1

I'm combining opamps to reduce the frontend opamp noise by a factor 0.707. This works in LTSpice, and I understand why it should work. My circuit amplifies a signal that is symmetric around 2.5V and should have a gain of 101.

But, I see two options, and I'm wondering whether LTSpice is telling me the truth or I'm in simulation purgatory.

Option 1 (left) Two complete parallel input circuits. The non-correlated output noise is combined through the 300 Ohm resistors.

Option 2 (right) I'm combining the parallel opamp circuits using one gain resistor R3 and paralleling the non-inverting inputs of both opamps. The non-correlated output noise is combined through the 300 Ohm resistors.

My first instinct was option 2 because it eliminates the extra uncertainty of two gain resistors, but I'm wondering if the combination of the noise sources at the non-inverting input has any effect at all.

LTSpice tells me there is no difference at the output.

Though when I measure the noise not at VOP/VOM but right before the 300 Ohm resistors, there is a huge difference. The combined circuit gives me a ridiculously high noise figure that's 2000 times higher, while option 1 gives the expected original non-parallel noise that's factor SQRT(2) higher.

Comparison between both circuits

tobalt
  • 18,646
  • 16
  • 73
Graafvaag
  • 61
  • 4
  • 2
    Could you please take a look at [this answer](https://electronics.stackexchange.com/a/28255/95619)? – a concerned citizen Nov 11 '22 at 12:33
  • This is a fundamental misunderstanding of the idea. It is the input-referred noise that improves, and thus, the whole SNR of the system. The output noise are effectively added but in RMS terms. Your 2nd option is no good (unless you're trying to have some ft-doubler or something), you'll get -3dB less noise, and you'll get 2x the current consumption. Let me know if you want me to expand on this.. – Designalog Nov 11 '22 at 13:16
  • Thank you. I'll post cleaner schematics in the future. My Altium schematic is very clean and organized, but my LTSpice doodles are usually mostly functional and not meant to be pretty. At least not any more than to help me draw them. – Graafvaag Nov 14 '22 at 09:01

2 Answers2

2

It is a standard way to reduce noise. Given the large gain, you can consider much larger resistor for the averaging at the output..That way the outputs will be less loaded due to mismatched opamps and gain resistors.

It only works for low source impedance because the input current noise rises by the square root of N, as you add more parallel stages.

The second approach will not work in practice, because the input offset voltages of the amps will differ, so they will disagree on the gain node voltages and saturate.

tobalt
  • 18,646
  • 16
  • 73
  • I love this answer. The second approach will have both opamps fighting over the feedback voltage. This completely explains my findings and can be considered 'the answer' to my question. Thank you. - I want to add that the 300 Ohms resistors are part of the next gain stage, where originally a single 150 Ohm resistor was used. – Graafvaag Nov 14 '22 at 09:02
  • > "That way the outputs will be less loaded due to mismatched opamps and gain resistors." Don't understand this. If op-amps are the "same", and mismatch from the resistors would be at most 1or 2 % ? Op-amps would have theoretically exactly the some "behavior", so "quasi" any value resistor (low values included) array should be ok. – Antonio51 Nov 14 '22 at 09:31
  • 1
    If the opamps have 1mV difference in input offset voltage (Vos) and both have A=100, there is 0.1V between both outputs. This creates a DC current through the averaging resistors. (300 Ohms in my example) Higher resistors cause less DC current. I think this is what tobalt meant. – Graafvaag Nov 14 '22 at 10:15
  • @Graafvaag yes this is exactly what I meant. Glad it helped. – tobalt Nov 14 '22 at 10:19
0

Don't know if I am wrong with this simulation. Made with microcap v12. For reference.
However, it shows a significant noise reduction at input.

Calculated noise for 1 to 8 circuits paralleled.
Noise is reduced as \$1/sqrt(N)\$. N: number of paralleled op-amps.

Op-amp gain of 100 ... Average network added.

enter image description here

Antonio51
  • 11,004
  • 1
  • 7
  • 20
  • "...significant noise reduction at output & input"?? Only at the input! The input noise will be reduced in this example, but the experiment isn't very useful as the outputs are shorted against each other. These scheme of noise reduction ONLY works with an averaging network at the output of all paralleled amplifiers. – Designalog Nov 11 '22 at 15:14
  • I add an averaging network as in this https://www.analog.com/en/technical-articles/paralleling-amplifiers-improves-signal-to-noise-performance.html – Antonio51 Nov 11 '22 at 16:04
  • @ErnestoG Just a question: the averaging at the output is not made by op-amps output internal impedance? – Antonio51 Nov 11 '22 at 16:33
  • those impedances are not physical, plus they're within their feedback loop. – Designalog Nov 12 '22 at 09:29