

Can you further limit the bandwidth on your scope (e.g. to 1 kHz) to reduce the scope noise?
Otherwise it would make sense to increase the gain of the amplifier.
(you want at least factor 3 less noise on the scope than the noise floor of the filter amplifier).
At the moment it looks that both contribute the same amount of noise.

with best regards

Andreas
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de

MDO3054 noise floor, including BNC cable.
One end of the cable connected to scope’s input and the other end shielded by a non-sorting BNC cap.


Mhm,

the output impedance of the amplifier is 150kOhm at 0 Hz and gets lower at higher frequencies.
The scope has 1 Meg.
perhaps it is sufficient to use the unpowered amplifier instead of open BNC to reduce the noise level on the scope.

with best regards

Andreas
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
the output impedance of the amplifier is 150kOhm at 0 Hz and gets lower at higher frequencies.
The scope has 1 Meg.
perhaps it is sufficient to use the unpowered amplifier instead of open BNC to reduce the noise level on the scope.

I tried that, but there’s no visible difference.
In fact, noise floor doesn’t change even when the input is shorted right at the scope’s input.
It seems that this scope’s noise floor is approximately 680uVpp…


I’ve tried a different scope, a MICSIG TO1104.
This scope allows user setting of BW limiting LP filter. Lowest possible frequency is 30KHz.
Vertical scale can also be set to 500uV/div, allowing better resolution on low-level signals.

The first picture below shows scope’s noise floor, BW limited at 30 KHz.
Scope connected to amplifier’s output, amplifier powered off.

The second picture shows scope + amplifier noise floor, amplifier powered on and input shielded with non-shorting BNC cap.

I have no low noise voltage source or batteries available at the moment; Will test as soon as possible.


Regards,
Pedro Couto
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2385
  • Country: de
Hello,

so with the MICSIG you gain 3 dB on the noise floor.

But obviously the display of the RMS value seems to be the value before bandwidth limiting.
(is obviously too high as it should be around 0.16-0.2 of the peak-peak value with gaussian noise).

With best regards

Andreas
 

Offline pmcouto

  • Supporter
  • ****
  • Posts: 95
  • Country: pt
Sorry for the delay; My daytime job is keeping me quite busy these days…

Pictures below show noise measurement results of 8xNiMH and 2xLiIon batteries (approximately 10.8V and 8.4V).
Batteries were placed inside a metal box and terminals connected to a panel BNC jack. “Battery Box” connected to Amplifier using a short (20cm) BNC-BNC cable.

The NiMH batteries I could find in my lab were very old (~10 years) and the terminals were tarnished, showing signs of corrosion. Although I was able to bring the cells back to life (well, sort of…) and clean the terminals using a soft wire brush, I don’t trust them – Cell degradation may be contributing to measured noise.
I’ve ordered some new Eneloop batteries and will repeat the test when I receive them.


Project files attached:

SCH – Schematics (numbers in green are Farnell order codes)
PCB – Gerber and Drill files
DG – Drill guide 1:1 (for Hammond 1590T)
FP – Front panel artwork 1:1 (for Hammond 1590T)

Note
Depending on switch manufacturer and model, SW1 and SW2 labels “Measure”/”Charge” and “BIAS”/”ON” on PCB silkscreen and front panel artwork may be reversed.


Regards,
Pedro Couto
 

Online mimmus78

  • Supporter
  • ****
  • Posts: 667
  • Country: it
I finally decided to reprint my PCB of the Andreas based noise amplifier along with other PCB I made and today I completed the assembly.

As expected it worked.

I enclose here photo of the assembled PCB and a sample of LM399 and KX LTZ1000 reference.

LM399 is very noiseless track as it was running on batteries.

KX has a little bit of noise as its range is the lowest of my scope and it was also running on mains powered power supply.

I have a couple of more PCBs that I will be happy to give away just shipping costs ... so if you want just ask me.

PS: before someone make me notice ... I know there is italienglish text on the pcb.
« Last Edit: October 30, 2017, 09:39:13 pm by mimmus78 »
 
The following users thanked this post: Andreas, MiDi


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf