DIY 0.1 to 10Hz Noise Amplifier

[trtr6842]

I'm starting some projects where it will be useful to measure low frequency noise, so I designed and built this 0.1Hz to 10Hz noise amplifier.

(I know my handwriting sucks)


It has a gain of 2,000 V/V, which is plenty to get above the noise floor of my RTB2000 scope.  My main goal was not ultra-low noise performance, as nothing I'm working on requires measuring noise less than 1µV p-p.  Instead I wanted something that settled within a reasonable time, and wasn't picky about finding the perfect low-leakage input capacitor.  To achieve that, I went with a multi-stage design.  I also wanted it to run off a single 9V battery with decent life.




The input stage is just the high-pass RC network with some diode clamps.  While breadboarding the design I found out the hard way that hot-plugging the input to a +10V supply can very easily fry the input to the first opamps.  Also with higher input voltages, the input resistors have to dissipate a significant amount of energy, so I used a total of 4x 1206 sized parts. 

The first amplification stage is 4x parallel OPA2188's with a gain of 200.  OPA2188's are reasonably priced and have pretty good noise performance.  There are less noisy options, but all I've seen consume quite a bit of current, which would kill the 9V battery life.  A gain of 200 is plenty to get the input completely clear of the op amp noise floor, but is low enough to allow typical electrolytic capacitor leakage currents to cause minimal offsets with the 2k input resistance.  Assuming a partially depleted 9V battery, the power rails will be about ±4V, so the input capacitor leakage can be as high as 10µA before the output hits the rails.  This is no problem for most electrolytics, and this strategy worked well as every cap that came on my stuffed boards worked just fine.

Right after the 200x input stage comes a high pass filter.  This removes any DC offsets from input cap leakage and allows more gain to be applied later.  After that is is a 4th order 10Hz high-pass filter, and a final 10x gain stage.  The output is high-passed at 0.1Hz as well, taking into account a 1MΩ scope input impedance.

Overall the performance is good enough for me, but not ultra-low noise.  With a shorted input I get about 261nVp-p.  That's plenty low for what I test, which is more like LM399's in stead of LTZ1000's.  It's also been very consistent, and the FFT of it shows no significant spikes.  If I can make the assumption that my amplifier and the device under test both have truly random noise sources, I could just do the math to subtract my amplifier un-correlated noise from any measurement's I'm taking that are below 1µVp-p, but I honestly don't expect I'll need to do that with what I'm working with.

Intrinsic Noise with Shorted Input



131kPt FFT of shorted input, done with scope data and a python script, since my scope won't do such low frequency FFT's



I tested the gain by using a 1Hz 1Vp-p signal and a 1MΩ / 100Ω resistor divider, and the gain was spot-on.



I tested the frequency response with a 200 second 0.05Hz to 20Hz logrithmic sweep.  The low frequency cutoff is a little high, but not bad.  (The vertical cursors represent 0.1Hz and 10Hz on the sweep)



The inter-stage high pass filter makes the settling time pretty quick, about 30 seconds for a 100µV step.  It can take up to 2 mins when first plugging into a new DUT, but that's not bad at all.  Better than waiting forever for the input cap to re-form!


The current consumption is 5.66mA at 9V, so that should give me a good 50 hours or more on a single battery, which seems pretty reasonable for my use.

Overall I'm really happy with the design, and it has proven very useful so far.  A while back I built a 4-20mA signal calibration source based around an LM399 and some goofy configurations of some NOMCA 8 resistor networks.  It also has a +10V output, and turns out its not too bad at about 6.5Vµp-p.



Another thing I learned is that the cheap 79L12 linear regulators I'm using on a project are about 15x noisier than their 78L12 counterparts! (1mVp-p vs 70µVp-p)  I didn't expect that, but I confirmed it on a sample size of 5 boards, so that's good to know.

I got 5 of these made through JLCPCB's SMT service, so I have 4 extra if anyone is interested in board-only or fully-enclosed one!

[mawyatt]

Nice work 

Curious as to why you chose SMD chip caps for use in the Sallen-Key Filters and output DC blocking cap. Are these ceramic because they are physically small size? Also are they C0G/NP0 types, or one of the higher dielectric constant ceramics?

Best,

[trtr6842]

I chose ceramics simply for size and convenience.  They are X7R types, nothing special.

My thought process was yes, they will be very imperfect, but their impact should be fairly small since they all come after the initial 200x gain stage.  The most critical ones are the two 10µF ones used in the sallen-key high pass filter since any noise the generate from capacitance variance could show up in the 0.1Hz to 10Hz bandwidth.  However, 10µF film caps are too big to fit in this enclosure.  1µF film caps are a bit more reasonable, but that would push the resistor values to 1.6M and 3.3M.  That's not too bad for thermal noise since it's after the 200x stage, and is actually a really good option.

For the low-pass caps, they are only dominant at 10Hz and above, and their effect is divided by the 200x first stage gain.

Overall I think the 260nVp-p is still dominated by the input resistor and opamps, but exploring film caps could be interesting if I ever need to improve this circuit!

[mawyatt]

Yes, the Sallen-Key HP would be the most sensitive with the 10uF X7R.

Agree the 10uF film are huge (used in our LCR DC Bias adapter), so maybe a 2.2uF Film, 10uF Tantalum, Electrolytic, or Poly. All these should be better than a X7R cap, altho larger.

Awhile back we developed a DC Bias adapter for our lab bench LCR meters to test some various caps including SMD ceramic types. This allowed studying DC Bias effects, but we also found that the High K ceramic types not only vary significantly with DC bias & temp, but just about everything, even the quality X7R from Samsung!! Just taping on them creates a huge piezo electric effect, not to mention PCB stress and so on. We decided to only use the High K ceramic caps for supply decoupling and bypassing and nothing else.

Best,

[MasterT]

High pass filter is likely to be off, since it is not expecting 3k /( 2-4 ) output impedance from first stage

[mawyatt]

The HP input impedance is too high for a few K source impedance to have much of an effect.

Best,

[trtr6842]

For the high-pass sallen-key I couldn't use polarized caps because they could see some significant negative voltages.  With larger supply rails I could have done some creative biasing and used them, but that was beyond what I could do with a single 9V battery.  I am actually working on some simple mods right now and will include changing the 10µF X7R caps to 1µF film ones and will update the resistors accordingly, I'll post those results soon.

For the high pass error in cutoff frequency, its definitely not the extra 250Ω of input impedance, here's an LTspice sim with input impedance of the high-pass filter, and the frequency response.  The sim steps the input resistor from 1µΩ to 250Ω and almost no effect is visible.  My guess is it's either tolerance on the 10µF ceramic caps, or I did the math wrong on where to put the vertical cursor!!!

[MasterT]

The HP input impedance is too high for a few K source impedance to have much of an effect.

Best,

My excel tells 10u has 1.6k at 10 Hz.

I recently get opa2182 to evaluate, and noticed that if I set G=1000 with 1k & 1M opa behaves as "normal" - noisy. But if I tried to put a cap in parallel with 1M ( 10 nF or so to get 10Hz) broadband noise drops as it should, But to my surprise in the  low frequency 0.1-10 Hz band noise jumps up! I can't say it was the first time I've seen abnormality when AZ interact and getting upset by caps - at the input or outputs, so I make a rules for myself not to connect any caps directly to az w/o isolating R in series. Any new az I buy first checked for this phenomenon.

[mawyatt]

The HP input impedance is too high for a few K source impedance to have much of an effect.

Best,

My excel tells 10u has 1.6k at 10 Hz.

As shown by the simulation above, the source impedance doesn't have much effect as we mentioned earlier.

The 10uF cap is working in a very high impedance circuit as can be seen by the 82K and 330K resistors, thus the source impedance of <1K doesn't matter.

Best,

[mawyatt]

For the high-pass sallen-key I couldn't use polarized caps because they could see some significant negative voltages.  With larger supply rails I could have done some creative biasing and used them, but that was beyond what I could do with a single 9V battery.  I am actually working on some simple mods right now and will include changing the 10µF X7R caps to 1µF film ones and will update the resistors accordingly, I'll post those results soon.

Another possibility is a non-polarized electrolytic, or back to back regular electrolytics.

Best,

[trtr6842]

I tried a couple mods on one of the other boards, and here are the results:


Up first was swapping the input 1206 thick film resistors for metal foil ones.  The effect was negligible.



Next I left the metal film input resistors, and also modded the high-pass sallen-key circuit.  I changed the caps to 1µF 50V film caps, and replaced the resistors with 10x the original value for the same frequency response.  Again, the effect was negligible, so those 10µF caps weren't a significant cause of noise, or if they were, the slight added noise from the 10x resistor values cancelled it out.



I went back to look at my initial noise analysis too.  I realized my gain stages were different than stated, they are 500x then 4x!  (pdf had outdated values, original writeup and .jpg schematics were correct).  Same overall concept, but I was back-and forth on what combinations to do and I ended up forgetting what I actually ordered.  Anyways... I did a full LTspice noise sim of the circuit as-ordered.  I modeled the OPA2188's as an ideal noiseless op amp with a 4.9k input resistor for an effective flat noise density of 9nV√Hz, which is what they're spec'd at.  The overall noise of the whole circuit with ideal passive components was calculated at 31.44µV RMS integrated from 0.05Hz to 20Hz. 

All of my measurements are within a couple percent of that!  So that means all the generic thick-film 0603 resistors and X7R caps I used are acting fairly ideally, and/or the input noise is dominated by the input stage.  The same LTspice sim shows that the 0.05Hz to 20Hz noise just after the 500x stage is 37.46nV RMS, slightly higher than the output noise.  This is due to the higher bandwidth before all the filtering.

[Kleinstein]

The design looks very reasonable. I like having the input AC coupling at a lower frequency than the later low pass filter.
The 10 µF X7R capacitors should see very little voltage and under these conditions they may be OK. They still may have quite some dirft with time and temperature and this could effect the bandwidth. The tolerance may also be an issue. The bandwidth for noise tests a tricky point anyway, as the noise BW is different from the -3dB BW. This is also a problem with data found in datasheets.

I don't see a major noise noise besides the OP-amps and R22+R31. So I would have expected a less noise as the specs for a single OPA2182 are at 120 nV_PP.
With 4 OP-amps in parallel the resistor noise is expected to be even more then the amplifier noise.
Another possible noise source is the input capacitor, as the leakage current is expected to be somewhat noisy.
For a test it may make sense also to also measure with a short behind the capacitor.

For comparing the noise performance the RMS noise is often more practical than the PP noise as it has less scattering. If the PP noise is used one should use "PPA" and not "PPP" as this would be PP over a longer interval.

At least from the plan the gain of the final stage is still x 10 and not x 4. The extra gain could about explain the higher than expected noise.

[trtr6842]

So it turns out my original write up had all the right values, first stage is 200x, second is 10x.  The pdf I attached had outdated values for a 5,000x design.  In reconciling those mixups, I found the cause of the bad low-frequency response:  R11 should be 82k in stead of 160k, so the Q of that filter was too low.

I fixed my LTspice noise sim, and I actually should be getting about 27.8nV RMS, so my circuit has about 14nV RMS of extra noise somewhere, my gut is that it's current noise through the input 2k resistor, since that isn't present in my simulation at all.

For noise measurements, I'm mainly going off of the standard deviation measurements, since those are simply AC RMS calculations.  My scope has a significant DC offset, so the RMS values are skewed, but the standard deviation (σ) values are perfect.  I confirmed that by taking RMS, σ, and mean measurements, and they all lined up where RMS^2 = mean^2 + σ^2.  While I'm talking about the scope, it's noise floor with the 0.0005x attenuation factor shows <=65nVp-p and <=7nV RMS when using the high-res acquisition mode, which is what was used for all previous screenshots & measurements.  As for the "Vppp" and "Vppa" measurements, I had no idea if those were any sort of standard notation, I just made that up on the spot.  My logic for taking the highest Vpp value of 10 separate 12-second captures is twofold:  The high pass filter is 3rd order, and removes a lot of noise beyond the 10-second span, so doing longer Vpp captures doesn't make too much sense.  Also, 10-second @ 1s/div captures are standard on datasheets, and my scope simply has a 12 division horizontal span.

On to some more test results:


Reducing the input resistance to 1k offered a good improvement, down to about 200nVp-p.  In theory the 2k input resistor had 18.2nV RMS of thermal noise, and the 1k one has 12.86nV RMS.  Ignoring op-amp current noise, swapping the two should have brought the noise from 31.27nV RMS to 28.5nV RMS.  Instead, the noise dropped down to 26.28nV RMS, indicating that 11nV RMS of current noise was also removed, putting the total 4x averaged input current noise of the OPA2188 at around 3.4pA√Hz, assuming a 10Hz bandwidth. 

Shorting R22 and R31 effectively makes this circuit very similar to the other designs I've seen around here which have around 100Ω of input impedance.  With some care that can be used, but input impedance that low with clamping diodes scares me.  I am very hesitant to reduce the input resistance, since I see that as dangerous for the potentially sensitive DUT's that I'll be plugging this into.  Since reference aging is so common, hot-plugging noise testers like this seems to be the default to avoid power-cycling the DUT.  Since the input cap is so large, a design with only 100Ω of series input impedance before the clamp diodes would draw a surge of 100mA with a decay time constant of ~220ms from a +10V reference.  Depending on the reference buffer (if there even is one!), that might not be trivial.  The 2k||2k design I made would only draw 10mA from a +10V reference, which is within the safe range for many common buffer op amps. 

An easy solution would be to integrate a precharge and bypass switch.  Have something like 2k||2k  always connected to charge up the input cap in about 10 seconds, then flip a switch to bypass the input resistance (R22 and R31 in my case).  I don't like that idea because I know I would forget to turn off the bypass switch before connecting to a new DUT!!

I'm curious if taking advantage of the high voltage supply handing of the opamps could be useful...  For example, change the input clamp circuit so that it clamps at -0.7V for negative voltages, but at something like +24V for positive voltages using a zener.  Add a schottky in series with the positive supply rail after the large decoupling caps to prevent reverse current, and that should allow the input to swing up without drawing huge surges of current.  As the input cap charges up the opamp supply rails would settle and return to the 9V battery supply, and then everything should work as normal.  I might have to breadboard that!

[trtr6842]

The frequency response was all cleaned up when I fixed the Q of the high-pass sallen key.  R11 should be 82k if the 10µF caps are used, or 820k if the 1µF caps are used.

[mawyatt]

I tried a couple mods on one of the other boards, and here are the results:


Up first was swapping the input 1206 thick film resistors for metal foil ones.  The effect was negligible.



Next I left the metal film input resistors, and also modded the high-pass sallen-key circuit.  I changed the caps to 1µF 50V film caps, and replaced the resistors with 10x the original value for the same frequency response.  Again, the effect was negligible, so those 10µF caps weren't a significant cause of noise, or if they were, the slight added noise from the 10x resistor values cancelled it out.

The 10X resistors should produce root(10) more voltage noise per resistor, so yes likely they are contributing more noise and offsetting the film capacitor effect. A better test would be to use a 10uF film cap with the same resistors as used with the X7R capacitors, but they are big. You might want to try an 10uF electrolytic, which is much smaller than even the 1uF film you've shown.

Another crude test would be to tap on the case with various capacitors for comparisons, the X7R are notorious for mechanical vibration sensitivity (piezo effect), the other types are much less prone to such.

Anyway, if the X7R caps are working acceptability, then that's good news

BTW what grade of X7R cap did you use?

Best,

[trtr6842]

Since all the filtering is after the initial 200x gain stage, noisy components there don't really contribute much to the input-referred noise.  My LTspice sim shows 27.89nV RMS noise with ideal 10µF / 10µF 82k, 330k for the high-pass sallen-key, and 27.915nV RMS  (Both values are input-referred). 

The specific caps are GRM21BR61H106KE43L, but again I think that everything after the 200x stage is essentially trivial. 

My testing with the lower and shorted input resistor is interesting though, since it's suggesting that the opamps might have higher than expected current noise, or that my clamp diodes are somehow introducing current noise...

[mawyatt]

An easy solution would be to integrate a precharge and bypass switch.  Have something like 2k||2k  always connected to charge up the input cap in about 10 seconds, then flip a switch to bypass the input resistance (R22 and R31 in my case).  I don't like that idea because I know I would forget to turn off the bypass switch before connecting to a new DUT!!

Think you could use a Push to Test button which would Set and close the input relay, which would shunt the larger input series resistor with a smaller resistor. Add a high value resistor in parallel with the input, so when the input test source is removed the large voltage across the input cap will drive the input op-amps to rail, then detect such and use this to Reset the relay (also could include a simple timer (CD4060) to time out Reset as a "Just in Case".

Anyway, just a thought on the series input resistor shunt relay operation.

Best,

[Kleinstein]

If a high initial input current is the problem, there is the option to use 2 JFETs and a resistor as a current limiter in bettween instead or parallel to R22+R31. This could give a 2-3 mA current limit with some 500 Ohm of resistance.

With a gain of 200 from the input stage the resistors in the filter part should not give relevant noise, even with 1 µF and 3.3 M. Anyway the resistor part would only be relevant at the lower bad edge not with the full 10 Hz BW.

3.4 pA/sqrt(Hz) would be quite a lot of current noise. The specs are way more optimistic, but the specs on AZ amplifier current noise are sometimes a bit scetchy. The current noise may depend on the circuit (e.g. capacitance directly at the input) and symmetry between the inputs. The current circuit at least looks symmetric for the higher frequencies, which is normally a good thing.

I don't think a higher supply would really help much. Chances are the current noise and input bias may get worse with a high supply.

[trtr6842]

Think you could use a Push to Test button which would Set and close the input relay, which would shunt the larger input series resistor with a smaller resistor. Add a high value resistor in parallel with the input, so when the input test source is removed the large voltage across the input cap will drive the input op-amps to rail, then detect such and use this to Reset the relay (also could include a simple timer (CD4060) to time out Reset as a "Just in Case".

Anyway, just a thought on the series input resistor shunt relay operation.

Best,

I like the "push to test" idea!  I think rather than a timer, simply re-setting the bypass switch whenever the 200x stage hits the rails could work.  Any time I unplug the input it rails, and could be a reliable way to ensure the bypass gets turned off.  It's unfortunate how little circuitry fits in a Hammond 1590B enclosure though!  The dual comparator, flip-flop, and analog switch required to do that would take up almost as much room as the rest of the circuit!!

[mawyatt]

Since all the filtering is after the initial 200x gain stage, noisy components there don't really contribute much to the input-referred noise.  My LTspice sim shows 27.89nV RMS noise with ideal 10µF / 10µF 82k, 330k for the high-pass sallen-key, and 27.915nV RMS  (Both values are input-referred). 

The specific caps are GRM21BR61H106KE43L, but again I think that everything after the 200x stage is essentially trivial. 

My testing with the lower and shorted input resistor is interesting though, since it's suggesting that the opamps might have higher than expected current noise, or that my clamp diodes are somehow introducing current noise...

The 200X front end gain certainly helps and the correct approach to making things "downstream" almost insignificant. Even tho I'd wager that if you lightly tap those X7R caps the output will jump all over the place, even sweeping a heat gun across them will likely induce some response, that's our experience with High K dielectrics in high sensitivity signal paths and why we avoid them.

Of course YMMV.

Thanks for the cap information, that's a muRata 10uF 50V, which is a very good source!! BTW its a X5R and not a X7R!!!

Best,

[trtr6842]

If a high initial input current is the problem, there is the option to use 2 JFETs and a resistor as a current limiter in bettween instead or parallel to R22+R31. This could give a 2-3 mA current limit with some 500 Ohm of resistance.

With a gain of 200 from the input stage the resistors in the filter part should not give relevant noise, even with 1 µF and 3.3 M. Anyway the resistor part would only be relevant at the lower bad edge not with the full 10 Hz BW.

3.4 pA/sqrt(Hz) would be quite a lot of current noise. The specs are way more optimistic, but the specs on AZ amplifier current noise are sometimes a bit scetchy. The current noise may depend on the circuit (e.g. capacitance directly at the input) and symmetry between the inputs. The current circuit at least looks symmetric for the higher frequencies, which is normally a good thing.

I don't think a higher supply would really help much. Chances are the current noise and input bias may get worse with a high supply.

I like the idea of a low-noise current limiter.  I've actually never worked with JFETs, so I'll have to read up on this and get some parts to prototype with.  Any common JFETs parts that you can recommend of the top of your head?  Preferably through hole for easy prototyping, but I've got plenty of SOT-23 breakouts as well.

I agree that something sketchy is going on at the input, and probably related to current noise.  I wonder if the clamp diodes are contributing at all, that's something I could test, at least for a short-circuit input.  Maybe the 10nF feedback caps are causing issues too.  They're not crucial to the overall frequency response, so I can try removing them.

For the supply changes, I wasn't thinking about increasing the supply voltage, I was considering letting an input transient push the supply voltage up temporarily, since most low-noise opamps are good to +36V.  This wouldn't reduce the peak input current on a hot +10V plugin, but with the right clamping circuit only the opamp bypass caps would have to be charged up, so the time constant would be ~20µs instead of 200ms!

[Kleinstein]

For the current limiter one would like low threshold JFETs, like SK209 or SK3557. No need to have low noise, but these are the ones that come with a low threshold.  For a test J113 or similar PN4393 would be candidates in a TO92 case.

[trtr6842]

The 200X front end gain certainly helps and the correct approach to making things "downstream" almost insignificant. Even tho I'd wager that if you lightly tap those X7R caps the output will jump all over the place, even sweeping a heat gun across them will likely induce some response, that's our experience with High K dielectrics in high sensitivity signal paths and why we avoid them.

Of course YMMV.

Thanks for the cap information, that's a muRata 10uF 50V, which is a very good source!! BTW its a X5R and not a X7R!!!

Best,

Since the ceramic high-pass caps seem to be such a concern, I ran this rough test for microphonics.
Board on the left is CH1 on the scope.  The high pass sallen-key parts are 1µF film, 1µF film, 820k, and 3.3M.
Board on the right is CH2 on the scope.  The high pass sallen-key parts are 10µF X5R, 10µF X5R, 82k, and  330k.

Both boards exhibit some microphonics, but from my time using the one I already have in an enclosure once I stop touching the test setup everything stays very consistent.

[trtr6842]

For the current limiter one would like low threshold JFETs, like SK209 or SK3557. No need to have low noise, but these are the ones that come with a low threshold.  For a test J113 or similar PN4393 would be candidates in a TO92 case.

Thank you for the suggestions, looks like Mouser has J113's in stock so I'll be ordering those! So far this is my favorite idea for current limiting so I will be trying it out as soon as I get parts.

[trtr6842]

I tried removing the input clamp diodes D2 and D3, and they offered no improvement when the input resistance was only 100Ω (Cfg#6).  But when the input resistance was 1kΩ noise is better with them removed vs installed (Cfg#7).  In that case σ dropped from 25.79nV to 25.31nV, meaning the diodes contributed about 5nV RMS of input noise, which is significant, but input protection is important.  I think if the JFET current limiter idea works well then I could just rely on the opamp internal clamp diodes safely.

[mawyatt]

Since the ceramic high-pass caps seem to be such a concern, I ran this rough test for microphonics.
Board on the left is CH1 on the scope.  The high pass sallen-key parts are 1µF film, 1µF film, 820k, and 3.3M.
Board on the right is CH2 on the scope.  The high pass sallen-key parts are 10µF X5R, 10µF X5R, 82k, and  330k.

Both boards exhibit some microphonics, but from my time using the one I already have in an enclosure once I stop touching the test setup everything stays very consistent.

Whats the gain from the HP to the DSO, checked the first schematic and it shows just 10X, is that correct? Also is the DSO set to it's highest sensitivity?

We grabbed a Samsung 0805 10uF X7R and lightly tapped it with a plastic tuner stick. Scope is 500uV/Div (187).

Then used a noname 0.1uF 50V Ceramic Disc (don't know dielectric tho), same scope settings (188).

Edit: Here's another with the tuning stick (Blue plastic Bourns type) dropped ~ 1/4" onto the Samsung 10uF 0805 case. Same tests with film and electrolytic capacitors show no response.

Anyway, looks like those muRata caps are quite good indeed

Best,

[trtr6842]

Whats the gain from the HP to the DSO, checked the first schematic and it shows just 10X, is that correct? Also is the DSO set to it's highest sensitivity?

The overall gain from input BNC to output BNC of the amplifier board is 2,000 V/V.  My test setup has no attenuation from amplifier board to scope, just a straight BNC cable.  The scope is set to it's highest sensitivity, which is 1mV per division.  With the 0.0005x attenuation factor that displays 500nV/div, which is true to the amplifier board's input.  The scope is also set to high resolution mode, and the high oversampling ratio bring it's intrinsic noise down to <70nVp-p and <7nV RMS, so the scope is well below the noise floor of the amplifier board.

Your test data in interesting, thanks for sharing!  You actually reminded me of something by calling me out on the X7R X5R mixup... I think I picked X5R caps on purpose!  JLCPCB had 4.7µF 1206 X7R caps, but I had a dilemma: Do I pick the larger more stress-prone package for the better dialectric, or do I pick the smaller package with a potentially worse dialectric?  Given my BNC connectors are only supported by the PCB, I went with the smaller package.  I have no idea if that was the right choice, but at least that was my thought process!

I also tried removing the feedback caps from the first stage amplifier: C6, C12, C16, C22.  There were a few comments about auto-zeroing opamps potentially acting odd with input caps, so I gave it a shot.  With an Rin of 1k the effect was negligible (Cfg#8).  With an Rin of 100Ω there was a slight improvement: σ = 19.45nV down to σ = 18.03nV (Cfg#6 vs Cfg#9).

[mawyatt]

Here's an example of slightly blowing across the Samsung 0805 10uF capacitor with scope set to 500uV/Div as shown.

These are some of the reasons we invested in lab bench type LCR meters to investigate the various behaviors of different capacitors, and frankly surprised how bad the High K dielectric types are, at least the ones we've tested.

Way back about 5 decades ago we had a device fail miserably during classic vibration testing, traced to High K ceramic capacitors in the signal path. This was an expensive lesson, as we had to redesign a few PCBs to accommodate larger film capacitors (back then PCB design was really expensive), since we've been shy of utilizing High K ceramics in any sensitive signal paths. Ironically the C0G/NP0 ceramic types are superb in every respect, even utilized in high resolution DMMs integrator ADCs, where one would think the Keysight/HP and Keithley folks would use some exotic expensive types 

BTW gotta love these nice low noise DSOs, we just need more resolution now, (read creating excuses justifications for a 12 bitter)

Best,

[Kleinstein]

Some of the C0G capacitor are really good, but not all C0G are the same. For the small capacitance range up to some 10 nF they out-perform film type capacitors in most aspects, maybe except leakage.
In my test the good COG where about where PTFE capacitors are claimed to be, but with a much smaller form factor to reduce parasitic capacitance.
Even the not so great C0Gs are about on par with PS and PP capacitor when it comes to loss.

[trtr6842]

BTW gotta love these nice low noise DSOs, we just need more resolution now, (read creating excuses justifications for a 12 bitter)

Mine's a 10 bitter!  It was the same price as the tek scopes in my range but had a lot of cool extra features.


So inspired by the JFET current limiter idea, but too impatient to wait for JFET's to arrive, I managed to come up with a MOSFET input protection circuit that appears to be working really well!
Here is the basic mod:  A series n-ch and p-ch MOSFET pair, each biased to be on when the input signal is near 0V. 



I was able to rework it onto the board fairly easily:




I did some quick functional testing by feeding the input of the amp board with a 19Vp-p sinewave from a signal generator.  I measured the voltage just after the MOSFET pair, and it was clamped well below the ±4.5V supply rails, which is perfect!  There is actually no current flowing besides opamp bias currents, so there is no risk of blowing up the opamps internal protection diodes.



I then turned off 9V to the amp board, and the voltage simply goes to zero, with no increased current draw.



So with the protection features confirmed, I made sure that it is still accurately conducting signals that should be in range.  I fed in a 100µVp-p test signal, and the output looked perfect.



Noise performance with a shorted input is great!  about 140 - 160nVp-p, or 17.92nV RMS (Cfg#13).  This is by far the best input protection, and best noise performance of any protected variant.  17.92nV RMS corresponds to 10nV√Hz per opamp with a 11.3Hz bandwidth (10Hz with a 4th order LP filter), and the datasheet is 9nV√Hz, so I'd say I'm pretty close to optimized for these opamps. 


I actually have a couple LT1037's on order, so I'll have to test those.  It should be significantly lower noise, even just using one, but the current consumption might go up a bit.  I think with a re-design I could use cheap and low-power 5V chopper amps to do all the filtering.  Any half decent chopper won't have 1/f noise, and broadband noise won't be too important after the 200x gain stage.

[Kleinstein]

The later stages don't need super low noise and with a relatively low gain also don't need high speed. So one could get away with a slower non chopper precision OP-amp (e.g. OP202).

The input protection looks unusual, but seems to work real well

[trtr6842]

The later stages don't need super low noise and with a relatively low gain also don't need high speed. So one could get away with a slower non chopper precision OP-amp (e.g. OP202).

The input protection looks unusual, but seems to work real well

Hahaha the OPA2202 happens to be better and way cheaper than the OPA2188's I'm using, I'll have to pick some up for the input stage!  I'll have to pick some up and give them a try.

[mawyatt]

Clever solution 

Best,

[trtr6842]

So I wasn't exactly right about the input protection circuit limiting current...  It does limit input voltage perfectly!  Here's an XY plot with input voltage on the X azxis and  voltage after the limiter on the Y axis:

The asymmetrical clamp limits are due to the very different threshold voltages between the Si2301 and the 2N7000.  The clamp thresholds also follow the supply rails up and down as I varied the supply voltage.  With modern low-threshold MOSFETs clamping should happen about 1.25 - 1.5V from the rails, which is enough to work with a well depleted 9V battery.


However I found out that inputs to the OPA2188 's are all protected by back-to-back diodes, so the input current goes higher than it should.  Here is an XY plot with input voltage on the X axis and input current on the Y axis.  The gentle slope around 0V is from the 2k input impedance (which I moved to after the limiter BTW).  However as soon as the input voltage gets to ±0.6V the input current magnitude increases.

Luckily my circuit had about 1k on the inverting input, so the overall current didn't damage anything.  For this version, its an OK solution.  For a lower-noise version, 1k gain setting resistance would be too noisy, but going lower would increase the input clamp current too much.

Now the OPA2182 does NOT have input clamp diodes, and has about 1/2 the noise as the OPA2188.  I'm going to try that one next.  Four sections of the OPA2182's should get into LT1037 territory as far as noise performance, cost about the same in total, and draw a little bit more current.  However from what I have found the OPA2182 is one of the only candidates that doesn't have back to back input diodes, so my next revision of the board will include an optional JFET current limiting circuit.

[trtr6842]

I got some OPA2182's in yesterday and swapped them in for the 1st stage amplifier.  I changed the input gain resistors to 120Ω and 30k for a gain of 250x, and I increased the gain of the second stage to 20x for an overall gain of 5,000x.

The OPA2182's have "MUX Friendly" inputs without the back to back clamping diodes, so now my MOSFET input voltage limiter provides adequate protection, and the input current is only due to the 2kΩ input resistor to ground.  This is the XY plot of input current vs input voltage.  For this test the input 2200F cap was shorted.



Noise with a shorted input is much better, 91nVp-p typical!  Current draw is up to 7.5mA, but that should still be good for ~60 hours on a 9V battery, which should be good given how quickly this design settles on powerup.

[Andreas]

Hello,

I typically measure the noise floor with (new) 8xNiMH cells. (~10V bias)
The leakage current of the input capacitor (the AC part of it) also contributes to the noise floor.
It also needs ~1-2 days until the leakage current is low enough for doing measurements.
So when not used I connect a 9V block to the input to keep some bias voltage on the input capacitor.

with best regards

Andreas

[trtr6842]

From your experience, how noisy are NiMH cells?  I don't have any on hand but might have to get some.

For this design, DC leakage isn't too important due to the two-stage design.  The 1st stage can tolerate ±7µA of leakage and not saturate it's output, and the first stage is immediately followed by a high-pass filter, cancelling out any DC offset before the remaining gain is applied.  Now if the leakage currents have any noise in the 0.1 to 10Hz range, then they'll get multiplied by 2kΩ and the voltage gain of 5,000x, but I have no data on the noise component of electrolytic leakage currents.

[MasterT]

The same rules for electrolytic capacitors's noise as for any semiconductor : higher absolute DC leakage would proportionaly generate higher AC low freq. noise level.
 I get very good results with this 2 caps:

UKL0J102MPD
RNL1C152MDS1

[Kleinstein]

How noisy the leakage current is, is a good question. As a minimum there should be resistor like noise, but the actual noise is likely higher. It is reasonable to expect shot noise, like with semiconductors or tunneling junctions. So current noise from randomly coming electrons and thus proportional to the square root of the current.

There is also the chance to get added current spikes from chemical reactions in the formation of the oxide layer, e.g. if a weak spot appears (weak inhibitor moves) and gets repaired (extra oxide formed).

For the resulting noise voltage it is no only the 2 K resistor to ground that is relevant, but mainly the capacitance itself. In this design the input cut off lower than 0.1 Hz and thus little effect of the resistor.

Batteries are normally considered rather low in noise. They however reactor to mechanical stress and thermal effects. So they should have some time at rest and thermal shielding. NiMH may be attractive because of a relatively stable voltage with discharge state and low internal resistance. 

To some degree electrolytic capacitors can behave similar and also react to stress.

[Andreas]

From your experience, how noisy are NiMH cells?

Depends.
Of course you have to do some pre-cautions.
Keep the cells at constant temperature (so do not use immediately after charging).
Old (defective) cells have some tens of uV.
New (good) cells do not contribute much to my noise floor of my LNA which is < 0.2uVpp.
Part of it is due to leakage current of the input capacitor.

But noise floor can significantly increase if the input capacitor has too much leakage current:
https://www.eevblog.com/forum/metrology/lowest-noise-op-amps-for-low-frequency-low-level-ac-coupled-signals/msg2218644/#msg2218644

with best regards

Andreas

[mawyatt]

From your experience, how noisy are NiMH cells?

Depends.
Of course you have to do some pre-cautions.
Keep the cells at constant temperature (so do not use immediately after charging).
Old (defective) cells have some tens of uV.
New (good) cells do not contribute much to my noise floor of my LNA which is < 0.2uVpp.
Part of it is due to leakage current of the input capacitor.

But noise floor can significantly increase if the input capacitor has too much leakage current:
https://www.eevblog.com/forum/metrology/lowest-noise-op-amps-for-low-frequency-low-level-ac-coupled-signals/msg2218644/#msg2218644

with best regards

Andreas

Wonder if one could use the increased noise level as a quick test of a NiMH cell, rather than having to do a discharge profile? Should be much quicker to evaluate and might even work on cells in a stack, test individual cell within unloaded stack.

Best,

[Gerhard_dk4xp]

From your experience, how noisy are NiMH cells?  I don't have any on hand but might have to get some.

<   http://www.hoffmann-hochfrequenz.de/downloads/NoiseMeasurementsOnChemicalBatteries.pdf       >

in short: the bigger the better.

Cheers, Gerhard

[trtr6842]

I ran a test using two 18650 LiPo cells.  I discharged them down to 4V each to get closer to the flat portion of their discharge curve.  I probed the output of the first amplifier stage so I could see how large the leakage current through the cap is.  This cap is prettymuch new and I did only a few minutes of aging before this test.

The basic setup:  A small cardboard box to limit airflow on the batteries, but this whole setup was just on my desk as I was still using the computer.



Here are the results.  I'm paying attention to the low values between the large disturbances.  The first stable time was starting at t = 3 hours where I left my desk and everything was still in the room.


Overall I don't think this design is too picky about cap leakage.  The amplifier was able to settle down even with as much as 75nA and only 3 hours under bias, and useable for noisier measurements like bandgap and some buried zener references with way less time under bias.

[Andreas]

Hmm,

I am missing a cookies box (see LT AN124) or
a steel paint can (see TI avt_081307)

with best regards

Andreas

[Gerhard_dk4xp]

I built some amplifiers in the usual way: CS stage with many/large JFETs, cascode
against Miller effect, OpAmp and feedback to the 1 Ohm source resistor.
Unless you are happy with audio bandwidth or have a GHz opamp, that thing
develops negative input resistance as soon as the feedback loop is closed.
An inductive signal source with the right L, and you have an oscillator.
I also checked designs from others who swore that would be stable. It was not.

I finally gave in and dismissed the feedback. I enforced the source current by
using a current mirror. Then the source had to be decoupled by LARGE caps
to get common source, ac-wise.

It turned out that every electron that defects through the electrolytic produces
a noise pulse at the amplifier output.  OSCONS were worst, 1/f corner in the KHz
and levels at a few Hz that the 89441A FFT analyzer had dynamic range problems.

Nippon Chem. alu electrolytics were much better but still not usable.
The only thing that borderline worked was an AVX wet slug tantalum 4700uF/25V,
€/$ 100 a pop. Vishay even wants 3 dB more money.

It seems, that circuit is the optimum cap noise current detector.

JFETs have a pos and a neg TC that may cancel.
At least in simulation, 16 pcs. CPH3910 have their thermal sweet spot at 45 mA total.
That allows simply grounding the sources.
The new TI FETs seem to be best at IDss, too much current to be practical
if you want 16 pcs in par. for noise reasons.

Gerhard
 

[trtr6842]

Hmm,

I am missing a cookies box (see LT AN124) or
a steel paint can (see TI avt_081307)

with best regards

Andreas

Do Hammond diecast enclosures count?  Or are only cookie tins or paint cans good at scaring away the noise?



I couldn't easily probe the output of the first stage with the board in the enclosure, and that's how I measured my leakage current.

[MegaVolt]

Or are only cookie tins or paint cans good at scaring away the noise?

Somewhere I came across an explanation of why tin cans are so good. They have 3 layers of tin-iron-tin. The result is better shielding than a material of the same thickness but made from the same material.

[trtr6842]

Or are only cookie tins or paint cans good at scaring away the noise?

Somewhere I came across an explanation of why tin cans are so good. They have 3 layers of tin-iron-tin. The result is better shielding than a material of the same thickness but made from the same material.

Hahaha I was joking about that... I doubt the material really matters much this close to DC, thermal and air current effects are probably way worse than any sort of EMI issues here.

[Andreas]

Do Hammond diecast enclosures count?  Or are only cookie tins or paint cans good at scaring away the noise?

The enclosure is not bad. But how do you get your DUT (voltage reference) and Power supply within your (shielded) Hammond case?
By the way: If my 34401A is too close to the amplifier (< 0.5m). I get heavy mains hum on my output signal. (good visible with a FFT)

A tin can has advantages against magnetic fields at low frequencies. (but of course also a thick walled case gives some shielding).

with best regards

Andreas

[MasterT]

Or are only cookie tins or paint cans good at scaring away the noise?

Somewhere I came across an explanation of why tin cans are so good. They have 3 layers of tin-iron-tin. The result is better shielding than a material of the same thickness but made from the same material.

Hahaha I was joking about that... I doubt the material really matters much this close to DC, thermal and air current effects are probably way worse than any sort of EMI issues here.

 RF is interfering with my noise test ,  Wi-Fi - easily getting into since packets transmission about 10 Hz - demodulated  on any P-N junction with 2.5 cm piece of wire attached to.  Resistors on a breadboard about same size. Worst OPA is LM4562:

[trtr6842]

The enclosure is not bad. But how do you get your DUT (voltage reference) and Power supply within your (shielded) Hammond case?
By the way: If my 34401A is too close to the amplifier (< 0.5m). I get heavy mains hum on my output signal. (good visible with a FFT)

A tin can has advantages against magnetic fields at low frequencies. (but of course also a thick walled case gives some shielding).

with best regards

Andreas

That's why I kept my amplifier size as small as possible, it can be put in whatever larger enclosure makes sense for the DUT.  Also, for a low-impedance DUT output I don't know how important it is for both amp and DUT to be in the same enclosure.  As long as DUT and amp are each stable I think their noise will dominate noise added by a halfway decent connection.

For the 60Hz interference, what order low-pass filter do you have on your amplifier?
But even if these amps don't reject 60Hz, as long as it doesn't inter-modulate down to the 0.1to 10Hz range it would not be hard to remove the 60Hz spike from the FFT and then run an IFFT to correct the time domain signal, if it even contributes significantly to the overall noise.  Or just applying a digital notch filter would achieve the same effect.

[Andreas]

Hello,

I have a 4th order low pass.
In my case I have 50 Hz mains frequency.
Before doing some IFFT I rather put my cookies box to a quiet place.

with best regards

Andreas

[trtr6842]

I Just got Rev X2 of this LNA built up:




The performance is better than the modified X1 revision a typical 78nVp-p and 10.5nV RMS shorted input noise floor.



The updates schematic is attached.  It looks different to fit on US Letter sheets, but is pretty similar to the modified X1 configuration.  I added a BNC connector for the DC output of the first stage to monitor input capacitor leakage.  There is also a status LED the lights up when that signal is at least ~V away from the both supply rails, indicating that the first stage is not railed.

I'm very happy with the electrical performance, but the mechanical form factor isn't great.  I had to over-drill the holes for the BNC connectors because otherwise the board would not fit into the box.  I also put the switch on the wrong side of the board, requiring longer battery wires.  An extruded enclosure with PCB slot would be nicer, but I already had these hammond diecast boxes on hand, so I figured I'd work around those.

[iMo]

Why not U1B (stage 2) with 40x gain such you'll get 10000 in total?

[Kleinstein]

There is little need for even more gain and with more gain one may run into clipping earlier, e.g. with a more noisy source. Already the gain of 5000 often higher than needed.

[Andreas]

Already the gain of 5000 often higher than needed.

Oh come on.
A modern wideband oscilloscope has ~1mVpp input noise.
To measure a serious reference like ADR1000 you really want a higher amplification to get below 200uVpp noise floor.

@trtr6842
I have seen that you are using "hires mode" with a relatively low (10kS/s) sample rate.

what exactly does this software filter do? (I am not familiar with R&S)
I hope it does a sliding average over neighbouring measurement points and not average complete traces.
where is the Edge frequency of this filter?
Have you done a FFT?
when averaging 256 neighboured points the sample rate would lead to around 40 Hz which is too close to the 10 Hz Bandwidth.

with best regards

Andreas

[trtr6842]

Already the gain of 5000 often higher than needed.

Oh come on.
A modern wideband oscilloscope has ~1mVpp input noise.
To measure a serious reference like ADR1000 you really want a higher amplification to get below 200uVpp noise floor.

@trtr6842
I have seen that you are using "hires mode" with a relatively low (10kS/s) sample rate.

what exactly does this software filter do? (I am not familiar with R&S)
I hope it does a sliding average over neighbouring measurement points and not average complete traces.
where is the Edge frequency of this filter?
Have you done a FFT?
when averaging 256 neighboured points the sample rate would lead to around 40 Hz which is too close to the 10 Hz Bandwidth.

with best regards

Andreas

Here are some specs for my RTB2004:
I left the probe settings according to a gain of 5000, the same as all my tests with the LNA.

Shorted input, 20Mpts capture, 10.9kS/s, Hi-res mode:
23nVp-p, 2.54nVrms (115µVp-p, 12.7µVrms 1x gain equivalent)

Shorted input, 20Mpts capture, 1.67MS/s, Sample-mode:
136nVp-p 16.1nVrms (680µVp-p, 80.5µVrms 1x gain equivalent)

So based on my last test where the LNA showed 13.2nVrms of noise, if you remove the 2.54nVrms of scope noise, the LNA actually has 12.95nVrms of input-referred noise.  Increasing the gain to 10kx would give a shorted input noise of 13.01nVrms, which is not a significant improvement for me.  Other scopes may vary though, but since my scope has such low intrinsic noise I'd rather have a wider measurement range.  With an old 9V battery (drained to 6V) I can measure signals up to about 1mVp-p.  Since I can't do that with scope naturally, that's pretty useful.  (Although this particular version has the 250x output as well, drastically increasing the range).

Essentially I get no benefits from a gain of 10,000x, but that's just my setup, others may vary.


As for sample mode vs hi-res, there is absolutely no concern for bandwidth.  The hi-res mode performs a moving average, and is reflected in the sample rate.  You can see from the two measurements above the oversampling ratio is about 153 (1.67MS/s vs 10.9kS/s).  In both cases the ADC is running at 1.67MS/s, but the decimated datarate of the hi-res capture is 10.9kS/s, 1000x higher than the 10Hz range we need.

Here is a basic amplitude test with the scope in sample mode:


Here is Hi-res mode, with an even lower sample rate than my tests.



5.0185V Mean Amplitude in Sample,
5.0144V Mean Amplitude in Hi-res
Only a -0.0071dB drop, perfectly acceptable given the benefits of lower input noise.


Back to the LNA, Here is a 131kpt FFT of a 60-second capture with a shorted LNA input:

You can clearly see the 10Hz LNA high-frequency cutoff, and then below that the noise floor of the scope beyond that.

Here is a 20Mpt FFT of a 240 second capture with shorted LNA input:

[2N3055]

Already the gain of 5000 often higher than needed.

Oh come on.
A modern wideband oscilloscope has ~1mVpp input noise.
To measure a serious reference like ADR1000 you really want a higher amplification to get below 200uVpp noise floor.

@trtr6842
I have seen that you are using "hires mode" with a relatively low (10kS/s) sample rate.

what exactly does this software filter do? (I am not familiar with R&S)
I hope it does a sliding average over neighbouring measurement points and not average complete traces.
where is the Edge frequency of this filter?
Have you done a FFT?
when averaging 256 neighboured points the sample rate would lead to around 40 Hz which is too close to the 10 Hz Bandwidth.

with best regards

Andreas

Even a 1mV P-P scope noise gets to be attenuated to 1µV P-P effective with 1000X preamp.

For my SDS2000X HD  with setup for 1000X amp I get  140-150 nV P-P  and 14-15 nV RMS effective scope noise. 2MS/s
Picoscope 4262 with setup for 1000X amp I get  23 nV P-P  and 3 nV RMS effective scope noise. 1MS/s

+-8µV full screen for Siglent and +-10µV full screen for Picoscope. Siglent can go down to 500 nV/div (+-2uV full screen), but noise stays the same.

That is on time base that captures 10s worth of data making sure it captures low frequency noise of scopes. On Siglent ERES 3 is on (lowpass filter in effect, i cannot be bothered to calculate frequency now sorry, but some kHz..) and so is on Picoscope (200Hz).

With good modern scopes with low noise inputs 1000X amp is more that enough.

[Andreas]

  The hi-res mode performs a moving average, and is reflected in the sample rate.  You can see from the two measurements above the oversampling ratio is about 153 (1.67MS/s vs 10.9kS/s).  In both cases the ADC is running at 1.67MS/s, but the decimated datarate of the hi-res capture is 10.9kS/s, 1000x higher than the 10Hz range we need.

Hello,

I was not aware that the instrument shows the "effective" sample rate instead of the raw value.
So your FFT pictures show that the LNA determines the frequency response.

with best regards

Andreas

[trtr6842]

Hello,

I was not aware that the instrument shows the "effective" sample rate instead of the raw value.
So your FFT pictures show that the LNA determines the frequency response.

with best regards

Andreas

To be fair, I wasn't aware of that either until I did those tests! 
Also the hi-res mode doesn't perform a moving average, it averages N samples and then treats the average value as a single data point.  I guess that is essentially equivalent to an N moving average followed by a decimation by a factor of N though... Either way, just wanted to clear that up.

For the FFT's, the 131kPt one uses the "display data", so it has the hi-res mode factored in, but with an even lower sample rate than shown by the capture.  It does show that the scope still has plenty of bandwidth for 10Hz though.

The 20MPt one uses the acquisition data, and that is the raw acquisition data with no hi-res averaging or decimation applied.

[Sariel]

Hey,

This is a very nice low noise amplifier.
Well done!

Could you please share the P/N of the enclosure?
And also what is the connector type of the 9V battery on the board (I am not familiar with it).

Thanks,

[AnalogTodd]

For those interested as to why a cookie tin or paint can is so often used for shielding, I would recommend picking up a copy of "Noise Reduction Techniques in Electronic Systems" by Henry W. Ott. Chapter six is 'Shielding Effectiveness of Metallic Sheets' and covers why one would use a particular material for shielding.

Basic idea has to do with frequency you are trying to shield against will determine the type of material you want. For high frequencies, copper and aluminum enclosures work very well. At low frequencies, you want to provide a low reluctance magnetic shunt path to divert fields around the circuit being protected. A high permeability material is optimal (MuMetal is a good example), steel (or tin) will work better at low frequencies as it does well magnetically. Using multiple enclosures spaced a distance apart can provide better results than a single thicker enclosure.

[thermistor-guy]

...A high permeability material is optimal (MuMetal is a good example), steel (or tin) will work better at low frequencies as it does well magnetically...

Is there any benefit in wrapping a steel can with copper tape?

[Andreas]

Hello,

shielding is mainly related to skin effect.
the skin  depth is proportional to square root of the product of conductivity times permeability.
So copper has a factor 10 better conductivity than a tin plated steel.
But permeability is only 1 against 50 - 1000 of steel.

So as end effect: you need a factor 3-10 thicker copper tape than your steel can to get the same shielding effect at the same frequency.

with best regards

Andreas

[iMo]

This is the "portable" box I designed long time back (1/f noise measurements).. Inside we had a second shielding layer made of copper foil.
https://www.eevblog.com/forum/metrology/a-portable-box-for-1f-and-low-noise-measurements/msg4652194/#msg4652194

[svetlov]

if possible, try an opamp in this circuit OPA209 series (OPA2209)
 or more quiet OPA2211
interesting to see what will happen

[AnalogTodd]

This is the "portable" box I designed long time back (1/f noise measurements).. Inside we had a second shielding layer made of copper foil.
https://www.eevblog.com/forum/metrology/a-portable-box-for-1f-and-low-noise-measurements/msg4652194/#msg4652194

That sort of box was something I was looking at when doing low noise measurements. Thick steel to really try and squash the low frequency stuff and inside some copper to help against high frequency stuff. Fortunately, I was able to get MuMetal in 50mil thickness to create a box for shielding. That was set inside a box made from double-sided copper, all nestled in a cookie tin. A lot better in terms of weight and 'portability'. Even with it all, you still get some line frequency stuff getting in the measurement, but it's a huge improvement over no shielding.

Page eight shows what my box looks like: https://www.analog.com/media/en/technical-documentation/app-notes/an-159.pdf.

[thermistor-guy]

Fortunately, I was able to get MuMetal in 50mil thickness to create a box for shielding. That was set inside a box made from double-sided copper, all nestled in a cookie tin. A lot better in terms of weight and 'portability'. Even with it all, you still get some line frequency stuff getting in the measurement, but it's a huge improvement over no shielding.

Page eight shows what my box looks like: https://www.analog.com/media/en/technical-documentation/app-notes/an-159.pdf.

Presumably, the tin's lid is a loose fit, with some gaps, so there will be EMI leaking through the tin-lid interface. OTOH, the lid is quick to remove, which is handy.

Can any simple modification reduce the EMI leakage, while keeping the convenience of quick release? Apply conductive tape, for example, to make the lid fit a little tighter?
(Reminds me of copper gaskets on GTEM cell access doors).

[Kleinstein]

Some of the cooky boxes have a relatively tight seal. Mainly the paint on the outside that is not ideal and may not give an electric contact for the lid.
If really critical one could likely solder spingy wires to the inside to make a few extra contacts.

For testing with a bread board or similar it is hand to have the bock upside down, so that one has easy acess from all sides and can add the cover if needed. The more normal way around is more suitable for mounting connectors, though a bit wobbely with the thin metal.

[EC8010]

Biscuit and baccy tins have a good seal but often have an insulating coating. That caught me out when I made an electrometer in a baccy tin and had buckets of hum. Soldering a 50mm multi-strand wire from lid to tin solved that (electrostatic) problem.

Tins are usually 0.1mm thick steel, so they provide electromagnetic screening >200kHz (penetration depth as mentioned earlier). You have to keep them away from 50Hz stuff. My bench has a steel frame, so I put a few neodymium magnets on the front girder and stick amplifiers to that, effectively putting 2mm of steel in the way of hum. And (more importantly) distance.

A gain of 5000 is about right to overcome oscilloscope self-noise. Oscilloscopes are designed for bandwidth and that means low capacitance FETs, but low 1/f noise requires paralleled devices. In short, oscilloscopes produce significant 1/f noise. You need the amplified noise of your LNA to be more than three times oscilloscope noise at the frequency of interest for oscilloscope noise to be negligible, and that's what leads to needing a gain of 5000. But with that much gain, you have lots of high frequency noise to overload the oscilloscope, so you also need an LC low-pass filter (RM cored inductor and polypropylene capacitor work well, plus damping resistor to prevent the filter peaking at its cut-off). I use 1kHz for general stuff with a x100 amplifier, but 100Hz for really low noise stuff with x5000. 100Hz is still high enough to be able to measure white noise and 1/f noise on the same FFT. If the 'scope can implement a low-pass filter as part of its decimation, that helps in taming the record length. I export the FFT to a spreadsheet for analysis.

[iMo]

Getting rid of 50Hz hum is extremely difficult, and we had a lot of troubles with it in my 1/f days.
Mind the "penetration depth" is the depth where the 50Hz is attenuated only to 37% (1/e) of its outside value.
For copper the 50Hz penetration depth is ~10mm and for steel/iron ~60-80mm.

[mawyatt]

This is where MuMetal can help, recall we had various thickness sheets of this in the labs way back when and made kludge type boxes for sensitive electronics & measurements.

Also, for sensitive measurements we moved everything away from the test bench and only had the absolute necessary equipment at hand, often powering things with batteries. This wasn't only for very low 1/f measurements, as very low phase noise measurements at higher frequencies required the same attention.

Best,

[mawyatt]

This is the "portable" box I designed long time back (1/f noise measurements).. Inside we had a second shielding layer made of copper foil.
https://www.eevblog.com/forum/metrology/a-portable-box-for-1f-and-low-noise-measurements/msg4652194/#msg4652194

Nice, portable as in get out the forklift as it looks like a tank

Best,

[AnalogTodd]

Getting rid of 50Hz hum is extremely difficult, and we had a lot of troubles with it in my 1/f days.
Mind the "penetration depth" is the depth where the 50Hz is attenuated only to 37% (1/e) of its outside value.
For copper the 50Hz penetration depth is ~10mm and for steel/iron ~60-80mm.

Yes, but there's a lot more to look at than just "penetration depth" when it comes to attenuating fields. Realistically, for a lot of noise measurements you are worried about AC fields as DC fields don't add to your measured signal. Now you need to think about absorption loss and reflected loss and how those change as a function of frequency. Add in there that you can use low reluctance magnetic materials to divert magnetic fields around a region instead of attenuating or reflecting, and you get a myriad of things to review depending on what you are trying to shield against. Because of their "penetration depth" copper and aluminum are great for absorption loss at high frequencies, but are less effective against low frequencies. For low frequencies, you can get good low reluctance materials like MuMetal but the permeability starts dropping at ~1kHz and matches copper/aluminum at ~100kHz.

This is where MuMetal can help, recall we had various thickness sheets of this in the labs way back when and made kludge type boxes for sensitive electronics & measurements.

Also, for sensitive measurements we moved everything away from the test bench and only had the absolute necessary equipment at hand, often powering things with batteries. This wasn't only for very low 1/f measurements, as very low phase noise measurements at higher frequencies required the same attention.

Best,

MuMetal is definitely good for low frequency shielding. Set up a transmitter and receiver 0.1in apart and put a shielding metal in between. At 1kHz, 30mils of MuMetal gives ~18dB of attenuation where the same thickness of copper is ~4dB. At 10kHz, the MuMetal is ~25dB compared to copper at ~20dB. But when you get to 100kHz the 30mils of MuMetal is ~29dB attenuation and it only takes 10mils of copper to give the same attenuation. Hence the multiple layers of different materials that can be used for a wide frequency range shield setup.

This doesn't even touch on how additional distance also attenuates or concerns about material saturation.

[TimFox]

Be a little careful when using Mumetal^TM and other high-permeability nickel alloys for magnetic shielding.
This is a matter of quantitative design.
Permeable metal does not absorb external fields, but re-directs them around the shielded interior.
As a very rough hand-waving calculation, to avoid saturating the magnetic material with the Earth's DC field, consider a rectangular box of finite thickness, oriented for convenience with the direction of the external field perpendicular to the face of the box. 
The total flux that hits that face is the external field multiplied by the area of the face.
Going around the box, that flux is squeezed down into an area given by the perimeter of the face times the thickness of the metal.
That flux density B in the bottleneck formed by the thickness must be below the saturation of the material, which for nickel alloys such as Mumetal is roughly 7,500 to 8,000 Gauss.
Iron alloys have less permeability, but higher saturation, roughly 20,000 Gauss.
Using CGS (Gaussian) units, the magnetic field H just inside the wall is the value (in Oersted) corresponding to that flux density B in the metal, which for the linear approximation is B/mu for non-saturated metal. 
In those units, the flux density in air inside the box is B = H.
For critical applications, it is common to use an iron alloy for an outer layer, with a nickel alloy inside that to achieve extremely low fields.
If the metal saturates on the DC field, it will look like a nonmagnetic layer, with only eddy-current shielding (and relatively high resistance).
In a good design, the magnification of the external field due to the geometry will be a smaller factor than is the reduction in the field due to the high permeability, and the shielding will be useful.

[Kleinstein]

Much of the reduction in the permeability of Mu metal with higher frequencies is due to eddy currents. This effect is already included in the usual model for the shielding thickness. The relevant permeability is the one wihtout the eddy current effect and thus less drop to higher frequency for actual permeability without the eddy current effect.

Mu metal is relatively expensive and need care handling: bend it too much and the permeability goes down quite a bit from the high ideal values. For a larger system the normal electrical steel (Fe-3%Si) could be an attractive option too, if magnetic shielding is needed.

[EC8010]

Mumetal is good in theory, until you discover its cost and practical problems. Distance is best for electromagnetic hum. Goes down by between square and cube of distance. Don't forget alignment, either. Worst is when coils are aligned, best when they're at 90 degrees, and that applies to PCB tracks. The attached was obtained by 3ft distance between device under test (breadboard in shortbread tin) and nearest powered mains transformer. The roll-off after 50Hz is due to the 100Hz low-pass LC filter preceding the oscilloscope. The other low frequency peaks are presumed to be aliases between 50Hz and other oscilloscope clocks (60Hz display, etc).

[mawyatt]

Be a little careful when using Mumetal^TM and other high-permeability nickel alloys for magnetic shielding.
This is a matter of quantitative design.
Permeable metal does not absorb external fields, but re-directs them around the shielded interior.
As a very rough hand-waving calculation, to avoid saturating the magnetic material with the Earth's DC field, consider a rectangular box of finite thickness, oriented for convenience with the direction of the external field perpendicular to the face of the box. 
The total flux that hits that face is the external field multiplied by the area of the face.
Going around the box, that flux is squeezed down into an area given by the perimeter of the face times the thickness of the metal.
That flux density B in the bottleneck formed by the thickness must be below the saturation of the material, which for nickel alloys such as Mumetal is roughly 7,500 to 8,000 Gauss.
Iron alloys have less permeability, but higher saturation, roughly 20,000 Gauss.
Using CGS (Gaussian) units, the magnetic field H just inside the wall is the value (in Oersted) corresponding to that flux density B in the metal, which for the linear approximation is B/mu for non-saturated metal. 
In those units, the flux density in air inside the box is B = H.
For critical applications, it is common to use an iron alloy for an outer layer, with a nickel alloy inside that to achieve extremely low fields.
If the metal saturates on the DC field, it will look like a nonmagnetic layer, with only eddy-current shielding (and relatively high resistance).
In a good design, the magnification of the external field due to the geometry will be a smaller factor than is the reduction in the field due to the high permeability, and the shielding will be useful.

Yes earth's fields can be problematic or useful depending on ones needs!!

We utilized the earths natural magnetic field to detect intruders with a special designed cable and signal processor back in 70s. The long 100M cables had a special core of select metals to map the earths localized magnetic fields into uniform pattern. Windings around the core were arranged and flipped in phase every ~1M to help with far field rejection. The cores were also selected and enhanced for a magnetostrictive effect, so magnetically "clean" intruders would produce a minute stress change in the surrounding ground which induced a minute (nanovolt levels) signal using the earths fields as a "permanent magnet" and the cable magnetostrictive effect would change under the minute stress changes, inducing tiny voltages across the windings. This worked beautifully, and we could detect and characterize various intruders, even "clean" intruders crawling less than 100mm/minute, with the Signal Processor.

Interestingly the cables actually enhanced the effects that MuMetal tries to minimize, that being the change in magnetic properties due to induced stress.

The cables were called MILES, the original signal processor MAID, and later version MSP. Honeywell also produced a commercial version called BLISS 1000 BLS-1000.

Interesting story behind the MILES cable and MSP development if folks are interested.

Best,

[MegaVolt]

Interesting story behind the MILES cable and MSP development if folks are interested.

It is very interesting.  Tell me, please.

[mawyatt]

Interesting story behind the MILES cable and MSP development if folks are interested.

It is very interesting.  Tell me, please.

Well the early cable development is second hand as we weren't part of this, and likely transpired in the late 60s. The cable was originally developed to detect magnetic type intruders, personnel with weapons, tanks, cars, and vehicles which it did, but the developers noticed that it could detect magnetically "clean" intruders and realized this was due to the cable's magnetostrictive effect. They set about to enhance this feature using proprietary cable cores which became the MILES cable and develop a more sensitive signal processor/detector which became the MAID.

The MAID Signal Processor worked well but did have issues with false alarms, especially during storms, and a replacement development program was started. We got involved in 76~77 as a consultant, later hired and developed the algorithms and circuits, a colleague conceived the unique preamp idea discussed later. The requirements were difficult as the entire signal processor needed to fit in a 150mm cube, have built in batter backup, consume less than 50mw and pass full military requirements including conducted and radiated susceptibility. The program we had was known to Sandia Labs and a group of researchers from major universities working with a new algorithm based around a Adaptive Recursive Filter (ARF), they were also working on a program but we didn't initially know, and they were given all our reports and design details. If you've ever worked on USG programs you'll recognize this as we were supposed to fail and Sandia Labs would be the design of choice for the future production.

Early in the design we realized/discovered to achieve the enormous dynamic range required that by pushing the signal integration right up to the front end was the best possible architecture and this also compensated for the natural rate sensitivity of the magnetic sensor which was proportional to rate of change of the earth's magnetic field which is locally disturbed by a magnetic type intruder, and also by the rate of change of ground pressure due to the magnetostrictive cable effect. Since the cable impedance was very low, pushing an integrator to the input requires a very large feedback capacitance, we used a large wet-slug tantalum selected for low leakage. The amp inside the integrator was a low noise selected dual transistor from National and a selected low power op-amp, and overall called a Rate Compensated Preamp. The transistor was biased for optimum low frequency noise and the base bias current was allowed to flow thru the cable, this created a means to detect a cable disconnect, degradation or cut. BTW this push the integrator right upon the input signal was used decades later at RF/MW in the PolyPhase Mixer which is one reason why it has such good NF & DR. After the integrated signal various frequency bands were selected by active filters with various characteristics, and the energy within these bands and various ratios used to detect and classify target intrusions.

We sent some folks out to collect MILES cable data from various USG facilities around NA to get typical signatures for different terrains, nearby railroads, various aircraft flying end of runway, sonic booms, thunder storms, winds, frozen ground and so on, even setting up a special field site at our facility for long term recordings (false alarms).

Some false alarms did show up and always about 6am. The field had signs to Keep Out, so we set up a camera to be triggered by the alarm and discovered a guy walking his dog across one of the cables. Left him a friendly reminder, and no more alarms!! Local HS kids were hired to try and run, jump, crawl, pole vault, anything they could do to get past the cable, even showed them where the cables were buried and they couldn't defeat the system!!

We were informed about Sandia and the various university efforts based upon the Adaptive Recurrsive Filter just before our 1st field test at Eglan AFB, Florida during "Smoke Week". Sandia Labs showed up with a large Grumman van (like UPS uses), with a diesel generator running a VAX 11780 and a large Igloo size metal box, we had our little analog 150mm cube box, and a couple chart recorders!!!

When asked about the size and power requirements we were imposed with, Sandia folks said they'll shrink everything into a series of custom chips and such...sure!!!

We beat or equalled every test parameter during the test, most importantly the probability of detection and false alarm rates were much better overall and the Sandia/University folks were not happy!!

Later the 2nd test was conducted at Rome AFB in late January for the frozen ground tests, same result. Honestly believe the single most important reason we were successfully was the Rate Compensated Preamp, which was the integrator right at the front end. The system had 146dBV gain available and the ratio of various energy band levels acts as a dynamic AGC giving the system the best possible chance of detecting an intruder while keep the false alarm rate very low. The Sandia & Universities ARF was very clever, but without the advantages of our preamp, they were at a signal disadvantage right from the start.

Anyway, we equaled or beat the ARF in every measurement category with our little analog 150mm cube with the Rate Compensated Preamp, and later this system was selected for production and replaced the MAIDs in the field 

Sorry for the long note, maybe the moderators want to move this, unless its OK with other folks.

Best,

[mawyatt]

Google doesn't show much, but did find some references:

References Honeywell BLS-1000 (Commercial version of MILES cable and MAID (later MSP) Processor)
https://apps.dtic.mil/sti/pdfs/ADA185001.pdf

MILES/MAID References
https://apps.dtic.mil/sti/citations/ADA056703
https://ieeexplore.ieee.org/abstract/document/6393530

Automated system to characterize the MILES cables we developed
https://apps.dtic.mil/sti/citations/ADA092200

Best,

[svetlov]

in this video disassembly of a similar module - see the design

[trtr6842]

I watched this video when it came out, just after I finished the first couple rounds of testing for my LNA!

The DC servo circuit is interesting, very cool that it successfully ignores input cap leakage while maintaining full input range.

One thing stressed in that video is the high pass filter response.  The 1K 1mF input RC gives a 1 second time constant filter with a -3dB point of about 0.16Hz and a first order roll-off.

The high pass filter in my amp is 0.1Hz but 3rd order, plus another 1st order roll off at 0.036Hz.  im thinking I should modify my high pass filter to be 0.16Hz 1st order (simply adjust the RC and remove the sallen key feedback) and see how the noise figure changes.  I can fairly easily decrease the cutoff frequency for the input and output to at least 0.01Hz without increasing resistor noise I think.

[Kleinstein]

The 1 second time constant is a bit questionable. The interaction with the DC servo may actually also effect the cross over. So the actual high pass for the input stage is 2 nd order. The transition also looks relatively steep - more than just 2 x 1st order combined, but more like a proper Butterworth or similar filter. The DC servo is a nice idea, but the way it is build is not very good.

A design with a longer time constant at the front and the fine filtering later should have a little advantage noise wise, as there is less noise from the transition region. So the 0.036 Hz from the front and 0.1 Hz later should be the better solution.  Having the high pass filter to cut away more of the single is more like cheating, not improving on the noise performance. Especially with modern instruments there is also the possibilty to do the actual filtering for the intervall of interest only in the digital domain (FFT or DSP based filtering) and have the analog hardware a little wider bandwidth on both ends.

[julian1]

Would it make sense to stack/move all gain (200x + 10x) before the high-pass filters to reduce the potential influence of the non-c0g/film 10u caps?.

This could be done by using a sallen-key as DC block to cancel the Vos/offset of the first stage 200x gain, and making it a low-pass where 'good' c0g 100n caps could be used.
Then follows the 10x inverting gain.
And then follow other filtering - the high-pass with 10u caps, and optional/extra low-pass sallen-key filters.

Eg,
200x gain -> sallen-key DC block with c0g 100nF -> 10x gain -> low-pass and other high pass sallen-key with 10u x7r/x5r caps.

Edit. probably won't work because the only the HP sallen-key not the LP can work as a DC block.

[trtr6842]

No, it wouldn't make sense, because then the tolerance for input capacitor leakage would also go down.  The whole reason why I split the gain into two stages is so that the input cap can be leaky and the amplifier won't saturate.  Also, since the 1st stage gain is significant, any non-ideal effects of those non-C0G/film caps are divided by a factor of the 1st stage gain.  If you go back to the early posts of this thread, you can see that I replaced all the 1µF ceramic caps with film ones and there was no significant change in the noise floor performance.  Also, if you look at the specs of the 1st stage opamps (OPA2182's), you can see that the noise floor is dominated by their input noise, meaning that the rest of the circuit is pretty transparent and quiet.  That means if I wanted to significantly improve the noise performance of this LNA I should focus on finding a quieter opamp, or add more OPA2182's in parallel, but both of those things are tricky.  All quieter opamps I've found have input clamping diodes, which makes input protection very hard.  More OPA2182's in parallel decreases battery life for only marginal performance boosts.

[Hawaka]

Very nice and small design!

Do you plan to release the full documentation so that one could easily order and built a unit?

[bobuhito]

trtr6842, nice work!  Did you ever try LT1037 opamps on the input like you mentioned?

I suspect LT1037's higher input bias current (and higher input current noise) vs OPA2182
might actually cause more noise (even though the voltage noise spec is 50% lower for LT1037)
since input current flows through the 100 ohm resistors and through the hard-to-predict
electrolytic cap...

[Andreas]

Hello,

I have a (one) LT1037 in my LNA in the first stage (factor 100). (2nd stage is LT1012 also factor 100)
But with input high pass 3300 uF/1kOhm because the LT1037 is optimized for low noise around 1K impedance.
The main problem is the noise coming from leakage currents of the 3300 uF electrolytic capacitor.
(input voltage dependant).
with selected capacitors I have typically 120 - 200 nVpp noise floor (7-10V input voltage).

with best regards

Andreas

[bobuhito]

Hello,

I have a (one) LT1037 in my LNA in the first stage (factor 100). (2nd stage is LT1012 also factor 100)
But with input high pass 3300 uF/1kOhm because the LT1037 is optimized for low noise around 1K impedance.
The main problem is the noise coming from leakage currents of the 3300 uF electrolytic capacitor.
(input voltage dependant).
with selected capacitors I have typically 120 - 200 nVpp noise floor (7-10V input voltage).

with best regards

Andreas

Hello,

LT1037 datasheet says 0.1~10 Hz noise floor is 60 nVpp (130 maximum),
so I guess you're saying half of your noise is from the electrolytic capacitor.

Do you get much less noise when you short your input voltage to zero?
After dielectric absorption settles (which might take a few days!), you
might get much less noise compared to your current 7-10V tests.  I'm
guessing you've tried different capacitor brands and picked the one with
the lowest noise.

In case you've posted such results or your schematic somewhere else,
please let me know of course.  Thanks!

[Kleinstein]

It is not just the capacitor as a noise source. There is also the current noise from the amplifier input in combination with the impedance of the capacitor at low frequency. So the LT1037 needs the rather large capacitors at the input.  AFAIR the simple design used the input coupling capacitor to set the lower frequency limit. This way the noise of the resistor to ground also gets partially effective. Ideally the input coupling has a lower cross over (e.g. larger resistor) and a later filter sets the 0.1 Hz limit.

[trtr6842]

Very nice and small design!

Do you plan to release the full documentation so that one could easily order and built a unit?

You can contact me for ordering a kit!  I can offer bare boards, assembled boards, or full units.

trtr6842, nice work!  Did you ever try LT1037 opamps on the input like you mentioned?

I suspect LT1037's higher input bias current (and higher input current noise) vs OPA2182
might actually cause more noise (even though the voltage noise spec is 50% lower for LT1037)
since input current flows through the 100 ohm resistors and through the hard-to-predict
electrolytic cap...

I never did try the LT1037, or any other opamps that aren't already in the shared results.  Any opamp with input clamp diodes (which is most of them!)  will not be protected.  Since input protection is an integral part of this design, I don't plan on trying any opamps with input clamp diodes. 

[Andreas]

Hello bobuhito,

most answers (including .pdf schematic) are somewhere here in the metrology section. Either in some LNA or LTZ1000 thread.
(I am too lazy to search for it).

I never measure with a short at the input of LNA. This would give a unrealistic low noise floor.
I always use batteries; formerly NiMh now as my NiMh are ageing and getting "loud" LiIon batteries with a voltage near the reference that I want to measure.
(yes the noise floor is input voltage dependant).
And I always have a battery connected to the input of the LNA to keep the input capacitor under bias since I do not want to wait 1 week for a measurement.

with best regards

Andreas

[Gerhard_dk4xp]

I never measure with a short at the input of LNA. This would give a unrealistic low noise floor.
I always use batteries; formerly NiMh now as my NiMh are ageing and getting "loud" LiIon batteries with a voltage near the reference that I want to measure.
(yes the noise floor is input voltage dependant).
And I always have a battery connected to the input of the LNA to keep the input capacitor under bias since I do not want to wait 1 week for a measurement.

Shorting the amplifier input does not show an unrealistic noise floor. It shows the
floor of the amplifier itself. An amplifier with a noise density of ~220pV/rtHz MUST see
a low source impedance of a few Ohms at most and the noise of the 10K bias network
must be shorted through this source impedance. 60 Ohm is already 1nV/rtHz.

I also would not consider Li-Ion "loud". Show me a different 8V voltage source whose
noise is 13 dB below 1nV/rtHz. This was measured with an ancestor of the preamp
in the picture. It had only 10*10UF Wima polypropylen caps in the input. Clearly
not enough, as you can see by the 30 dB/decade noise rise. This noise rise stops
finally when you get the full broadside of the Bias resistor.
That also shows that you must measure noise density and not some integrated
value where the lowest few bins dictate the result.

Amplifier == 10 * 2 * ADA-4898 in par.

This is NOT 1/f noise; you get the 1/f rise if you short the coupling capacitor on the
OpAmp side to GND. (here approximated by 60 Ohms, black trace)
Red & green are the batteries with & without 47 Ohm DC load.

With the amplifier in the pic, the propylen caps have been replaced by an AVX wet slug tantalum.
It did cost abt. 100€/$ :-(  Vishay wanted twice that.

There was no problem with leakage currents in this low-impedance environment.
In fact, I have made a copy of the ribbon mic amplifier from ArtOfElectronics ed3
with 10*1000uF Panasonic "SEPF" electrolytics. Only single ended with 16 *
Zetex transistors instead of 64 pcs. I get 70 pV/rtHz  as promised.
(single ended instead of differential)

I'm working on yet another FET version to take advantage of the cross correlation
of my Agilent 89441A. The high noise CURRENT of the many BJTs in parallel would
generate a voltage drop across the source impedance that would be common to both
channels and would not average away.


<        http://www.hoffmann-hochfrequenz.de/downloads/NoiseMeasurementsOnChemicalBatteries.pdf      >

There are also battery measurements performed by Walls at NIST, time-frequency group.
I think the filename is 1111.pdf, but it is a moving target on their server.



regards, Gerhard

[Gerhard_dk4xp]

2 pics were gone:

[miro123]

Hello,
I have followed this interesting thread. I have learned some new design tricks.
I have one question - why measuring of DC voltage sources are performed on this way?
Why not use differential measurements between 4 or more different DC sources and use cross corelation info to extract the behaviour.

I apologies in advance if I highjacked this interesting thread, or should I start a separate thread?
I asking this question because I find that all my six ADR1399 e pretty close to each other. many application like Gerhard battery noise measurements are also possible in DC differential mode. Why is all those application are doing AC decupled way? It should be a reason. The sub-100mV DC differential measurement is area where the modern integrated ADCs shines.
BR,
Miro

[Andreas]

@Miro,
I think most people read application notes like AN83 AN124 DN6 from LT. And of course also other manufacturers.

@Gerhard,
perhaps I was somewhat unclear: what I wanted to say:
My NiMH cells are ageing and thus getting noisy.
That is why I use LiIon cells (Mine are low noise).
Of course a battery is a low impedant input to the amplifier.
But the input capacitor is a part of the amplifier (which has leakage current dependant noise i.e. dependant from input voltage bias).

with best regards

Andreas

[trtr6842]

Hello,
I have followed this interesting thread. I have learned some new design tricks.
I have one question - why measuring of DC voltage sources are performed on this way?
Why not use differential measurements between 4 or more different DC sources and use cross corelation info to extract the behaviour.

I apologies in advance if I highjacked this interesting thread, or should I start a separate thread?
I asking this question because I find that all my six ADR1399 e pretty close to each other. many application like Gerhard battery noise measurements are also possible in DC differential mode. Why is all those application are doing AC decupled way? It should be a reason. The sub-100mV DC differential measurement is area where the modern integrated ADCs shines.
BR,
Miro

Often times the initial DC offset between references of the same type is much greater than the 0.1Hz to 10Hz noise they each have.  For example, the ADR1399 has a listed initial tolerance of +250mV / -300mV (550mVpp), but the noise is only 1.44µVpp.  The noise is 38 thousand times smaller than the DC offset, so  whatever ADC you feed that into would need over 112dB of dynamic range, which is very high.  Now I know it's possible to hand pick your references with better DC matching, but even if you match them to 55mV, 1/10th of the initial spread, you still need an ADC with very high dynamic range.  So even with a differential measurement, it makes sense to AC couple the signal and amplify it.  If you are going to do that, you end up pretty close to the common AC coupled LNA design you see around here, and then you might as well just measure one device at a time for simplicity.

[Alex Nikitin]

I’m getting about 500nV p-p noise with a passive 0.1-10Hz filter and Hioki DM7275 voltmeter, no amplifier, and 1.5uV p-p noise from 10V Fluke 731B reference is measurable in this setup.

Cheers

Alex

[trtr6842]

I’m getting about 500nV p-p noise with a passive 0.1-10Hz filter and Hioki DM7275 voltmeter, no amplifier, and 1.5uV p-p noise from 10V Fluke 731B reference is measurable in this setup.

Cheers

Alex

What sample rates are you using with each DMM?

[Alex Nikitin]

What sample rates are you using with each DMM?

1 NPLC, roughly 20 Hz

Cheers

Alex

[Kleinstein]

The 1 PLC (20 ms in Europe) integration already effects the bandwidth. With the usual AZ mode the meter only reads the input a little less than half the time. This adds some sensitivity to frequencies around 25 Hz. Together with a 1st order filter this gives a frequency response that looks a bit complicated. Not a simple 10 Hz upper limit, though a somewhat comparable equivalent noise bandwidth.

The effective frequencies anyway depend on the filter details. IFAIK there is not strict standard of the filter response and the comparison of noise data from different filter (including the AC coupling) setups can thus be a bit tricky.

[Alex Nikitin]

Yes, there is (obviously) some aliasing, I did check the response with a sweep generator, though for a noise signal the effects of narrow band variations should be reasonably benign. On the first screen shot below the sweep is linear, 60s, from 2Hz to 20Hz, with the filter, at 1NPLC, on the second it is 2Hz to 62Hz, and on the third is 2Hz to 62Hz without the filter.

Cheers

Alex

[trtr6842]

That's definitely still useful, if you have such a low-noise DMM already!!
Very cool to see multiple approaches at this. 

[andrewtaylor]

Hi Richard,

great project, congratulations!
Is it possible you add the PCB layout files (I think you might have orderd your pcb from JLCPCB) on your project homepage later?

would appreciate this.

[trtr6842]

Hi Richard,

great project, congratulations!
Is it possible you add the PCB layout files (I think you might have orderd your pcb from JLCPCB) on your project homepage later?

would appreciate this.

Thanks!  You can order bare PCB's, assembled PCB's, or enclosed LNA's from me if you'd like one!  will not be sharing the gerbers or placement files.  It turns out this design is a good alternative to the $500 Euler precision LNA, and I don't want my entire design to be copied with zero effort by some manufacturer, I hope you understand.  Instead I'm selling my design for much less to recoup some of the development costs, but still making it accessible to hobbyists and engineers.  Please contact me if you're interested! 

I have a new batch with some extra features arriving in about 1 week, I'll be adding updates on this thread soon!

[MegaVolt]

There are also battery measurements performed by Walls at NIST, time-frequency group.
I think the filename is 1111.pdf, but it is a moving target on their server.

https://tf.nist.gov/general/pdf/1133.pdf

[trtr6842]

I've received the latest batch of LNA boards, this revision has a few updates, and I've shipped out orders for 5 of them so far.

Notable new features:

• Metal EMI shield over the 1st gain stage circuit, clipped on to allow rework/modification
• Threaded BNC connectors for easy panel mounting and better shielding
• Negative input warning LED
• Protection against swapped inputs/outputs (accomplished through rework)
• Wire pads for easy gain switch installation

New board with EMI shield


EMI shield clip close-up


New Board with EMI shield removed


Bottom side rework and labelling


Fully enclosed unit.  This one has a custom 0.1-100Hz bandwidth for someone interested in using it to measure small 60Hz signals.
The red LED on the left lights up if the LNA is powered on and the DC input voltage goes below -10mV. 
The input protection limits the current to ±2mA, which buys me some time when I inevitably hook up my input backwards!


The new Hammond 1590Y enclosure fits everything nicely.  The foam keeps the battery from rattling around, a good re-use of the packaging these PCB's came in.
The threaded BNC connectors make mounting the board much simpler. 


In past revisions I definitely stressed input protection, since having such a large input capacitor paired with low input impedance is a recipie for trouble.  However, I neglected to protect the signal outputs from any accidental connection to DC sources of up to +30V.  With past revisions that would surely damage or completely destroy the LNA.  On these boards I reworked some series resistors and TVS diodes on the bottom side of the board.  These limit and clamp any injected current, and they do not significantly change any output characteristics when used with 1MΩ scope inputs.

x250 output protection: Series 4.99kΩ 1206 resistor and a TVS diode:


Main output protection: 2kΩ series resistor.  The DC blocking cap also helps with protection.


Since I had a few orders for these boards, I wrote an automated test script so that every LNA would be fully and consistently tested before I shipped them out.  It checks gain at input current, gain at 1Hz, shorted-input noise floor, and it does an FFT based frequency response measurement. Below is a frequency response measurement example.  I know the lines are fuzzy, and that immediately makes all of us noise-freaks think that there is something fishy going on, but that's just part of this FFT method I'm using!!!


Here is how the FFT method I use works:

I use my RTB2004 to record the un-attenuated signal generator output and the LNA outputs.  The signal generator output starts at 0V, then steps to +4V halfway through the capture.  I take each channels time-domain data and use python to calculate the FFT of each.  Then I divide the magnitude of each output by the magnitude of the input, and I get a pretty good frequency response plot!  However, since the input function I'm using is just a step function, it does not have much energy in the higher frequencies, and I quickly run into the noise floor of my scope.  I actually do two of these captures, one with a 180 second capture time, and one with a 6 second capture time.  This covered the 0.1Hz to 10Hz range well, but you definitely can see lots of fuzz up near 500Hz. 

I use this method because it is much faster than injecting one sine wave at a time and measuring the gain.  That takes hours at these low frequencies, but only a few minutes with this method.  For me that's well worth a little fuzz on the resulting plots.  I am working on some more advanced excitation signals to try and get cleaner results.  I've tried a sin(x)/x input, and that cleaned up the higher frequencies, but at the expense of the low frequencies.  One promising input signal is a sum of approximately log-spaced input frequencies.  If each input frequency is a multiple of some base frequency, then the pattern repeats cleanly and can be loaded as an arbitrary waveform in my signal generator.  Focusing the spectral energy on a log spaced of points could result in a group of clean data points that would make a good bode plot, but it leaves a lot of points with no energy, so those would have to be removed from the plot to keep it clean.  Anyways, I will share how that goes if I get around to it!

[gf]

One promising input signal is a sum of approximately log-spaced input frequencies.  If each input frequency is a multiple of some base frequency, then the pattern repeats cleanly and can be loaded as an arbitrary waveform in my signal generator.

Since the peak amplitude is limited, the key to good SNR is a signal with a low crest factor.
The script I found here some time ago tries to find a phase for each tone of a multi-tone signal, so that the crest factor is minimzed.
It seems to work nicely as long as the spectrum is populated sparsely enough.
Since it uses FFT internally, the frequencies of all tones are integer multiples of a base frequency, which meets your desire, too.

[trtr6842]

Interesting script!

Here is what I've got: The script is trying to add 100 sines together, approximately log-spaced from 0.02Hz to 10Hz.  Since a lot of the lower ones overlap when rounded to the nearest multiple of the base frequency, the script ends up giving 70 tones.  To get a small enough resolution bandwidth on the FFT, the pattern is repeated 4 times before the FFT is computed.  Right now each sine has the same amplitude, but I do want to start shaping the amplitude profile to optimize SNR for the expected LNA frequency response.

Basic sum-of-sines, no phase shift:  Vpp = 28.6x the amplitude of a single sine


Flipped phase for every-other frequency:  Vpp = 23.7x the base amplitude


Linear phase shift, each frequency's starting phase is evenly spread out over 360°, Vpp = 16.9x the base amplitude, which feels pretty good considering there are 70 tones added together.  I'm curious what kind of Vpp to tone amplitude ratios the random iterative script can accomplish.


Here is the script I used to generate the time domain signals:

Code: [Select]

import numpy as np
class WaveformMath:       
    def sum_logspace_sines(n_points, f_max_ratio, n_f, n_cycles=1):
        # n_points is how many time-domain samples you want.  The resulting signal will have fewer tones, since tones that are too close together will be skipped
        # f_max ratio is the ratio of the max frequency to the min frequency (the lowest frequency is normalized to 1.0, so this is how you define the max frequency)
        # n_f is the number of frequencies you'd like
        # n_cycles is how may times you'd like the pattern to repeat within the number of points given

        x = np.linspace(0, n_cycles, n_points)  # 0.0 to 1.0 normalized sample time
        y = np.zeros(n_points)
        f_log = np.logspace(0, np.log10(f_max_ratio), n_f)  # ideal set of log-spaced tones
       
        f_approx = []
        f_max = 0
       
        for f in f_log:
            fa = (f*n_cycles//1)/n_cycles  # round down to nearest frequency
            if fa > f_max:
                f_approx.append(fa)
                f_max = fa
               
        for i in range(len(f_approx)):
            # y += np.sin(2*np.pi*f_approx[i]*x)  # straight sum
            y += np.sin(2*np.pi*f_approx[i]*(x+n_cycles*i/len(f_approx)))  # linear shifted phase based on index
           
            # if i%2 == 0:  # Flipped sign every other frequency
            #     y += np.sin(2*np.pi*f_approx[i]*x)
            # else:
            #     y += np.sin(2*np.pi*f_approx[i]*x*-1)

        # print('Number of tones: %d' % len(f_approx))
        # print('peak: %.3f' % (max(y)/2-min(y)/2))
        y = 2 * y / (max(y)-min(y))   # normalize Vpp to 1.0
        # print(y[0])  # print initial DC offset (DC average is stil 0)
           
        return x, y


[gf]

Right now each sine has the same amplitude, but I do want to start shaping the amplitude profile to optimize SNR for the expected LNA frequency response.

If I remember correcly, the script I refereced also lets you specify the desired amplitude (gain) for each frequency.

EDIT: I digged out the script again. When I feed it with your parameters (70 log-spaced frequencies in the 0.02...10Hz range, quantized to multiples of 0.02 Hz), then I get crest factors of about 2.1 (each run gives different results, as it starts with random phases). I have not yet considered different amplitude/gain for each frequency.

Modified script:

Code: [Select]

%% script to do a mutlitone low crest factor noise signal

nx = 10000;
fx0 = zeros(nx/2+1,1);
% fx0(floor(logspace(log10(0.02),log10(10),100)/0.02+0.5)+1) = 1; // round to nearest
fx0(floor(logspace(log10(0.02),log10(10),100)/0.02)+1) = 1; // round down
ifr = find(fx0 != 0);
nf = length(ifr);
gains = ones(nf,1);

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% initialize
phi = 2*pi*rand(nf,1); % random phase
fx0 = zeros(nx/2+1,1);
fx = fx0;
fx(ifr,:) = gains.*exp(1i*phi);
% fx0(ifr,:) = gains;

%% main iteration loop
% numIter = 1000;
numIter = 500;
for i = 1:numIter
  % make conjugate symmetric
  fx1 = [fx; conj(fx(end-1:-1:2))];
  % time domain
  x = real(ifft(fx1));
  % clip at a target crest factor of 2
  % xmax = 2*rms(x);
  xmax = 2*sqrt(mean(x.*x));
  x( x > xmax)= xmax;
  x( x < -xmax) = -xmax;
  % go back to frequency domain
  fx1 = fft(x);
  % transfer phase
  fx(ifr,:) = gains.*fx1(ifr,:)./abs(fx1(ifr,:));
end
% final time domain signal
fx1 = [fx; conj(fx(end-1:-1:2))];
% time domain
x = real(ifft(fx1));
fprintf('final crest factor = %f\n', max(abs(x))./std(x));

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

clf

subplot(3,1,1)
plot(x/max(abs(x)))
grid on

subplot(3,1,2)
plot([-nx/2:nx/2-1]*0.02,fftshift(abs(fx1)))
grid on
xlim([-10 10])
ylim([0 1.1])
xlabel("Hz")

subplot(3,1,3)
plot([-nx/2:nx/2-1]*0.02,fftshift(arg(fx1)))
grid on
xlim([-10 10])
ylim([-pi pi])
xlabel("Hz")
ylabel("phase")

EDIT:

Linear phase shift, each frequency's starting phase is evenly spread out over 360°, Vpp = 16.9x the base amplitude, which feels pretty good considering there are 70 tones added together.

Do you really mean peak-to-peak, and not peak? And "base amplitude" means just the (single-sided) peak amplitude of the sine?
Then 16.9 would imply a crest factor of 16.9/2/(sqrt(0.5)*sqrt(70)) = 1.4283, which would very good for the sum (a single sine wave already has crest factor sqrt(2)).

If you mean this value, then it would be only approx peak, not peak-to-peak.

Code: [Select]

# print('peak: %.3f' % (max(y)/2-min(y)/2))

[trtr6842]

Linear phase shift, each frequency's starting phase is evenly spread out over 360°, Vpp = 16.9x the base amplitude, which feels pretty good considering there are 70 tones added together.

Do you really mean peak-to-peak, and not peak? And "base amplitude" means just the (single-sided) peak amplitude of the sine?
Then 16.9 would imply a crest factor of 16.9/2/(sqrt(0.5)*sqrt(70)) = 1.4283, which would very good for the sum (a single sine wave already has crest factor sqrt(2)).

If you mean this value, then it would be only approx peak, not peak-to-peak.

Code: [Select]

# print('peak: %.3f' % (max(y)/2-min(y)/2))

Thanks for sharing your results!

To try and clarify, when I said "Vpp=16.9x the base signal amplitude", I was incorrect, thanks for catching that!. I actually calculated the Vpp of the end signal and divided it by the Vpp of a single tone.  Each tone is calculated as sin(2*pi*N), so the amplitude is 1.0, and Vpp would be 2.0.  So that value would be same as the overall single-ended amplitude divided by the single-tone single-ended amplitude (what a tongue twister!)

So for crest factor, I think the actual with the linear-phase method is 2.853.  I added this code at the end of my script, after the normalization (which doesn't affect crest factor):

Code: [Select]

        vpp = (max(y) - min(y))
        vp = vpp/2
        ac_rms = np.std(y) # AC rms and standard deviation are mathematically equivalent
       
        print('peak to peak: %.3f' % vpp)
        print('peak (SE): %.3f' % vp)
        print('std-dev (AC RMS): %.3f' % ac_rms)
        print('Crest Factor: %.3f' % (vp/ac_rms))
and I got:

peak to peak: 2.000
peak (SE): 1.000
std-dev (AC RMS): 0.350
Crest Factor: 2.853

So your iterative phase guessing is about 35% (2.6dB) better.  I'm curious if that would make a noticeable difference, I'm not sure.  How quickly does your example script run for the 10kpt 1000 iteration script you shared?

[gf]

peak to peak: 2.000
peak (SE): 1.000
std-dev (AC RMS): 0.350
Crest Factor: 2.853

So your iterative phase guessing is about 35% (2.6dB) better.  I'm curious if that would make a noticeable difference, I'm not sure.  How quickly does your example script run for the 10kpt 1000 iteration script you shared?

About 0.4 seconds, on my 12 years old notbook, without plotting. And about 2 seconds with nx=100000.
Don't know if 1000 iterations are really necessary, but the original value was obviously too low, so I increased it generously to 1000.

Crest factor 2.85 is not bad, IMO

[gf]

I do want to start shaping the amplitude profile to optimize SNR for the expected LNA frequency response.

Tried to add-in the inverse frequency response of a 3rd order highpass with 10 0.1 Hz cutoff.
Crest factor only around 1.6, but the resulting signal is of course strongly dominated the lowest (0.02...0.1 Hz) frequencies.

EDIT:

Alone the 0.02 Hz tone contributes ~98% of the total signal power, and the frequencies <= 0.1 Hz together about 99.6% of the total power 

I don't know what the exact aim is. Weighting with the inverse filter response will lead to a flat ouput. But if the aim is to get the same SNR at each frequency, then the amplitude of the stimulus tones should be rather weighted by the input-referred noise PSD.