Hi,

Consider a standard op-amp circuit (example) either inverting or non-inverting. Ignoring the 1/f noise then the

noise gain = 1 + Rf/Rg
noise band width = Gain band width product / noise gain
input referred RMS noise voltage = input referred noise density * noise band width

which is white noise. What is the nature of this noise? Is it spread out across all of the noise band width?
For an audio application, all else being equal, does a 20MHz GBW op-amp produce more audible noise than a 10MHz GBW op-amp?

I read that you can create another pole, for example with an RC filter on the output, thereby limiting the noise band width but is there any point to that if we can ignore noise above ~20kHz? If it's true that we can ignore noise above 20kHz then we can you use 20kHz as NBW for the calculation and not bother with the RC filter on the output?

 

as the name implies, GBW is a product of gain and bandwidth.
if the bandwidth is 20kHz and gain is 50 then GBW is 1MHz.
and if you have part with 20MHz GBW that is used for signals up to 20kHz, maximum gain is 1000. product with 10MHz GBW will have max gain of 500.

 

Hello,

Does this article help?
https://sound-au.com/noise.htm

Bertus

 

Refer to the noise density figure (quoted in V/√Hz) and multiply by the square root of the bandwidth to get the output in Volts.
Audible noise by definition has a bandwidth of 20kHz, so for a circuit with a flat gain response, then the output audible noise will be the noise density multiplied by 141. An amplifier with more bandwidth will make the same audible noise as one with the same noise density.
But then there is 1/f noise, which increases with decreasing frequency at 6dB/octave below a figure known as the 1/f noise corner.
MOSFET op-amps have a high 1/f noise corner (so make more 1/f noise).
1/f noise is a problem with circuits that have more gain at low frequencies (RIAA weighted preamps, for instance)
And then there is thermal noise with a density of 2√(kTR) which comes from the resistors in the circuit. The thermal noise of the input resistor is multiplied by the gain of the circuit. The thermal noise of the feedback resistor isn't.
And then there is current noise which adds to the voltage noise density Vn=√(vn^2+(R.in)^2) where R is the input resistance.
Bipolar op amps make more current noise, FET op-amps make more voltage noise. The noise density of a good JFET op amp will be below 1fA/√Hz

 

What is the nature of this noise? Is it spread out across all of the noise band width?

It is white noise, which is spread basically evenly across the spectrum

For an audio application, all else being equal, does a 20MHz GBW op-amp produce more audible noise than a 10MHz GBW op-amp?

For equal op amp input noise specs, and the same audio bandwidth filter, the noise would be the same.
So you want an op amp with a low input noise spec.

If it's true that we can ignore noise above 20kHz then we can you use 20kHz as NBW for the calculation and not bother with the RC filter on the output?

Probably could be ignored, but it's good practice to add a simple 1-pole RC filter to roll it off above 20kHz.

 

Oops, I forgot to square root the NBW.

The thermal noise of the input resistor is multiplied by the gain of the circuit. The thermal noise of the feedback resistor isn't.

Aha, so that's why according to LTSpice an input resistor of lower value has a higher noise density contribution than the feedback resistor of some higher value!

@crutschow Thanks, yes I would consider a low noise amplifier, but playing around with different configurations it can be a waste of money if the resistor noise dominates. I need to study this in more depth.

edit: typos

 

You need to have a VERY good op-amp before the resistor noise dominates. But watch out if you are using high-value resistors because it is very easy for the current noise density multiplied by the resistance to be the dominant source of noise.
Also remember that although the input resistor to your circuit may generate noise, the actual resistance to be used in the noise calculation is the input resistance in parallel with the output resistance of the previous stage.

 

I made a spreadsheet for doing some calculations, screenshot below, and was under the impression that resistor noise would often dominate, although in the attached example not massively. Now that I look at it, did I forget to include resistor current noise? Probably, I'll have to check.

Edit: oh wait, resistor noise current is technology dependent isn't it? Then I didn't include it.

 

I made a spreadsheet for doing some calculations, screenshot below, and was under the impression that resistor noise would often dominate, although in the attached example not massively. Now that I look at it, did I forget to include resistor current noise? Probably, I'll have to check.

Edit: oh wait, resistor noise current is technology dependent isn't it? Then I didn't include it.

What value of resistance did you use to calculate the resistor noise?

 

The video does some justice to contributions Motorola team and others in semiconductor field made.
Thanks for the topic. glad to see testing.
The HP3555B set for 15kHz flat band, resistors 10k and 101k giving10X gain and output measured in microvolts RMS.

 

I forget to include resistor current noise?

Resistor current noise is just its voltage (Johnson) noise divided by its resistance.
It's not a separate noise, unlike op amp inputs, which have the two separate noises.

 

What value of resistance did you use to calculate the resistor noise?

The resistance used is Rin||Rfb=Req=6.67kOhm. The formula used is Enr = sqrt(4*k*T*Req*NBW).

I've attached the spreadsheet if you're interested. It is in ods format from LibreOffice. Let me know if there are compatibility issues and I'll upload it in a different format.

Note that there's a sheet for inverting and non-inverting opamps. After I was done I noted that the noise calculations are identical. On the positive side of that I now understand when people say that inverting amplifiers are noisier than non-inverting amplifiers, that is, for the same signal band width.

 

Resistor current noise is just its voltage (Johnson) noise divided by its resistance.
It's not a separate noise, unlike op amp inputs, which have the two separate noises.

Thanks. I do remember reading something about technology dependent resistor noise. Perhaps then that was referring to pop-corn noise or some other noise not yet considered here. Or maybe I'm confusing specs.

 

. I do remember reading something about technology dependent resistor noise

Yes.
For example, the old carbon composition resistors are generally noisier than film or wire-wound type resistors for a given resistance value.

 

Thanks. I do remember reading something about technology dependent resistor noise. Perhaps then that was referring to pop-corn noise or some other noise not yet considered here. Or maybe I'm confusing specs.

Thin film are better than thick film. Thick film (which are Ruthenium Oxide) have some voltage dependence, and are a little noisier.
In truth, choosing the wrong type of resistor (from what is currently available) makes only a tiny difference compared to choosing the wrong type of capacitor.

 

it is very easy for the current noise density multiplied by the resistance to be the dominant source of noise.

I had misinterpreted this. There's no additional current noise, as crutschow later also notes, it's one and the same thing for resistors. So then we can make Rin/Rfb lower values so long as the op-amp has enough drive current and compare resistor noise to op-amp noise. Then we can achieve better noise performance at the expense of power consumption.

Thanks you, gentlemen.

 

I had misinterpreted this. There's no additional current noise, as crutschow later also notes, it's one and the same thing for resistors. So then we can make Rin/Rfb lower values so long as the op-amp has enough drive current and compare resistor noise to op-amp noise. Then we can achieve better noise performance at the expense of power consumption.

Thanks you, gentlemen.

It’s when you have a high source impedance that you really have to think about it.
Otherwise, lowering the feedback resistors continues to reduce the noise until the op-amp runs out of drive.
I find it remarkable that the 45-year-old NE5532 seems to have the right balance of current noise and voltage noise to achieve excellent performance, and very few newer devices have surpassed it, and it’s now really cheap

 

It’s when you have a high source impedance that you really have to think about it.

A really good point. I ran into it doing a sim with cascaded op-amps. The results was strange until it occurred to me that the coupling capacitor between the stages has a high impedance at low frequencies.

 

A really good point. I ran into it doing a sim with cascaded op-amps. The results was strange until it occurred to me that the coupling capacitor between the stages has a high impedance at low frequencies.

High resistance generate noise, high reactance don’t.
But if you have a circuit with a high resistance to ground, and a coupling capacitor, then the high resistance is only shunted by the low resistance of the previous output at frequencies above the cutoff frequency of the high-pass filter that it forms, leaving the high input resistor to generate noise at frequencies below that.
This is particularly true on magnetic cartridge preamplifiers, where the 47k input resistor is normally shunted by the 600Ω resistance of the cartridge, but above 150Hz the inductance of the cartridge removes that shunt.

 

Hi,

Consider a standard op-amp circuit (example) either inverting or non-inverting. Ignoring the 1/f noise then the

noise gain = 1 + Rf/Rg
noise band width = Gain band width product / noise gain
input referred RMS noise voltage = input referred noise density * noise band width

which is white noise. What is the nature of this noise? Is it spread out across all of the noise band width?
For an audio application, all else being equal, does a 20MHz GBW op-amp produce more audible noise than a 10MHz GBW op-amp?

I read that you can create another pole, for example with an RC filter on the output, thereby limiting the noise band width but is there any point to that if we can ignore noise above ~20kHz? If it's true that we can ignore noise above 20kHz then we can you use 20kHz as NBW for the calculation and not bother with the RC filter on the output?

Hi H,
I know very little about opamps, but somewhere rattling around my head is where two opamps were in parallel, but opposite, for cancelling out noise.
If this is all wrong, then just ignore it.
C.