Author Topic: DIY low frequency noise meter and some measurement result of voltage references  (Read 68326 times)

0 Members and 1 Guest are viewing this topic.

Online splin

  • Frequent Contributor
  • **
  • Posts: 734
  • Country: gb
If you are looking at the wide band noise-- yes you are right.  If you are only concerned with DC-10Hz, then the current noise [@ 0.1 to 10Hz, in pApp] divides into the LF noise spec-- and that is 30nVpp for this amp.

Sorry if I didn't make it clear - I was referring to the current and voltage noise density graphs on page 7 from which came the

Quote
I make it < 500 ohms @ 10Hz (2.5pA/rt(Hz) v 1.2nV/rt(Hz)),  < 175ohms @ 1Hz (8pA/rt(Hz) v 1.4nV/rt(Hz)) and < 100ohms @ .1Hz providing you balance the inputs.
[/quote]

I didn't bother to quote the .1Hz figures but for completeness they are approx 30pA/rt(Hz) and 30nV/rt(Hz). I also haven't integrated the current noise over .1 to 10Hz but it will clearly significantly exceed the 5nV rms (30nVpp) opamp voltage noise with 500ohms source impedance.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
The LT6081 is similar to the LT1028: good for a really low impedance source only, e.g. less than 500 Ohms (balanced) at higher frequencies an less than about 100 Ohms in the LF range. So this is not practical with an AC coupled low frequency input.

Having balanced inputs often also adds extra noise from the extra resistor to balace the inputs.
 

Offline TiN

  • Super Contributor
  • ***
  • Posts: 3998
  • Country: tw
  • xDevs.com/live - 24/7 lab feed
    • xDevs.com
Anyone would be willing to make up schematics with abovementioned LT6081 to try? :)
I'd get PCBs and prototype assembly done for it to test in HW, including extra PCBs for contributors.
Plan to build something to test my LTZ's anyway, so one way or another I'd be making something, but rather make something reasonable from folks who understand all this opamp usage better than me.
YouTube | Chat room | Live-cam | Have documentation to share? Upload here! No size limit, firmware dumps, photos.
 

Offline zlymex

  • Frequent Contributor
  • **
  • Posts: 556
  • Country: cn
LME49990(dated 2009) was the improvement of LT1028. LT6081 is the same or worse than LME49990 as far as the noise is concerned. Both of them having the same low frequency noise of 30uVp-p, but the current noise of LT6081(unbalanced) is worth than LME49990.

I've considered using LT1028 or LME49990 as the first stage of the noise meter, but failed to get anything better than current ADA4528-2 mainly because the very large Ib and the incapability to make balanced input for one stage amplification. The typical Ib(Abs) of LT6081 is very similar to LT1028 and LME49990.
« Last Edit: August 18, 2016, 01:28:56 am by zlymex »
 

Online BU508A

  • Frequent Contributor
  • **
  • Posts: 639
  • Country: de
Re: DIY low frenquency noise meter
« Reply #79 on: August 18, 2016, 05:29:16 am »
Hi,

Chopper stabilized amplifiers have flat 1/f noise to DC but higher broadband noise.  You can combine a chopper stabilized amplifier with normal amplifier to get the best noise characteristics of both.

There was a design note (DN 36) from LT which describes such a combination.

Also there was a very interesting design note (DN 42) from LT with a comparison of chopper amplifiers vs. bipolar OpAmps.

And at last there is an application note (AN 20) with some considerations about instrumentation low pass filters with some comments in it about capacitors.

I've attached all PDF documents here below except the AN20, because it was too big (2.7MByte). Here is the link: http://www.linear.com/docs/4115

Andreas


“Chaos is found in greatest abundance wherever order is being sought. It always defeats order, because it is better organized.”            - Terry Pratchett -
 
The following users thanked this post: EmmanuelFaure

Offline TiN

  • Super Contributor
  • ***
  • Posts: 3998
  • Country: tw
  • xDevs.com/live - 24/7 lab feed
    • xDevs.com
Alright, I cannot let zlymex get all the glory, so let get some more nanovolts flying.

How about we build few known LNA designs to test in practical environments?

Two with quarupled opamps. LNA1 with Analog instrumentation amp, AD8428. Paper about it here. Second is discrete one, using ADA4522-4, copycat of zlymex's design.



And third one is classic from Jim with few changes, based off Linear AN124.



I do realize gain of these amps are different, so perhaps option to set equal gain should be considered, or just leave them as is, using "black box" test approach, to see which one has less noise.

 :-DMM
YouTube | Chat room | Live-cam | Have documentation to share? Upload here! No size limit, firmware dumps, photos.
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
The AD8428 is a BJT based design and thus has high current noise. So like the LT1028 it is good for low impedance sources (e.g. < 1 K, better even < 300 Ohms) only and thus not a good choice for something like an AC coupled low frequency amplifier. Having more units in parallel only makes it even lower impedance. The AD8428 might be an interesting option at high frequencies.

For combining the output of 4 amplifiers there is no need for precision resistors - just about any resistor will do - if you want even carbon type and 20% tolerance.

For an AC coupled LF amplifier it is usually better to have the coss over frequency of the input RC lower (and thus higher resistance) and set the lower frequency limit at a later stage or in software.
 

Offline zlymex

  • Frequent Contributor
  • **
  • Posts: 556
  • Country: cn
Those are very nice design of TiN, though I worried about the current noise for MK I circuit as well.
AD8428 is specified as 150pAp-p LF current noise, parallel four makes 300pAp-p, this will create 150nVp-p voltage noise on an 500 Ohm input impedance, and this noise voltage will overwhelm the 40nVp-p typical voltage noise of the opamp(20nVp-p if 4 in parallel).
However, the current noise spec of opamps may well be wrong(may be either under-stated or over-stated), so it is worth trying it out.
 

Offline TiN

  • Super Contributor
  • ***
  • Posts: 3998
  • Country: tw
  • xDevs.com/live - 24/7 lab feed
    • xDevs.com
Let's say nice design once it's on the table and working nice. I'd be careful on cheering without any practical validation, yet.

In such case just with few parts swap it could be then DC coupled and used as null-detector between referenced for example. :)
YouTube | Chat room | Live-cam | Have documentation to share? Upload here! No size limit, firmware dumps, photos.
 

Offline zlymex

  • Frequent Contributor
  • **
  • Posts: 556
  • Country: cn
That's right, DC coupled LNA. I've been considering DIY an altra-low noise(<160nVp-p), variable voltage reference(0.9V-10.1V) as the base for that.
This will allow opamps with small voltage noise but large current noise to be used as the LNA.
« Last Edit: September 28, 2016, 01:30:47 pm by zlymex »
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
With a BJT type amplifier the high current noise is normal and well understood. So I won't have much hope for a miracle. There might be quite some scattering in the 1/f part, but this is more like a few bad parts, not a few magic ones that don't follow established theory.

Current noise in AZ OPs is a much more complicated thing as noise could depend on details like capacitance at the inputs or even decoupling and supply voltage. Also measurements are more tricky at lower levels and not all chips might be the same here, as the bias current can show quite some scattering. So with AZ OP the current noise specs can be tricky and may have errors. So here you can be lucky and find some that that are much better than specs - but also overly optimistic specs.

So the AD8428 is more like a good amplifier for the 100 Hz - 100 kHz Band, or maybe with DC coupling for the 1 Hz - range. However in the low range, there will be trouble from 1/f noise and chopper amps should be better. So not really good for a null meter either.

For the AN124 like design the 2N4393 JFET is an unusual choice. A good low noise JFET would be the BF862 (at least some of them have good 1/f noise). The ADA4822 does not make much sense with a 100 K input resistor - it needs smaller resistors (and thus larger caps) to take full advantage of the low noise specs of the OP. The choice of resistors on the drain side of the JFETs is also a little strange (I would expect more equal values for R33 and R34). Also near 10 mA looks like a lot of current for the FETs. At so much current I would consider limiting the voltage by using a cascode or bootstrapping to limit the heating and thus thermal "noise". 1/f noise often gets better with lower current per transistor and more of them in parallel. With the JFETs having a few in parallel is not that bad (except for costs and matching), as current noise is usually still low.
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 2334
  • Country: de
Hello,

I don´t know what exactly you want to do. But be carefully.

When lowering the input impedance of your amplifier too much you will get a voltage divider
by the output resistance of your noise source (typical up to 20 Ohms for a LTZ1000) and your input resistance.
So around 1000 Ohms are a practical lower limit.

With best regards

Andreas
 

Offline TiN

  • Super Contributor
  • ***
  • Posts: 3998
  • Country: tw
  • xDevs.com/live - 24/7 lab feed
    • xDevs.com
Well, finally I got around and did a layout of the board for MK-I (40x100mm FR4 4L).



Green shape is solder-mask opening, so metal shield cage can be soldered on top to provide shielding and protection from air drafts.
Signals can go via SMT EMI pass-thru filters.

Schematics in PDF

I'll revise MK2 and MK3, and will do similar size boards for them too.

Actually whole process was recorded and livestreamed on YT, so you can watch the whole thing in realtime, as well as my mambling.

Part 1, 2hour 28 minutes
Part 1, 2hour 28 minutes

Part 2, 4 hours (max limit on YT for HD, so first 1hour is trunicated :((

« Last Edit: October 30, 2016, 07:07:42 pm by TiN »
YouTube | Chat room | Live-cam | Have documentation to share? Upload here! No size limit, firmware dumps, photos.
 

Offline quarks

  • Frequent Contributor
  • **
  • Posts: 821
  • Country: de
Somehow I totally missed this, so now it is bookmarked.
 

Offline Nuno_pt

  • Frequent Contributor
  • **
  • Posts: 373
  • Country: pt
The mumbling from Illya is the best. :-)

Looks like I've 3 more boards to build slowly.
« Last Edit: October 30, 2016, 06:09:24 pm by Nuno_pt »
Nuno
CT2IRY
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
On how much of the OPs one can / should use in parallel depends on the ration of voltage noise to current noise. This ratio gives the source impedance that has lowest equivalent noise power or the best noise figure.
Using more OPs in parallel can shift the optimum to lower source impedance.
So more OP in parallel only make sense if the source (here the RC circuit) is low impedance.

Here we are looking at low frequency noise so one could take the 0.1 Hz to 10 Hz noise. Using the 1 kHz noise density can give slightly different numbers.

So the ADA4522 specs are 117 nV_pp and 16 pA_pp for the 0.1 to 10 Hz range. Thus an optimum source resistance of 7.3 K Ohms. However noise current specs on the AZ OPs are a little tricky - so the real world samples might be a little different.
With the AC coupled input the impedance is frequency dependent. As the noise is mainly white (not much, if any  1/f) it is more the higher frequencies (e.g. 1 or 10 Hz) that set the limit. It is the capacitor that sets the limit. For a 1 Hz frequency of interest each OP would like to have at least 20 µF of capacitance.
So it depends on the capacitor and the frequency of interest how many OPs in parallel make sense.

Similar things apply to BJT based OPs like the LT1007.
Here the LT1007 has best noise figure somewhere around 100 Ohms in the LF range (e.g. 10 nV and 100 pA at 0.1 Hz). In the higher frequency range the best noise figure is at about 600 Ohms. The AD8428 would be similar or lower in impedance - so it is very limited use in using them in parallel in the LF range.

As these amps have 1/f noise the lower frequencies are more relevant. So the LT1007 would like to have a quite large input cap, so more like 5000µF or more.

Paralleling might be more useful for JFET based OPs, like the OPA140. Even with something like 6-8 in parallel to get a similar 0.1 -10 Hz noise level as the ADA4522 the current noise should still be low in the LF range.

 
The following users thanked this post: zlymex

Offline TiN

  • Super Contributor
  • ***
  • Posts: 3998
  • Country: tw
  • xDevs.com/live - 24/7 lab feed
    • xDevs.com
I think it's cheap to try build all three variants and test them on practice using various DUTs at input. Also different input RC can be tested
YouTube | Chat room | Live-cam | Have documentation to share? Upload here! No size limit, firmware dumps, photos.
 

Offline VintageNut

  • Frequent Contributor
  • **
  • Posts: 520
  • Country: 00
I have a question about using the noise meter with an oscilloscope.

What is the noise floor of the Picoscope that is shown in the measurement in this thread?

The scope that I am using seems to have a noise floor of 640 uV p-p when set in 20MHz bandwidth mode. In 250 MHz bandwidth mode the scope noise is 1.4 mV p-p. The scope is set for 1X probe and is on the 1mV/div range which is the most sensitive range.

If the noise floor of the noise meter is somewhere between 90nV p-p and 170 nV p-p from reading this thread and some app notes on the web, the gain needed to be above the scope noise floor will be 10,000.

Does that sound correct?
working instruments :Keithley 260,261,2750,7708, 2000 (calibrated), 2015, 236, 237, 238, 147, 220,  Rigol DG1032  PAR Model 128 Lock-In amplifier, Fluke 332A, Gen Res 4107 KVD, 4107D KVD, Fluke 731B X2 (calibrated), Fluke 5450A (calibrated)
 

Online Andreas

  • Super Contributor
  • ***
  • Posts: 2334
  • Country: de
Hello,

the noise of the PicoScope 5000 series is specced to be 70uV RMS (450uVpp) in 16 Bit mode in 50mV range (5 mV/Div).
see datasheet:
https://www.picotech.com/download/datasheets/MM040.en-8.pdf

But you can reduce the bandwidth to further reduce the noise.
There is a 20 MHz hardware bandwidth filter and several software FIR filters.
Either resolution enhancement to 20 Bits (a 256 sliding average filter) or a configurable edge frequency (e.g. 1kHz).
With this filter the noise is at 15 uVpp referred to the input of the scope. See also:
https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/msg938724/#msg938724

The 10000:1 amplifier is needed to scale from nV to mV.
In my case the 10000:1 amplifier determines the noise floor around (100nVpp) with either 20 Bit resolution or the 1 kHz filter on the scope.

with best regards

Andreas

edit: 15uVpp scope noise alone with filter
« Last Edit: November 01, 2016, 12:49:38 pm by Andreas »
 

Offline VintageNut

  • Frequent Contributor
  • **
  • Posts: 520
  • Country: 00
Hello,

the noise of the PicoScope 5000 series is specced to be 70uV RMS (450uVpp) in 16 Bit mode in 50mV range (5 mV/Div).
see datasheet:
https://www.picotech.com/download/datasheets/MM040.en-8.pdf

But you can reduce the bandwidth to further reduce the noise.
There is a 20 MHz hardware bandwidth filter and several software FIR filters.
Either resolution enhancement to 20 Bits (a 256 sliding average filter) or a configurable edge frequency (e.g. 1kHz).
With this filter the noise is at 15 uVpp referred to the input of the scope. See also:
https://www.eevblog.com/forum/metrology/diy-low-frenquency-noise-meter/msg938724/#msg938724

The 10000:1 amplifier is needed to scale from nV to mV.
In my case the 10000:1 amplifier determines the noise floor around (100nVpp) with either 20 Bit resolution or the 1 kHz filter on the scope.

with best regards

Andreas

edit: 15uVpp scope noise alone with filter

Thank you. That is what I was looking to know. I can probably approximate this with the scope that I have.
working instruments :Keithley 260,261,2750,7708, 2000 (calibrated), 2015, 236, 237, 238, 147, 220,  Rigol DG1032  PAR Model 128 Lock-In amplifier, Fluke 332A, Gen Res 4107 KVD, 4107D KVD, Fluke 731B X2 (calibrated), Fluke 5450A (calibrated)
 

Offline VintageNut

  • Frequent Contributor
  • **
  • Posts: 520
  • Country: 00
I decided to see what the DMM7510 can do with the digitizer. Using 1000 samples/second, and a crappy banana plug with some cheap speaker wire as a short, the noise is 7.2 uV p-p. Respectable.
working instruments :Keithley 260,261,2750,7708, 2000 (calibrated), 2015, 236, 237, 238, 147, 220,  Rigol DG1032  PAR Model 128 Lock-In amplifier, Fluke 332A, Gen Res 4107 KVD, 4107D KVD, Fluke 731B X2 (calibrated), Fluke 5450A (calibrated)
 

Offline VintageNut

  • Frequent Contributor
  • **
  • Posts: 520
  • Country: 00
The DMM7510 cooked overnight and some today. Attached is the screen capture with statistics for the shorted input on the 100mV range.

10.04uV p-p, 4.02uV average, Std Dev 1.03uV, 33,000,000+ readings at 1000 samples per second, 9+ hours.

working instruments :Keithley 260,261,2750,7708, 2000 (calibrated), 2015, 236, 237, 238, 147, 220,  Rigol DG1032  PAR Model 128 Lock-In amplifier, Fluke 332A, Gen Res 4107 KVD, 4107D KVD, Fluke 731B X2 (calibrated), Fluke 5450A (calibrated)
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
To measure noise, the peak to peak reading is a little difficult, as it is by definition sensitive to singular events. So it will always be not that well reproducible. Today the better way to measure noise is using FFT and than measure the noise spectrum. A fast DMM like the 7510 is a good option here.
For a given frequency band (e.g. 0.1 -10 Hz as the typical for low frequency noise) one can also use the RMS value (= Std deviation) - this number is much better reproducible than the peak to peak value. To get the right bandwidth this would be more like 10 s windows with something like a 20 Hz sampling rate.

Doing measurements on very long times, there will be a mixture of noise and drift. To get rid of the drift one might want to add a lower frequency limit, by using a software high pass filter before calculation RMS values.
So the lone run gives a number for about the 0.03 mHz - 500 Hz range. The upper band limit depends on filters used in the DMM.
 
The following users thanked this post: TiN

Offline VintageNut

  • Frequent Contributor
  • **
  • Posts: 520
  • Country: 00
The raw buffer data is always available for the DMM7510. A post-processing filter can perform whatever the user wants.
I am not too worried about bandwidth. The noise meter that is being proposed has the band limit built-in.

On my oscilloscope the ratio of scope channel p-p noise to RMS is something like 10X, 600 uVp-p and 50 uV RMS. 
working instruments :Keithley 260,261,2750,7708, 2000 (calibrated), 2015, 236, 237, 238, 147, 220,  Rigol DG1032  PAR Model 128 Lock-In amplifier, Fluke 332A, Gen Res 4107 KVD, 4107D KVD, Fluke 731B X2 (calibrated), Fluke 5450A (calibrated)
 

Online Kleinstein

  • Super Contributor
  • ***
  • Posts: 5973
  • Country: de
The ratio of peak to peak value to RMS value depends on the bandwidth (especially the ration of lower and upper limit), the length of time window and the type of noise (white or 1/f). The usual number is something like a factor of 6 or a little more (e.g. 7).

A ration of 10 for the scope should be due to the high bandwidth ratio (e.g 0.1 Hz - MHz).

A noise measurements needs to keep in mind the frequency band - otherwise the values are not comparable with other instruments. For a very accurate comparison one would also need to note the type of filter (e.g. 2nd or 3rd order, steepness) - there is more than just the band limit to characterize a filter. The concept of noise equivalent bandwidth only works it the type of noise is known (e.g 1/f, white or other).

The noise meter as shown has some band-filters build in, but there will be also limits from the DMM / scope. As shown there is only a first of second order high pass filter - this may not be enough if there is heavy 1/f (an maybe 1/f² part) noise. The initial AC coupling adds some noise in the transition region. So ideally there should be an extra filter (e.g digital in the DMM or PC)  to set the lower frequency limit instead of the initial RC combination.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf