I'm designing some audio-frequency instrumentation for my workbench and am using TL074s for most opamp requirements. One section requires microvolt measurements and noise is therefore an issue.

It's often difficult comparing opamps on noise and other figures because they're often stated differently in the datasheets, so choosing something else, e.g. OP07, has several unknowns.

A better solution might be a discrete differential preamp using either FETs or bipolars, but a similar problem of calculating comparative performance arises.

Can anyone offer some broad suggestions as to the likely benefits of a discrete preamp, along with such factors as noise versus collector current and so forth?

 

Can anyone offer some broad suggestions as to the likely benefits of a discrete preamp, along with such factors as noise versus collector current and so forth?

I think it's generally pretty hard to beat the noise specs of a low noise op amp with a discrete design.
If the input circuit impedances are relative low (10Ω or less) than op amp input voltage noise is the important parameter.
For example, the low-noise NE5532 has an input voltage noise density of only 5μV/√Hz with an input noise current of 0.7pA/√Hz.

For higher impedances, the current noise can become more important since the input noise current is multiplied by the input resistance to give a voltage noise.
For comparison the TL074 has a noise voltage of 18nV/√Hz but a low input noise current of 0.01pA/√Hz.

The total circuit noise is the combined value of the input voltage and current noises so you need to calculate that to determine which op amp is better in your circuit.

 

Linear systems LSK389A has been used as a discrete front end to op amp.
The subject of beneficial use is specific to an application's requirements.
The extra configuration in order to have access to the transistor comes with an expense in performance.
https://www.linearsystems.com/lsdata/others/LSK389_App_Note.pdf

FET vs bipolar, both can be used but different configuration and layout has to be very good.
" modern opamps are so darned good that improving them in any way is a real challenge. " good, 10dB gain but 30dB might be limit.

 

> The total circuit noise is the combined value of the input voltage and current noises

Ah, thanks. Wasn't aware of the difference. I've dug up several pages with discrete-preamp/opamp designs, but at present am more interested in identifying the relevant parameters. At the end of the day the best solution (for UNDERSTANDING things rather than building things) might just be to breadboard a few different circuits and measure their performance.

Trouble is, once you get down to the limits of your 'scope, meters etc and take into account things like metal-film vs carbon film resistors, it becomes difficult to decide which are significant contributors and which are not.

The MF/CF choice seems simple: MF resistors are much quieter, I believe. As to FET vs bipolar I've no idea. As you say, modern opamps are so darned good that improving them in any way is a real challenge.

> The design achieves very low noise alaskasworld.com floor

Your link is to "AlaskasWorld Airlines is the fifth largest airline in the United States". Error?

> Linear systems LSK389A has been used as a discrete front end to op amp.

Spot on and thanks indeed! Yes, this is what I was after, something right at the bleeding edge of noise performance with test setups I can reproduce.

 

JFETs are quieter for high source impedances, Bipolars are quieter for low source impedances, MOSFETs are just noisy, Valves are quiet for high source impedances but not as quiet as JFETs.
Moving magnet cartridges are the real challenge as they change from low source impedance to high source impedance over the audio range.

 

Thanks for the distinction bewteen JFETs and MOSFETs: corresponds with what I've seen. And for emphasizing the role of source impedance: a very neat summary.

 

Doug Self's book is one of the best on the subject.
http://www.douglas-self.com/ampins/books/ssad3.htm

 

Without knowing your application it's hard to be specific, but before you go discrete look at the low noise op-amps. for example I've used these:

This noise is much less than that from a 1k resistor (\(4nV / \sqrt{Hz} \) ).

 

Thanks for the recommendation and esp the noise figure for the resistor (MF I presume) - a very useful reference point.

And crikey! Yes, the noise figure is astounding! One of the problems is that there are so very MANY opamps out there now it's difficult to know how to sort them. The manufs and distributors do their best with their excellent online selection tools, but its still a job of work going through them all, which is why I decided to investigate the discrete approach.

 

I note that the current noise isn't mentioned until you get much further down the datasheet. 1pA/rtHz isn't particularly impressive, as some of the JFET parts can achieve less than 1fA/rtHz. I note that some manufacturers don't even specify the current noise.
The LT1028 is a decent device for low source impedances, but with discrete designs using something like a ZXT951 you can achieve 200pV/rtHz, as much noise as a 40 ohm resistor. See table 8.1A of Horowitz and Hill.

 

> some manufacturers don't even specify the current noise

Quite so. I'd never come across the distinction between V and I noise until I posted this thread. For that alone it's certainly been a worthwhile exercize.

 

Voltage and current noise are uncorrelated, so that the total input noise is:
Noise = SQR(Vn^2+(Rin.In)^2)
or, including the thermal noise of the source resistor
Noise = SQR(Vn^2 + (Rin.In)^2 + 4kTR)
so the source resistance makes its biggest contribution in conjunction with the current noise, bigger than its own thermal noise.

 

Most grateful for a VERY abbreviated Noise 101 course.

Just as a matter of interest, some of the JFET datasheets specify noise in dB, e.g. 2N5458 from ON Semi quotes Noise Figure at 1kHz to be 3.0dB:

https://www.onsemi.com/pdf/datasheet/2n5457-d.pdf

Relative to WHAT I've no idea. Could you elucidate?

 

Noise figure is the ratio of the noise including the noise from the amplifier, to the noise of the source expressed in dB.
So Noise_at_input = thermal noise of the source = SQR(4kTR)
Noise_at_output = Noise = SQR(Vn^2 + (Rin.In)^2 + 4kTR)
So NF = 10 log ((Vn^2+ (Rin.In)^2 +4KTR)/(4kTR))
You'll sometimes see it expressed as an equivalent resistance
https://frank.pocnet.net/sheets/030/e/ECC88.pdf see Req on second page.

And then there's 1/f noise which is a whole different thing and far less predictable.

 

Phew! It'll take me a day or two to digest all of this, so I'll thank you again, both personally and on behalf of all others viewing the thread. Back in a while.

 

Doug Self's book is one of the best on the subject.
http://www.douglas-self.com/ampins/books/ssad3.htm

I'll second that. I was looking to see if anyone had suggested it and was going to post recommending it but you beat me to it.

 

Graham Cohen's microphone preamp is another good example of low-noise design.
https://www.leonaudio.com.au/double.balanced.mic.amp.notes.pdf
There's a photo of him in his lab in 2008 still using a slide-rule.

 

depending on the actual measurement, you may have to use different front ends. If you're measuring low-to-medium impedance nodes, bipolar input devices are best. Best performance may be achieved by using either a combo of VLN bipolar with opamp, or a dedicated low-Z LN chip, like THAT 1510/1512.
If you're measuring Hi-Z nodes, you need J-fet's. J-FET opamps are generally too noisy (the few exceptions are very expensive), so you should use a combo of VLN JFET's (LSK 389 or 489 - beware that 389, although having less noise, has larger Cgs capacitance) and opamp.

 

For very low input impedances a medium power bipolar transistor such as ZXT951 will give lower noise than a so-called "low-noise" transistor, because of its lower base spreading resistance.
PNP devices should be quieter than NPN because of the higher electron mobility in the N-type base regions, though, by that argument germanium should be quieter than silicon.

 

You should also be aware of the THAT Corporation and all their audio app notes, in particular http://thatcorp.com/datashts/dn109.pdf