Low noise voltage opamp oscillating

[pwnell]

Birds eye overview of my project: I am building a low noise voltage amplifier to amplify small signal noise, specifically I want to quantify Johnson noise in resistors.  The amp is not my design -  merely built it (I know it is working - but I might have botched something, I therefore cannot post it, but suffice to say it is dual channel, two cascaded low noise opamps each with negative feedback, and non-inverting, all using SMD components in a shielded box, powered by 4 D-cell 1.5V cells).  The power supply is partly of my own design as I never had the schematics.  See schematic attached.



I tested this amp by connecting a signal generator to the input and measuring the output - I get a clean gain of approx 2100 between 100Hz and 2000Hz with an input of 0.865mV RMS (Sine wave).  The supply rails look fine - after my addition of the TLE2426 and caps it does not drift. 

Here is the problem - I connect various carbon resistors to the input (via a shielded copper tube and BNC cable), then measure the output.  Here is what I get - random noise (as expected) for resistors of values:

Short
10MΩ 1% Carbon
330kΩ 1% Carbon
10kΩ, 5%, Carbon
1kΩ, 5% Carbon
100Ω, 5%, Carbon
10Ω, 1%, Carbon

But for these resistors:
1MΩ 5% Carbon
100kΩ, 5%, Carbon

I get a clipped square wave output.  Frequencies are ~1.178kHz and ~8kHz, respectively (1MΩ and 100kΩ).  Measuring at the input to the opamp I get a 60mV or so spike at that frequency, but what is oscillating in my circuit that can cause this?  There are no caps between the input and the non-inverting input, only between the output of each opamp and the inverting input.

[David Hess]

Check figure 17 on page 17 of the datasheet.  The 0.22 microfarad output capacitor shown will be unstable for some loads.

I would replace that capacitor with a 1 microfarad solid tantalum or 10 microfarad aluminum electrolytic.  The higher ESR of these capacitors aids stability.

[pwnell]

Still unstable

[magic]

Why are you posting a schematic of the PSU rather than the amplifier itself? Did you verify that it's the output of TL2426 which goes bonkers and not the opamps?

You can get a 100% stable ground by tapping into the middle of the battery chain. Problems will begin when some of the batteries get depleted before others, but it's good enough for testing at least (or you can even choose to ignore the problems and just build it that way).

And your schematic is wrong; you are supposed to connect COMMON to VEE and BAT- and OUT to GND and the capacitors. C2 makes no sense as drawn because it's shorted. Make sure the schematic corresponds to what you have actually built and vice-versa.

[pwnell]

Why are you posting a schematic of the PSU rather than the amplifier itself? Did you verify that it's the output of TL2426 which goes bonkers and not the opamps?

The output of the supply rails look stable to me - no oscillations or anything strange I can see.  I posted it as I designed it, whereas the low voltage amp I did not design and have no permission to repost.

And your schematic is wrong; you are supposed to connect COMMON to VEE and BAT- and OUT to GND and the capacitors. C2 makes no sense as drawn because it's shorted. Make sure the schematic corresponds to what you have actually built and vice-versa.

You are right - the schematic is wrong, the power supply circuit is correct (since it works) - see attached, I redrew it and removed the 0.22µF caps as I tested with and without them and it made no difference..

Any insights where in the amp I can look at?  Like I said - feeding it an input voltage over 100Hz - 2kHz it seems stable - properly amplifying the signal.  Only when I apply a single resistor and no voltage source to the input, and only for some values, does it oscillate.

[magic]

Any insights where in the amp I can look at?  Like I said - feeding it an input voltage over 100Hz - 2kHz it seems stable - properly amplifying the signal.  Only when I apply a single resistor and no voltage source to the input, and only for some values, does it oscillate.

Well, you said there are two stages so you could probe in the middle to see which of them is doing it. I kinda suspect the first.
I also suspect that if you apply a signal through one of the "evil" resistors then the amplifier will also not work right.

Maybe it could be something related that shielding pipe and cable and their parasitics, can you describe that part of the setup at least?
Myself, if I were to measure Johnson noise, I would just connect the resistor somewhere close to the input opamp and put the whole assembly in a grounded metal box.

[pwnell]

Could not test everything. So applying 2mVpp from my signal generator causes a stable output, regardless of the resistor in parallel. Not sure if I was to put it in series. However with only the signal generator attached, if I turn off the signal generator but leave it connected, the output switches to oscillating immediately and stays like that for 4-5 seconds before returning to a stable output of 0V.

Also I cut out the bnc input path and connected the resistor directly to the board - made no difference.

** Edited to show 2mVpp and not 2Vpp

[Kleinstein]

Some amplifiers, espeically JFET stages tend to oscillate with a certain input impedance. They may need some parallel or series impedance (e.g. Ferrite in series or capacitor in parallel) to make them stable. So this could be an internal problem to the amplifier.  Another possibilitly is expernal coupling (e.g. capacitive) than can be more severe with higher source impedance.

A oscillation frequency of only 1-8 kHz is rather low. Amplifiers normally tend to oscillate more near there upper bandwidth limit.

For testing an external generator and resistance, mainly the sereis conection makes sense. Parallel the ouput impedance of the generator (e.g. 50 Ohms)  would normally swamp the extra resistor of high values. If in doubt one may use the generator with an extra attenuator (e.g. -20 dB) at the output to make the ouput impedance even better controlled (the active amplifier part will have a frequency dependent autput impedance to some degree).

[pwnell]

I did some further testing.  When I connect my signal generator to the amplifier as per below, I get an interesting result.



The PSU is turned off, so the amp is unpowered.  I use BNC from Signal Generator to amp, and have a T to split it off from the BNC connector on the signal generator to the oscilloscope (magenta) where I terminate it into a 50Ω passthrough adapter.  I also have a BNC from the output of the amp to the oscilloscope (yellow).  What I noticed is that when I output a 2mVpp square wave @ 100Hz (frequency is insignificant, I have a rise time of about 10ns), I see ringing on the rising and falling edges of the *input* signal from the signal generator the moment I attach the BNC cable to the (unpowered) DUT.  The DUT has a 10Ω resistor between the input signal and the first opamp.  Is this normal?  Can this feedback be the cause for my woes when connecting a resistor of specific value to the input?



When the amp is powered and I feed that square wave in, the amp does not oscillate but it produces bad output waves (yellow) - is this acceptable for a 100Hz input or indicative of an issue?



Lastly, with the DUT connected to power, the oscilloscope and only a variable resistor as input (no signal generator, no external voltage), I found that the amp is "stable" until 42kΩ.  Between 42 and 43kΩ the amp starts showing sinusoidal output, quickly rising until it hits saturation and clips into a square wave.  The oscillation is repeatable at 42 - 43kΩ input.

[pwnell]

I made some progress.  I rewired much of the project to use way shorter cabling.  Nothing else changed.  I now understand what causes it to oscillate and clip but still do not understand where this is coming from.  Please see below.



In this test, I connected a 100kΩ resistor to the input of the voltage opamp.  Before this would cause the amp to oscillate at V+ and V-.  With my rewiring it is not oscillating any longer.  See the 60Hz power line noise that is picked up?  Why is that happening?  The circuit is supplied by a DC battery power supply - not connected to any AC equipment other than the oscilloscope probe.  I made sure all AC wires are as far away from the circuit as possible.

When I touch the metal chassis with my hand, this happens:



Most of the AC noise is gone.  Why would my body attenuate that AC signal?  Grounding happens automatically when I connect the oscilloscope probe so I cannot provide additional grounding. 

For interest sake - with the 1MΩ resistor attached, the unit again oscillates between V+ and V-.  Touching the ground plate with my hand removes this and shows that it is this AC induced noise that causes the amp to saturate and clip - now it is still showing AC 60Hz noise but it is no longer oscillating / clipping:



How can I make this low noise voltage amp immune to AC mains noise?

Here is a picture:

[magic]

Now it's making sense

Yes, it's picking up the electric field of mains wiring. It doesn't matter how far the cables are, if they are all around you will experience their field.
The solution is to surround the whole input node (opamp, cables, resistor) with a grounded conductor. It may be your hand, or both feet for better coverage, or a metal case with holes for cables

[pwnell]

The solution is to surround the whole input node (opamp, cables, resistor) with a grounded conductor. It may be your hand, or both feet for better coverage, or a metal case with holes for cables

Problem is - whenever I bring anything grounded to mains ground close to the circuit it just makes things way worse.  I used a large metal dome, connected to mains ground, and placed it over the circuit, PSU and cables - it was about 5 times worse than just leaving it in open air.

It is the best when I do not probe it with my oscilloscope, but instead use my macbook air unplugged from the wall, and a Scarlett 2i2 USB audio interface powered via USB-C.  Then I find minimal AC interference as long as I do not come close to the circuit.

[magic]

The shield should be grounded to the circuit's ground.
Ideally, you would have a star ground point near the first opamp and connect to it:
- the first opamp's gain resistors
- the second opamp's gain network
- the shield
- ground of the output connector

But you can probably get away with a lazier ground layout, as long as everything is contained on that one board and no big currents flow through the circuit.
One thing that won't work very well is grounding the shield to some random mains socket. Too much potential difference, even if it's just a fraction of a volt.

Here's how you do shielding around such things:
http://www.vk2zay.net/article/251
Except that the guy should have run a coax cable between his two cans instead of a pair of bare wires, or put everything in one can.
But he only cares about DC so maybe it's good enough for him as long as the opamp doesn't clip due to picked up noise.
(Or as long as he doesn't notice that it clips and just accepts a wrong result and never finds out it was wrong )