DIY low frequency noise meter and some measurement result of voltage references

[zlymex]

Requirement
 - able to test low frequency noise of voltage references
 - 0.1Hz to 10Hz bandwidth
 - portable, easy to use
 - low floor noise, 0.16uVpp level(same as Linear described in AN124f by Linear)

General considerations
 - self-contained, not necessary depending on oscilloscope or DMM for reading
 - one 9V rechargeable lithium battery powered
 - charge port, oscilloscope output port provided
 - case: aluminium, 103mm by 76mm by 35mm, my usual type
 - LED display: 4.5 digits

Schematics

There are several modifications to the Linear one
1. Input capacitor C1
I bought those wet tantalum several years ago but tested not good recently. After applied 10V for 10 hrs, there is still leakage of several uA. May be the voltage I applied is much larger than Linear(2.5V) or simply the caps are bad. Either way, I give up. I have several 80uF and 22uF film caps, tested very good, but requires too many  to built up to 1000uF, and the volume is huge. Now the only option left seems to be MLCC. I bought some of those 47uF/50V before and there are still about 50 left. However, there are problems when I installed those capacitors, will be described later.


Add: later test showed that the current noise of ADA4528 is much smaller than specified. Therefore, a smaller value for C1 can be used such as 1000uF or 470uF.

2. Input resistor R1
It provides DC bypass for C1 to charge/discharge, and also function as the 0.1Hz HPF together with C1. The attenuation is -1.5dB at 0.1Hz because there is another HPF at later stage with similar property, summing up to -3.0dB at 0.1Hz.

3. Input protection
I first implement it without Rp2, D3 and D4, an opamp was fried.
Because the noise current pass thru those Rp1 and Rp2, keep them as low as possible, better not exceed R1 combined.
I use bc junction of low power transistors(such as 2SC1815, 2SA9012) for D1-D4, leakage current is around 1pA at -4V.
R2 and the switch provide slow charge for the capacitor when connect to a voltage source.

4. The amplifier
The magnification is 10000, same as Linear.
I don't want to use FET front end, and I'm not prepare to measure >10Hz.
My target is a dual opamp, voltage noise <=100nVpp  0.1Hz to 10Hz. There are many so called ultra low noise opamps that not satisfied this because they suffered from severe 1/f noise effect. For instance ADA4898 with 500nVpp noise, even if 20 paralleled, still >100nVpp.

Also, the current noise should be <=50pApp in 0.1Hz to 10Hz range, which generates <=50nVpp voltage noise at 1k impedance. Be noted also that there are many so called ultra low noise opamps that not satisfied this.

It seems that there are not many opamps left to satisfy these criteria except ADA4522-2, but I cannot find the source of purchase. I choose  ADA4528-2 instead with very similar performance except the supply voltage is a bit low.

5. Post opamp part
I use two amplifiers in parallel to further reduce the noise, the outputs are connected together by two 620 ohm resistors, and add an 33uF capacitor(C3) for 10Hz LPF. Now the LPF is second order.
Becasue C4 and R5 is another 0.1Hz HPF, there is no need for separate filter stage as in AN124f circuit.

6. Meter part
This part can be omitted if one decide to use an oscilloscope only for output. 
Unlike Liner that use paralleled bc junction and be junction of transistors for the low leakage diodes, I use only bc junction for the peak detection diode. The reverse break down voltage of a be junction may be very low, therefore Linear has to use added resistors and diodes for clamp.
R6a and R6b are just jumpers, value not important.
U4a and U4b are photMos AQY212GS, leakage only around 1pA at -4V and turn on current of only 1.5mA with sub-ohm turn on resistance.
(There is no photoMos symbol in the software I use, so that I use phototransistor symbol insdead)
U2B is the instrument amp because my LED meter is earth referenced.
I use manual reset switch only, although an automatic reset can be added if required.
The LED meter has 1.9999V range, operable from 3.4V to 20V and draw 18mA current. I modified the decimal point so that it display 199.99(uV)

7. Power supply
There are two lithium cells inside a 9V rechargeable battery, supply 7.2V to 8.4V for small current, nominal 8V.
U2A split this 8V to +4V and -4V. I choose RS3 a bit small so that it share the current necessary for LED meter which only draw current from positive rail.
D5 and D6 provide the negative supply for U1(max 5.5V for this ADA4528-2). This can be omitted if U1 is an ADA4522-2.
D7 and D9 provide charge route by input socket when power off. Caution should be taken to disconnect anything from the input socket when power off.
Rp3 provide the trickle charge current for C1 when power is off.

Here is the photo of the finished meter:

[zlymex]

Building process, modifications and lesson learned

Firstly, get together all the large parts such as the case, board, LED meter, battery and sockets, arrange them till satisfied. I'm not going to make a PCB since this is a test build, modification is inevitable.


Sockets are to be used for two opamps for easy replacement owning to test or damage. However, soldering of those MSOP is not an easy task for me.


Secondly, assembling. It almost out of hand since there the modification is continues and I'm running out of board space. I could have move the U1 toward left side but hindered by the centered LED. Anyway, here is the inside photo of workable first version.


I only installed 21 MLCC caps(the plan is 36, 1500uF), tested not good at all. It must be the severe piezoelectric effect of the MLCC that I greatly under estimated. I cannot touch or move the meter while the measurement is in progress. The touching of the table or even waking nearby seems affect the result. Here is the pulse I got when I slightly press the reset button.


Even if everything is idling, there is still exit low frequency wobbling probably because the very large TempCo or stress release or something.


Long story short, after I've tested many aluminum electrolytic capacitors, I found several very good one for the job, very low leakage at 10V(<15nA) and there is no aperiodic noise bursts at all mentioned by Linear. They are:
one Nippon Chemi-Con 1000uF 35V, one Nippon Chemi-Con 2200uF 35V, two Panasonic 3200uF 35V

I use 2200uF 35V as the final selection, here is the inside photo.


I'm very happy with this since the capacitor settle down very quickly, allowing me to switch references with no time especially for similar voltages.
As a comparison, it took Linear 24 hours to settle down their highest grade $400 price tag wet slug.

More modification on 14th May: Change value of R1, R2a, R2b, R3, R5 and C3 so that the frequency band is precise.
Also, added Rp3 and D7 for protection and R2 for slow charge of C1. After the modification, the floor noise of the meter is slightly increased from 90nVpp to 100nVpp, still well within the original planned 160nVpp.


Noise floor of my oscilloscope PicoScope6, input shorted by a special BNC cap, only 7.5nVpp, thanks Andeas proving a very useful software probe.


A quick way to settle down the input capacitor.
EE caps have very large DA and an 2200uF can be roughly models as:


When I connect a DUT to the meter, sometimes it takes more than 15 minutes for C1 to settle down so that U1 get out of saturation. The direction of the 'leakage' current has both ways: it can flow into the R1 or flow out of it depending on the history/present voltage of the capacitor. There is a quick way though for C1 to settle down fast: 'reverse' bias it. For instance, if C1 stayed at 8V for a long time(as happened when the power is turn off for long time and just turned on), I need to test 6.3V, then I'll short the input to ground for 3 seconds so that the voltage of C1 is about 1V, then I wait for one minute before actually connect to the DUT. The C1 will now settle down much quicker. However, there are times that the waiting is still too long afterwards, and I don't know which direction the leakage of the C1(the knowledge of the direction is important so that I can repeat the revere process). I solve this by adding two LEDs as can be seen below to  indicate the saturation status so that further steps can be taken.

Simple version of the noise meter

Major modification: omit the sample-hold and LED display, use one lithium cell and ADA4528-2, use smaller input capacitors since the actual current noise is very small.
The floor noise is 90nVpp(0.015uV RMS).

Operation Procedures
(Take 7V reference noise measurement for example)
1. Prepare the DUT and the noise meter(check for battery, charge if necessary)
2. Disconnect anything from the BNC input of the noise meter, connect to oscilloscope, turn the meter on, both LEDs should be lit.
3. Connect the cable to DUT(no connection to the meter yet), measure the voltage from the BNC plug to confirm the test voltage
4. Measure the voltage of C1 from the BNC input socket of the meter, should be around 8V(same as battery voltage).
5. Short the BNC socket of the meter for 3 seconds. This will make the voltage of C1 drop to around 1V.
6. Wait for 1 to 2 minutes so that C1 is 'reverse' exercised
7. Connect DUT, now the blue LED should be out indicating the negative saturation of the U1.
8. Wait for 2 to 3 minutes, the blue LED should be come back on. If the green LED goes out quickly, C1 is not exercised enough, goto step 5
9. If the waiting is too long(>5 minutes), then the exercise is excessive. Disconnect the BNC plug from the meter, turn off the meter(so that C1 is charged) for 3 seconds and turn on again, plug the BNC back. This step can be repeated.
10. If both LEDs are on for sometimes, the measurement can be performed.

[zlymex]

Some measurement result
Note also that the results here is only the noises that I tested using my DIY meter on particular voltage references that I possess.

1. Panasonic 3200uF/35V capacitor(I was told this is used for airbags), charge to 4.1V, 124nVpp
2. Panasonic NCR18650B lithium battery, charged to full more than six months ago, 4.1V, 115nVpp
3. A Chinese temperature compensated 6.3V zener, 2DW233, powered by 12V battery thru 1k resistor(5.7mA), 336nVpp, much better than a LTZ1000.
This ultra low noise characteristic of the 2DW23x series(from a particular maker) has been confirmed by many Chinese voltnuts before, but I don't believe this until I had my own test.


When current increased to 11.8mA, noise is reduce even further to 236nVpp.

4. Other measurement result is summarize in table below



5. Ordered by noise


6. Some words about 2DW23x
The one I tested is Diamond brand made by Shanghai 17th Radio Factory. I have a lot of other 2DW23x which are much inferior with noise figures ranging from 20uVpp to 100uVpp. The design and construction of this 2DW23x were completely changed although they still share the same datasheet.

Understandably the noise of a zener is inverse proportional to the square root of the zener current in theory,  and in practice I tested that 2DW233 follows this very well. The mystery is, how they achieve this kind of low noise?


I took apart one and took a photo with my card camera plus a magnifier. It seems to me that they are hand made because the die is not centered and wires are irregular.

[Andreas]

I first implement it without Rp2, D3 and D4, an opamp was fried.

Shure it was through the input?
The AD4528-2 is specified for 5.5V single supply and +/-2.75V dual supply.

Nice scope. Which PicoScope Model do you use?

I bought those wet tantalum several years ago but tested not good recently. After apply 10V for 10 hrs,

10 hrs may be not enough.
At the moment I try forming the input capacitor from 9.5 (my 9V block which is usually attached to the input) to 13V because I recognized large noise at above 10V. After 4 days now the noise level is settling from several uVpp (> 4uV) to now around 250nVpp. So you should keep the input capacitor constantly charged somewhat above the voltage that you want to measure.

With best regards

Andreas

[zlymex]

Hi Andreas,

About the burn down of the opamp, it happened very soon, I only have chances to test 2 or 3 voltages. I don't know exactly how, but it's Ok after I added Rp2, D1 and D2, there is no problem since. The normal supply for U1 is 5.2V, not 8V.

The PicoScope I use is 5442A. The one Dave's teardown(EEVBLOG #521) is 5443B

For those wet tantalum, I'm giving up. I cannot wait that long for a result because my intention is universally quick test, and the voltages may be different from time to time. My capacitor is constantly charged if switched off as can be seen from the schematics.

[Kleinstein]

For charging the input cap to 8 V when not active one could add a resistor to prevent bad things happen when connecting something to the input in this case.

It might also be a good idea to have a switchable series resistor at the input to prevent excessive load to the DUT. Some reference circuits don't like such a current spike.

Are the Phototransitstors good enough to discharge the peak detecting caps far enough - they might have a saturation voltage in the 50 mV range.

[Vgkid]

Thanks for the circuit. Do you have any information on those 2dw233 zener diodes?

[SeanB]

With those big rectangular Tantalums I will tell you the bad news. Internally they consist of 2 PCB assemblies, with regular rubber bung tantalum slug capacitor units soldered between 2 pcb end pieces, and then with PTFE wire leads to the lid section. Then this is slipped into a kapton sleeve and dropped into the can, which is then filled with a white silicone sealer, pretty soft and flexible. The lid is then put on after curing and soldered.

They are fiendishly expensive from Kemet now, and are sadly the one product they make in glass frit seal tantalum that is pure shyte, the proper glass frit seal capacitors are good essentially forever, or till the case corrodes through from the outside. The rubber bung style capacitor is used because the end cap is not soldered, so will withstand a solder bath immersion at least once. Sadly the seal is not gas tight, and with time the wet sulphuric acid electrolyte inside will diffuse through the rubber bung, and eventually the capacitor goes high ESR , and is considered failed, or leaks sulphuric acid.

Your ones having high leakage says they have dumped acid internally, and have this sitting inside the case giving the high leakage. Time to make yourself a small set of beautiful coloured ( the anode slugs are a lovely blue green colour, depending on the formation voltage and the degradation from storage) figurines in some stoppered glass test tubes, just by keeping them in there in some dilute sulphuric acid.

I was working on equipment that used these, and did rebuild a few using regular electrolytics inside the cases, as the mounting was designed to only fit these style units. I think I changed around 5000 dead tantalums over the course of a year, running through the pile of boards.

[Marco]

I don't know anything about the zener, but I was intrigued so I looked up the datasheet. Seems a simple planar zener, the third pin is superfluous.

[Andreas]

The PicoScope I use is 5442A. The one Dave's teardown(EEVBLOG #521) is 5443B

Hello zlymex,

there is still a lot of wideband noise (above 10Hz from the scope) in your screenshots. (i.e. the 73 nV)
I would either set the resolution to 16 Bits (if you use only 1 channel) and 20 Bits resolution enhancement. (in the channel menue).
Or: set the resolution to 16 Bits  and use a 1kHz or 100 Hz digital filter.
Maybe you have to increase the number of samples from 1 Meg to 5 Meg.
So you should get a nearly clean dash for the scope.

Edit: note the picture is with 50x magnifier.

With best regards

Andreas

[zlymex]

For charging the input cap to 8 V when not active one could add a resistor to prevent bad things happen when connecting something to the input in this case.

It might also be a good idea to have a switchable series resistor at the input to prevent excessive load to the DUT. Some reference circuits don't like such a current spike.

Are the Phototransitstors good enough to discharge the peak detecting caps far enough - they might have a saturation voltage in the 50 mV range.

Thanks very much, good points made for the first two, I'll modified the schematics and the circuit.

For the third one, I forgot to mention that I actually using photoMos in the circuit. There are no photoMos symbols in the drawing software(Multisim) so that I use phototransistors in the schematics instead.

[zlymex]

Thanks for the circuit. Do you have any information on those 2dw233 zener diodes?

Those 2DW23x series(2DW232, 2DW233, 2DW234, 2DW235) has been around for many year and are the only 'reference grade' zener in China. Previously known as 2DW7C and changed name about ten years ago. There are a lot of manufacturers making these devices and I have tons of them. The performance varies according to makers very much and generally are very poor (aging, tempco, noise). I've never use them in my designs/circuits.

However, someone discovered that there is this Shanghai 17th Radio Factory making this particular Diamond brand(there is a diamond symbol on face of each device) with exceptional low noise. I didn't believe it first, but people start buying/teardown/discuss about it since and more evidence for the low noise emerged.

Those 2DW233 I bought is from here: https://item.taobao.com/item.htm?spm=a1z09.2.0.0.gpf2Zk&id=35815633601
Noise comparison tests
 http://bbs.38hot.net/forum.php?mod=viewthread&tid=49306
http://bbs.38hot.net/forum.php?mod=viewthread&tid=84620
http://bbs.38hot.net/forum.php?mod=viewthread&tid=119921
http://bbs.38hot.net/forum.php?mod=viewthread&tid=120264
Teardown and analysis http://bbs.38hot.net/forum.php?mod=viewthread&tid=120731

There are many sellers at Aliexpress selling these cheaply but only buy those with diamond mark on the top such as
http://www.aliexpress.com/item/Free-shipping-2DW233-DIP3/32433963421.html
http://www.aliexpress.com/item/hot-spot-10pcs-2DW232-new-original-in-stock/32637084815.html
And preferably recently made. The first two digits on the bottom are year of make, mines are 14 and 15(2014, 2015)

The difference within the series is only the zero tempco current. For 2DW232, its 5mA. 2DW233 is 7.5mA. 2DW234 is 10mA. I prefer 2DW232 and 2DW233, but it seems not much difference, most of those actual zero TC points are larger than specified ranging from 7mA to 20mA.

I'm not giving much hope to these devices because the aging rate is a question mark.

[zlymex]

......
Your ones having high leakage says they have dumped acid internally, and have this sitting inside the case giving the high leakage. Time to make .......

I've just tore apart one, there are 5 smaller caps inside.

[zlymex]

I don't know anything about the zener, but I was intrigued so I looked up the datasheet. Seems a simple planar zener, the third pin is superfluous.

That is correct. There are two zeners in symmetrical back to back connection, therefore we can use it in either ways. The third pin is the common cathode and is connect to the case. It used to be planar structure but I'm not quite sure for these Diamond brand.

Edit: I took apart one, here is the photo by my card camera plus a magnifier

[zlymex]

The PicoScope I use is 5442A. The one Dave's teardown(EEVBLOG #521) is 5443B

Hello zlymex,

there is still a lot of wideband noise (above 10Hz from the scope) in your screenshots. (i.e. the 73 nV)
I would either set the resolution to 16 Bits (if you use only 1 channel) and 20 Bits resolution enhancement. (in the channel menue).
Or: set the resolution to 16 Bits  and use a 1kHz or 100 Hz digital filter.
Maybe you have to increase the number of samples from 1 Meg to 5 Meg.
So you should get a nearly clean dash for the scope.

Edit: note the picture is with 50x magnifier.

With best regards

Andreas

Thanks very much for the suggestion. Yes, there was lot of wideband noise that should not be appeared in the chart. I actually turn on the resolution enhancement(by default) but suspected the post-process so that I deliberately turn it off. I set it up according to your suggestion now I got around 20uVpp noise, much better than before. I then measured the noise floor of my meter, ranging from 83nVpp to 95nVpp, let's say it's 90nVpp.  I'lll update the charts above with blue remarks.

[Andreas]

Hello Zlymex,

I have defined myself a custom specific probe with attenuation 0.0001:1 (amplification 10000 fold).

So I do not have to calculate the factor 10000 manually in my measurements.
(remove ".txt" from attachment before importing to picoscope software in the probe menu).

With best regards

Andreas

[zlymex]

Hello Zlymex,

I have defined myself a custom specific probe with attenuation 0.0001:1 (amplification 10000 fold).
......

That's fantastic!
I imported it and now the noise floor of the PicoScope is only 7.5nVpp
I then measured that 2DW233 again @ 11.8mA, the noise is only 236nVpp and can be read directly from the peak to peak value at the bottom.

[Andreas]

Hello Zlymex.

yes nVpp
you should also write it on the blue text within the picture and not uVpp!!
Otherwise someone could be confused.

By the way: how do you shield your DUT?
Cookies box or tin can?

With best regards

Andreas

[Alex Nikitin]

Nice work, however I have to point out that the actual bandwidth of the circuit shown in the first post is somewhat smaller than required, only 0.16Hz to 6.5Hz (-3dB).

Cheers

Alex

[DuPe]

Thanks zlymex for sharing this.
I like it, since it is more straightforward than big Jim's AN124.

Cheers
Peter

[zlymex]

Hello Zlymex.

yes nVpp
you should also write it on the blue text within the picture and not uVpp!!
Otherwise someone could be confused.

By the way: how do you shield your DUT?
Cookies box or tin can?

With best regards

Andreas

Oh yes, thanks, updated.
I just wrap tightly around the DUT with some soft tissue. Voltage references are all low impedance and they will not be easily affected if not put something like tin can. However, they are very sensitive to thermal changes so I have to make sure there is no circulation/wind for DUT especially on leads and connections.

[zlymex]

Nice work, however I have to point out that the actual bandwidth of the circuit shown in the first post is somewhat smaller than required, only 0.16Hz to 6.5Hz (-3dB).

Cheers

Alex

Thanks for that, can you tell me how the bandwidth of 0.16Hz to 6.5Hz is calculated?

Edit: Now I understand why. I've modified the schematics and my implementation so that the bandwith now is 0.1Hz to 10Hz(-3dB).

[Andreas]

Nice work, however I have to point out that the actual bandwidth of the circuit shown in the first post is somewhat smaller than required, only 0.16Hz to 6.5Hz (-3dB).

Cheers

Alex

Thanks for that, can you tell me how the bandwidth of 0.16Hz to 6.5Hz is calculated?

Hello,

Thats something that I have overlooked too:
In my design the input capacitor is somewhat away from the 0.1 Hz edge because of the large tolerances of the electrolytics.
(so I have 3200uF * 1K  = 0.05 Hz for the input and the lower edge frequency is determined by foil capacitors with 20uF).

for the 0.1 Hz edge:
input  1/(2100 uF x 750R * 2 * PI) = 0.101 Hz for the 1st -3dB point
Output 1/(50uF * 30K * 2 * PI) =  0.106 Hz for the 2nd -3dB point.
so both high passes add to -6dB at 0.1 Hz

The -3dB point is around factor sqrt(2) higher so around at 0.14 Hz.

I would do a LTSPICE simulation for the whole.

With best regards

Andreas

[zlymex]

In that case, might probably increase the 750 to 910 Ohm, and increase 30k to 36k. And modify other resistors if necessary.

[Alex Nikitin]

In that case, might probably increase the 750 to 910 Ohm, and increase 30k to 36k. And modify other resistors if necessary.

As Andreas said, just do an LTSpice simulation of the circuit. That is what I did after I became suspicious of the frequency response with the values shown. 

Cheers

Alex