The source noise and the input stage noise are both amplified by the gain, as is the signal, so SNR ratio doesn't change, so long as there is enough gain to prevent later stages from contributing significant noise - the worst case would be an input stage with gain of 1 - the next stage will then also contribute similar levels of noise again - in this situation increasing the gain will improve SNR by raising the signal level at the next stage.
Or put another way the input referred noise of the next stage is reduce by the gain factor of the first - so imagine an amp with 2nV/√Hz of input stage noise, 10nV/√Hz of second stage noise, and an input stage gain of 20. That second stage then has input-referred noise of 10nV/20 = 0.5nV/√Hz, which combined with the 2nV gives 2.06nV/√Hz, basically only the input stage is contributing significant noise even though the second stage is a poor performer noise-wise compared to it.
Are you implying that the noise is only present on the non-inverting input?
Both inputs have noise current.
Both inputs have noise current.
I hope not, the 3dB extra noise for a differential amp is because there's two input devices in series (from signal viewpoint).
For gain distribution and noise you normally convert all the noise sources to an equivalent noise voltage first.
The previous post of mine talks of voltage noise only, though similar arguments apply to noise current too. Noise current is easy to tune by setting the bias so its sort of a separate consideration when the bias can be adjusted.
For gain distribution and noise you normally convert all the noise sources to an equivalent noise voltage first.
The previous post of mine talks of voltage noise only, though similar arguments apply to noise current too. Noise current is easy to tune by setting the bias so its sort of a separate consideration when the bias can be adjusted.
I can't agree with your last sentence, because the same bias also affects the noise voltage.
I don’t agree with your posting either.
To convert all noise sources to an equivalent noise voltage, you first have to identify the vatrious noise sources, add them in the poper way and only after having taken those steps you have converted them in an equivalent input noise.
The extra 3dB noise you are talking about for a diff input are already taken into account in the Amp’s spec.
When an amp is specified at 1nV/rtHz it doesn’t mean for each individual input.
But when adding two amps together in a differential way, you will have to add 3dB to their individual noise figure.
Hans
No I don’t, but this is a correct way to calculate noise gain.
Put all the noise at the inverting input and you will get of course the same result.
For an inverting amp with a gain G, noise gain is 1+G.
This is not only true for the noise from the amp but also for the voltage noise from Rg//Rf.
And for a non inverting amp with gain G, noise gain is G.
Hans
Hi edgarels.
There seems something suspect about the biasing on the input network. With bases attached at the input the emitter to emitter voltage is approx. 1.2 volts. However the subsequent transistors requires 1.2 volts base to base for those transistors to operate, requiring that the input devices be saturated in order for the subsequent devices to function.
For the record, I agree with everything @Mark Tillotson wrote except the last sentence of post #23.
You miss the step of transforming them to the input, which is what Mark's post #21 is all about. For simplicity, he only considers a subset of the noise sources and assumes they are already transformed back to the inputs of the individual stages.
The discussion is mostly about discrete design, so there is no Amp's spec.
In any case, when an op-amp is specified for 1 nV/√Hz, there is normally about 0.7 nV/√Hz coming from one input transistor, about 0.7 nV/√Hz coming from the other input transistor of the input differential pair and a very small amount coming from the rest of the op-amp. (Actually, each input transistor usually consists of a whole bunch of small transistors in parallel, but I treat them as one for simplicity.)
I think Edgar came to the same conclusion in post #13.
Besides this, I think the elephant in the room is the noise of the feedback op-amps. They have to work at a very low inverting gain because they are in the feedback, but even then, their voltage noise is essentially in series with that of the input transistors.
Hi Marcel,
I don’t think I did. Adding the noise sources in the proper should be read as transforming them to the input.
Hans
Friis's equation is in terms of noise figures or noise temperatures and available power gains, which gets awkward when there is no well-defined finite and nonzero impedance (usually noise figures and available power gains both tend to infinity when the impedance tends to zero or infinity). Mark's post #21 is essentially the same story for a low-impedance system.