Low frequency, very low level, DC biased, noise measurements

[janaf]

New thread for the discussion on low frequency, very low level, DC biased, noise measurements, like uV, nV noise from voltage references.

A paper on a JFET front amp:

Ultra low-noise preamplifier for low-frequency noise measurements in electron devices

http://www.researchgate.net/profile/Bruno_Neri5/publication/3087527_Ultra_low-noise_preamplifier_for_low-frequency_noise_measurementsin_electron_devices/links/00b7d52dd4825a9b72000000.pdf

[Kleinstein]

This is a very interesting circuit in the article. The good thing is that it works with a moderate size capacitor (thus foil types are available) at the input and also allows relatively high source impedance.
 
However there are two minor drawbacks:
1) the used JFETs are old types not available any more. Choosing the right Fets sets the 1/f noise corner. Finding good modern replacements is needed. The 2SK369 may be a candidate.
2) Performance likely very much depends on the thermal layout and setup. So it's more than just putting the parts together on a board. Especially at low frequencies thermal effects can add extra noise.

[janaf]

This is a very interesting circuit in the article. The good thing is that it works with a moderate size capacitor (thus foil types are available) at the input and also allows relatively high source impedance.
 
However there are two minor drawbacks:
1) the used JFETs are old types not available any more. Choosing the right Fets sets the 1/f noise corner. Finding good modern replacements is needed. The 2SK369 may be a candidate.
2) Performance likely very much depends on the thermal layout and setup. So it's more than just putting the parts together on a board. Especially at low frequencies thermal effects can add extra noise.

Even the 2SK369 is also obsolete? There is the Linear systems LSK369 which should be equivalent, also available in dual in one can.

The 2SK369 is available on ebay. Seller polida2008 who many have good experiences of, has singles in TO-92. Should be matched and thermally coupled....

[Andreas]

Even the 2SK369 is also obsolete? There is the Linear systems LSK369 which should be equivalent, also available in dual in one can.

The 2SK369 is available on ebay. Seller polida2008 who many have good experiences of, has singles in TO-92. Should be matched and thermally coupled....

Also on Reichelt:
http://www.reichelt.de/2SK-369/3/index.html?&ACTION=3&LA=446&ARTICLE=2198&artnr=2SK+369&SEARCH=2sk369

With best regards

Andreas

Edit: I would use FETs only if I really need a high input impedance.

[Marco]

JFETs have to be matched for this amplifier, precise Vgs matching is not critical since it's not a differential amplifier but if they're too far apart they won't be biased right. That said, the switch they use to charge the input capacitor in reasonable time would load a LTZ1000 too much as Andreas pointed out (the low value input resistor in Jim Williams application note probably does too).

If you just want to throw expensive parts at it the LT1128 is so good though I'd say just use it as a buffer. After that you can use a HPF with a low value resistor to the next stage without issue and use a switch to quickly charge the capacitor as well. Then use a LT1028 as a 10000x amplifier and a standard 0.1-10 Hz measurement setup (ala this). It's not elegant, but should work (LTZ1000 has low dynamic resistance right?).

[Andreas]

(LTZ1000 has low dynamic resistance right?).

Hello.

Not really compared to a LM399 (< 1 Ohm)  or LT1027 (< 0.05 Ohm) reference.
The Datasheet indicates 5mV change for 1mA change at the output = 5 Ohms.
The zener alone is even worse: around 20 Ohms.

If you also want to measure ordinary (precision) zeners (1N829A) you will have around 10-20 Ohms.
So for a noise amplifier I would not go below around 1000 Ohms as input impedance.

With best regards

Andreas

[Marco]

Still not enough to push the LT1128 current noise into relevance.

[Kleinstein]

The trouble with the current noise is the impedance of the coupling capacitor. This will be likely in the 1 K - 10 k range unless really huge caps are used. So using a high current noise amplifier will need a capacitor of considerable size. Allready the OP27 would still need something like 5 mF. So like a LT1028 would like something like 50mF or more. Finding a good cap may not be easy, as noise and leakage at 10 V are not the usual parameters for electrolytic caps.

I see several option that may work.
1) the JFET amp with good discrete JFETs and good thermal setup, just like first link showed. This may use a foil cap, like 10 µF. Just finding good Fets (low 1/f noise) can be a problem. The old 2SK147 seemed to be exceptional good, especially the 1/f crossover.

2) Using a large mF range cap and an BJT based OP, lke LT1037. The choice of Amplifier scales with the cap.  This was proven to work, if good caps are found. Protection of the source is a little difficult, but possible.

3) use a compound amplifier, like Jim Williams in AN124. This at least good a the higher frequency part, e.g. 10 Hz - 1 kHz. Though I don't see much advantage in the 1/f range of both amplifiers, like below about 3 Hz.

4) Using DC amplification to a high voltage (e.g. 100 V) and AC coupling after this. The following stage is less critical and could use a foil cap and a less critical JFET amplifier. This may be tricky with battery supply.

5) Possibly using a Chopper amplifier like LT2057 or similar, with AC coupling in the 10 µF range. Here things are not so clear, as there can be more tricky effects like noise foldback and higher EMI susceptibility. However the data look promising so far, though data on current noise are not as complete as one would like it.

[janaf]

(LTZ1000 has low dynamic resistance right?).

The LTZ1000 itself, as mentioned some 20ohm but for the whole circuit, it's very low, milliohms at most as far as I can see. I was surprised to hear that the output took so long to recover after a mA surge. Can't really understand it (not doubting that you are right!)

I put some 10mF+ elyth caps on charge overnight. It looks like the leakage is surprisingly high in the order of 50Mhom at 10V below rated voltage (63 and 80V caps). Typically around 1uA leakage at 50V. It seems the resistance does not increase at lower voltage but it's really hard to measure.

I also had a smaller 1000uF/50V at rated voltage, now running it with a 9V battery. Leakage  should be proportional to capacitance, some nA. I'm measuring on it now. The good news is that the noise seems very low. While there is a leakage, the current seems quite stable, drifting very slowly, much below 0.01Hz as far as I can see. So it may cause some offset on amp input, but offset drift should be slow. Right now, after a few hours on battery, measuring -88nA (current flowing from cap to battery), with drift in the order of 10pA/s, on the last digit of my DMM resolution. The current is still drifting towards zero. I have no idea where it will stop. It could be the battery leakage current that is seen. By the looks of it now, it may take days to stabilize. It may of course also be temperature drift. But then the mass of the caps is high so drift should be slow. The voltage is now dropping by less than 1uV/s (10Gohm input DMM).

[janaf]

The trouble with the current noise is the impedance of the coupling capacitor. This will be likely in the 1 K - 10 k range unless really huge caps are used.

Agree. Noise will be very dependent on input impedance / input resistor  cap and that in turn dependent on amp input current leakage.

Right now I'm trying to see if E-caps are usable (as indicated from the German forum). It looks like it would be very possible. Again, there are tradeoffs as leakage is proportional to capacitance, inversely proportional to voltage rating vs voltage used. So a big cap but not too big.

[janaf]

Could someone who has his/her math's fresh in mind do humanity a favor; come up with a simple function for ball park calculating RMS noise from 0.1Hz to 10Hz. I would really like to be able to calculate it, based on (integrating) white noise level and 1/f corner frequency.  8)Possibly also it's derivate, unless that's constant of some physical reason.

[janaf]

The current in the cap I'm testing, could be dielectric absorption too?
Which makes me wonder how that influences filtering.

It's now at -46.9nA, still with a very steady drift towards zero current (Has a 9V battery connected)

[Kleinstein]

Calculating the RMS noise shloud be realtively easy:
One can integrate the power (e.g. voltage square) over frequency. We can also treat the white and 1/f parts separately, as it is not correlated.
So the white part just gives bandwith (BW) times square of noise-voltage.
The 1/f part takes one more step: The Integral of 1/f² is  -1/f. Integrated from f_low to high frequencies this gives 1/f_low.
At the corner frequency f_c, the white and 1/f part are of same size.
So with just an white a 1/f part and integrating to well abov f_c one would get:

U_noise² =   U²_n(white) *  BW + U²_n(white)*f²_c/ f_low =  U²_n(white) * ( BW + f²_c/f_low)

So due to the 1/f part the noise power is inceased by a factor of ( 1 +  f_c² / (f_low *BW)).
For the noise density  (volts per sqrt(Hz)) take the square root of this factor.

For the LF amplifier there may well be a 1/f² part, e.g. from current noise time capacitor impedance. So we have a third anlog contribution
and get a faktor of

Sqrt(  ( 1 +  f_c² / (f_low *BW) + f³_c2 / (2*f_low² *BW))) ) to muliply the white noise density.



A "leakage" current in the 50 nA range sound good. The current itself is not so much of a problem, its just the noise that is possibly associated with it that may cause trouble. For filtering, the input AC coupling should have a time constant considerable lower than the measurement limit (e.g. 0.1 Hz). Its better to have a later stage or software filtering will set the lower limit. So 50 nA of leakage at a 10 K resistor to GND gives something like 0.5 mV of offset and limits the amplification of the first stage to something like 1000 times, possibly 100 times, which is OK.

[Andreas]

Hello,

when I selected my caps (according to the cirquit from german forum)
I had around 2-3 out of 10 capacitors (new old stock) with suitable low leakage current
after two days forming.

According to "branadic" you should use normal standard 85 degree types.
So no low impedance switch mode types which are optimized for low ESR but no low leakage.
Also I cannot understand that someone recommends the "OS-Con" type capacitors:
they are also not optimized for low leakage.

A higher nominal voltage than needed (e.g. 25V) will also help to reduce the leakage current around 7V.
But be aware that any heating (soldering) will increase leakage current.
So cool the pins of the capacitor with a pincer during soldering.

With best regards

Andreas

[dom0]

https://www.mikrocontroller.net/topic/207061?page=2#3410803

100 mHz - 100 kHz
.5 nV^2/Hz white noise share
.7 nV^2/Hz over full bandwidth assuming white spectrum, i.e. it also has a very low 1/f corner

[Rupunzell]

Not a simple or one device-design fits most topic. To achieve the lowest noise performance requires a LOT more than just circuit design, low noise devices, construction techniques, low noise power supply and ....

The design criteria must be very well defined for the device or item being measured and what the expected signal needs to be for the rest of the measurement system.

"Cook book" or off the shelf IC solutions often never achieve the lowest noise, highest signal fidelity and best overall performance. The best low noise designs are optimized for a specific application.

Bipolar transistors can be low noise depending on what the signal input might be. The do have a problem of noise current dependent on collector current. Data books once published noise contour maps to help designers optimize a particular device to a specific circuit and system.

JFETS don't really have this problem except the larger the JFET's gate area which results in lower device resistance and lower noise caused high input capacitance. One example of a large gate area JFET would be the Interfet process NJ3600 spec'ed at 0.35nV/root-Hz at 30Hz. The 30Hz spec is significant as low noise is difficult to achieve at low frequencies.  Design trade off cost for this device, input capacitance of ~600pF at VDS =10V.

Other low noise JFETS are ones made by Moxtek
http://moxtek.com/wp-content/uploads/pdfs/n-channel-ultra-low-noise-jfets/X-RayJFETs.pdf
These offer low noise and low input capacitance when operated at -100 degrees C.

The Physics folks tend to use germanium JFETS cooled to -xyz degrees C to achieve low noise for target arrays.
http://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20000031730.pdf

Microwave folks often use GaAs FETs and numerous other FETs to achieve low noise. LNA's also appear as parametric amplifiers, to Masers.

Single ended input devices are 3db lower noise than a diff pair front end, except the single ended input is more difficult to manage than using a diff pair.

Going beyond low noise devices, there is an entire world of grounding, shielding, field management and powering to allow optimum low noise performance. Suggested reading: Grounding and shielding techniques by Ralph Morrison. First editions of this book were focused more on noise, grounding and related, later editions omitted some of what was covered in the first editions but added sections on digital related problems.

Trying to get an extremely low noise front end to live with digital back end is always difficult. More often than not, the digital section is located else where from the input section and put to sleep as the digital switching noise tends to get back into the inout section causing added noise and other problems.

There are a host of power supply related, system related problems that hinder low noise and signal fidelity performance with any added digital processing or control system seriously compounding this problem.


This is a rather complex question with a lot of different and complex answers.

Bernice

[Rupunzell]

These are mostly forgotten today, Analog Devices 301 Parametric Op-Amp.

http://www.analog.com/library/analogDialogue/cd/vol1n2.pdf

http://www.analog.com/library/analogDialogue/cd/vol1n3.pdf



Bernice

[tggzzz]

The Art of Electronics, third edition, has a very long chapter on low noise techniques. It contains many practical and theoretical results, some of which are well-known, some of which came as a surprise to me.

For example, I would never have guessed that (for some applications) the lowest noise transistor would be a medium power BJT!

[janaf]

Input cap leakage, an epiphany, surely not known anywhere on earth before 

1) Voltage change due to discharge / stray resistance of caps is proportional to stray parallel resistance, RP and capacitance, C. The RP/C is typically constant for a capacitor type, regardless of capacitance.

This means that the voltage in a cap under self-discharge is a characteristic, regardless of size. IE all caps in a series should show the voltage declining at the same rate, regardless of size!

This characteristic rate can be simply measured for one size of cap, and should then be valid for all sizes in the same series (with the usual fine-prints about voltage ratings, reality, different manufacturing lines, boundary effect of can size.......)

Better: it also means that you can compare any two capacitors, different series in a very simple way. Pich for example a 10.000 uF e-cap and a 10uF polyester, charge two to the same voltage, keep one side connected, for example the negative, and measure the voltage difference or current flow between the two. The direction tells which cap leaks most, which is best, regardless of capacitance size!

So you can simply have a shootouts between series, see which one is best. Then if you like, select another cap size from the same series and it should have essentially the same RP/C.

Finally you can of course have a shootout between caps of the same size&series. 

This is only valid assuming your measuring device does not add/subtract any current. A good o'l analogue panel type meter would seem the safe choice.

[Kleinstein]

Here is another link to an LF JFET based Ampl.:
users.cosylab.com/~msekoranja/tmp/04447683.pdf

Its a little newer (2008) than the old one using obsolete 2SK147. There best choice of JFET was a rather expensiv IF9030 - at least it's still available. The Sk369 (still available , maybe as LSK369) is there second choice - still only 15 nV/sqrt(Hz) at 0.1 Hz.

If this really turns out to be true for new devices, a JFET amp with the 2SK369 should be a rather good solution, likely limited by thermal effects.

[Marco]

About getting rid of the expensive capacitors, there's an interesting article on Electronicdesign.com.

They replace an integrator with a DAC in a feedback loop ... they used a differential amplifier, but the concept should work for a normal non inverting amplifier as well (you use a non inverting integrator to create the reference voltage for the feedback network).

PS. dropping the stage amplification to say 100x and and replacing Rg with some resistors and a wirewound pot across a 9V battery could work too (amplification would shift a bit with trimming, but nothing too bad).

[Kleinstein]

For the expensive capacitor, it seem to be possible to use some types of normal AL-Caps, after selection for low leakage as well. So I don't see a real need for an expansive Ta cap. So it can come down to testing a few caps.

Such a large cap, still has the small problem, that there will be some "drift" due to temperature and time dependent "leakage". Since the low frequency limit should not come from the AC coupling anyway, this is not a big deal.
The large cap will likely need some protection circuit for the source and possible for the amplifier as well (in case of a short). In addition we may want a build in battery to keep it under voltage, even if not used. Still the size of the cap matters, especially if we want to go lower than 0.1 Hz as well: 1 mF at 0.1 Hz has an impedance of around 1.5 kOhms. This is significant for the current noise of a BJT amplifier. A FET based amplifier may get away with a more moderate foil type cap of something like 10µF.

Having a second low noise source to compensate the DC voltage may be an option. However we get the extra noise if that source, though this may be rather low. For this path I think a stack of 1.5 batteries may be enough: the fist DC coupled stage would then only see something like 1.5 V maximum and could use a 10 fold DC amplification at 20 V supply. After this amplification things a relatively simple.

If we don't mind some extra time for measurements and digital calculations afterwards, one could also use the classical way of using 2 separate amplifiers and (now digital) cross correlation. So the "amplifier" box would likely have two ADCs inside and give out digital values (e.g. 16 bit at 60-500 Hz). This also gives RMS noise as a function of frequency, but not direct peak to peak however. The good thing is, that it works even for values much lower than the amplifier voltage-noise - the limit is set by averaging time.

To a limited extend one can also calculate back the amplifiers noise in a single canal setup. This is especially true for an amplifier with low current noise, like a JFET one. However this may introduce systematic errors if noise levels change over time.

[dom0]

Did anyone ever try something like a Dicke switch?

The technique sounds very promising, but I really don't care near enough for 1/f ultra low noise to develop something in that direction ; an extended/modified version of the mikroncontroller.net link above serves me just fine for my needs.
(Although I wouldn't mind a similar spec'd differential amplifier with good n×50 Hz CMRR, but that's really only a minor nuisance for me).

Edit: switch selection might be an issue at <1 Hz (thermal offset drift due to coil heating), but should be no issue with low thermal EMF reed relays.

[MK]

For relays, use the latching type with a short pulse to change over, like the lymex design in one of these precision threads somewhere.

[MK]

another low noise fet is the BF862 suitably low capacitance for the gm it provides, and low noise too, just put 4-8 in parallel.