Trying to get lower noise with an op-amp based amplifier

Thread Starter

Elerion

Joined Sep 11, 2017
117
Hi everyone.

I’m trying to desing a simple low noise amplifier but I’m not able to reach low enough noise, and I suspect I’m missing something. I would appreciate some help to track the problem down.

The circuit is built around a TL972 (this opamp is 4 nV/sqrt(Hz)).
I used a non-inverting 101x voltage gain, with two poles.
LN_opamp_101x.png
C2 (with R2) sets a high pole at about 40 kHz.
C1 (with R1) sets a low frequency pole at about 7 Hz.
Both poles set the passband.
LN_opamp_101x_sim.png

Circuit works perfectly (built and tested), except for the noise expectations.
When I short input to ground, I measure 1 mV – 5 mV (rms) at the output, suggesting an EIN of 10-50 uV rms (due to the approx 100x voltage amplification).

There’re two problems:
1)- Probably 10 uV is not that bad, considering the circuit is not inside a metal enclosure (I was aiming for 5 uV max).
2)- The real issue is that noise level varies with apparently irrelevant things. Sometime the noise is 10-20 uV (quite stable), but then I move some meters away or closer, or move some wires around, and the noise gets 40-50 uV (stable again). And it’s possible to revert and partially control these changes, suggesting there’s a very concise reason behind which I couldn't find after several hours of trial and error.

Of course I did not use a protoboard. I soldered all components as tight as I could (no flying wires), and used 1000 uF close to each supply pin and ground. I also tried to short C1, which (maybe) improves very little. The power supply is linear, and no close SMPS was powered on.

Any tip? Anything I could try to achive lower noise with the same op-amp?
Thank you!
 

sghioto

Joined Dec 31, 2017
1,443
Any tip? Anything I could try to achive lower noise with the same op-amp
Did you read the data sheet about layout guidelines? Also noted you have R3 labeled as 1000K that's 1M why so high?
EDIT: What will the input be connected to when in use?
SG
 
Last edited:

Thread Starter

Elerion

Joined Sep 11, 2017
117
Did you read the data sheet about layout guidelines? Also noted you have R3 labeled as 1000K that's 1M why so high?
EDIT: What will the input be connected to when in use?
I did read the guidelines and tried to follow as close as possible (but there's no ground plane, as this prototype is not on soldered on a custom PCB yet).
The input will be audio circuits with an output impedance up to 10 kohm.
 

Papabravo

Joined Feb 24, 2006
14,393
You're not going to get anything like optimum performance on a prototype board. Have you done a noise analysis on the simulator?
What kind of noise are you observing?
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
Performance will likely improve with a proper PCB. Why do you need a gain of 101?
Is this a mic preamp?
I want to be able to measure the noise output of audio circuits, which is below the range of my measurement equipment.

You're not going to get anything like optimum performance on a prototype board. Have you done a noise analysis on the simulator?
What kind of noise are you observing?
Yes, I have simulated noise. About 100 uV output RMS (100 kHz bandwidth). EIN is about 2.1 uV RMS. Both white and pink noise.
 

OBW0549

Joined Mar 2, 2015
3,545
Couple of thoughts:

1) If you're getting changing levels of noise depending on moving around, moving wires, etc., I'd say you're getting pickup of stray signals from the surroundings-- mains frequency, RF, switching supply noise, and so forth. If you're dealing with extremely low noise levels and want to test residual noise performance, you've got to shield the absolute livin' daylights out of EVERYTHING.

2) Have you looked at the output of this circuit on an oscilloscope? What does the noise look like? You can tell a lot from the visual characteristics of the noise.

3) The TL972 lists its input equivalent noise voltage density as 4 nV/√Hz at 100 kHz, but it says nothing about input current noise density. And if this circuit is going to be driven from a high-impedance source such as 10 kΩ, it's likely the current noise that will cause the biggest problems rather than the voltage noise. With these input specs, the TL972 is clearly intended to be driven from a much lower impedance source.

4) The omission of an input current noise spec from the data sheet is likely telling, especially since the input bias current (750 nA max.) is so large. My guess is that the input current noise is simply horrendous, which is most likely the reason for not mentioning it in the data sheet.

5) In figure 6 on page 6 of the data sheet I note that the 1/f corner frequency for input voltage noise density is way up there, at 1 kHz (probably the reason for specifying noise at 100 kHz rather than the customary 1 kHz); be aware that this thing is going to produce a LOT of noise at low frequencies.

Looking at the TL972 data sheet I struggle to think of what I could possibly use it for; certainly not any application where I need low noise at low frequencies when driven from a high source impedance.

You might be better off with a decently quiet JFET input op amp such as an LT1793.

EDIT: you might also try an LT1028.
 

Audioguru again

Joined Oct 21, 2019
1,908
The TL97x opamp is weird;
1) Its output is class-A instead of push-pull class-AB like most opamps.
2) Its maximum recommended supply is only 12V instead of 30V or more like most opamps.
3) Its noise is low only when its source signal has a very low impedance (coil and magnet dynamic microphone?).
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
3) The TL972 lists its input equivalent noise voltage density as 4 nV/√Hz at 100 kHz, but it says nothing about input current noise density.
Good catch. Unfortunately, I don't really yet know how bad is the input current noise, as my measurements were done shorting the input to ground. I will have to wait until I have time to get the whole thing into a metal grounded enclosure.

You might be better off with a decently quiet JFET input op amp such as an LT1793.
EDIT: you might also try an LT1028.
Those are expensive parts. What about LF356? Has a very low bias and noise current. Not great EIN voltage, but it's much much cheaper opamp.

Its noise is low only when its source signal has a very low impedance (coil and magnet dynamic microphone?).
As I previously said, I don't really know if that is so. After I shield everything, I compare the output base noise into a low and a high impedance, and post the results.
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
you've got to shield the absolute livin' daylights out of EVERYTHING.
I have an small aluminium box which I could use, but there's no space inside for 2x9 V batteries, so I would have to drill the box to provide power. Would this spoil the shielding? I have metal conectors for DC power, but they are obviously 2 poles, and I need V+/V-/GND.
How can I do this test properly? Thank you.
 

OBW0549

Joined Mar 2, 2015
3,545
I have an small aluminium box which I could use, but there's no space inside for 2x9 V batteries, so I would have to drill the box to provide power. Would this spoil the shielding?
No, a few small holes won't make much difference-- or any difference.
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
Many audio preamp circuits can use a modern opamp in a circuit powered by one 9V battery.
Do you mean rail to rail opamps? By the way, TL972 is rail to rail. I don't think that working with such tiny signals would work with a single supply, though. In the case of TL972, the lower possible output voltage (VOL) is about 10 mV over negative rail (when VERY little current is demanded).
 

Audioguru again

Joined Oct 21, 2019
1,908
Any opamp can use a single-polarity supply if it is biased at about half the supply voltage so that its output can swing up and down, and input, output and negative-feedback-ground coupling capacitors are used.

Most audio opamps are not rail-to-rail.
 

Audioguru again

Joined Oct 21, 2019
1,908
All opamps produce higher level of low frequency noise (rumble).
C1 reduces the gain at frequencies of 7Hz and lower but I do not know if your meter can measure frequencies that low.
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
Any opamp can use a single-polarity supply if it is biased at about half the supply voltage so that its output can swing up and down, and input, output and negative-feedback-ground coupling capacitors are used.
Ok, you mean this, don't you?
LN_opamp_101x_single.pngI decided to use split supply because I thought that could lead to better performance.

perhaps the C1 (22µF) contributes to that
Why could that be? I think you are right, based on experimentation, but I don't really know why C1 could worsen EIN voltage. In any case, the difference is small. Taking noise measurements in various setups is not easy, due to the highly random behaviour below a couple of mV or so.
 

Thread Starter

Elerion

Joined Sep 11, 2017
117
In this post I'd like to share some results.

I placed the circuit inside a small aluminium enclosure, and repeated the measurements, using:
- a regulated power supply
- a pair of 9v batteries.

With the power supply I usually get 1 mV rms. Depending on where I stand, I can get readings as low as 600 uV, or as high as 2 mV rms, but never as high as 5 mV.

With batteries I always get less noise than using the power supply. I got readings as low as 450 uV, but sometimes as high as 1 mV (again, depending on where I stand).

One more test: opening the enclosure lid (so the circuit is shielded from all directions except one) clearly makes noise a bit higher when using the power supply, but seems to have no evident impact when using batteries.

This is the noise "look" on an average result using batteries.
noise batteries metal box.png

All tests where done with input shorted, so noise current is not playing any role. The specs (TL972) are 4 nV/sqrt(Hz), and I got approx 500 uV (when using batteries), which seem to be OK, but well below the specs (as voltage gain = 100, a 500 uV output means 5 uV rms EIN. On the other hand, based on the 4 nV/sqrt(Hz) datum, and approximating EIN voltage as 4nV * sqrt(20000) we get 566 nV, not considering 1/f noise, which is clearly present; let's assume 1 uV rms). My results are 5 times worse.

Based on your experience, do you think these results are realistically good?
I ordered some better opamps for more testing.
 
Top