I was looking into microphone preamps and came across some circuits which use discrete BJTs in front of opamps such as these two documents from THAT Corp:

https://www.thatcorp.com/datashts/AES129_Designing_Mic_Preamps.pdf ("CFIA Mic Preamp" starting page 46)
https://thatcorp.com/datashts/dn109.pdf

Other than THAT Corp trying to promote sales of their matched transistor products, what is the benefit of fronting an inamp with discrete transistors?

The second document implies there's a noise benefit. It states "...This gain allows the noise of U2A and U2B to be swamped by that of the differential
pair, which results in about 0.94nV/sqrt(Hz).." The opamp in the schematic (assuming the label "4570" means Renesas UPC4570) has a specified noise input of 4.5nV/sqrt(Hz), implying that the BJTs somehow reduce that. Although to me the word "swamped" means "dominates", or in other words that the noise of the differential pair would be much greater than the noise of the opamps.

Does anyone know the mechanism by which the discrete differential pair reduces the overall noise of the circuit? A simple 3-opamp inamp would be much simpler to implement.

 

I was looking into microphone preamps and came across some circuits which use discrete BJTs in front of opamps such as these two documents from THAT Corp:

https://www.thatcorp.com/datashts/AES129_Designing_Mic_Preamps.pdf ("CFIA Mic Preamp" starting page 46)
https://thatcorp.com/datashts/dn109.pdf

Other than THAT Corp trying to promote sales of their matched transistor products, what is the benefit of fronting an inamp with discrete transistors?

The second document implies there's a noise benefit. It states "...This gain allows the noise of U2A and U2B to be swamped by that of the differential
pair, which results in about 0.94nV/sqrt(Hz).." The opamp in the schematic (assuming the label "4570" means Renesas UPC4570) has a specified noise input of 4.5nV/sqrt(Hz), implying that the BJTs somehow reduce that. Although to me the word "swamped" means "dominates", or in other words that the noise of the differential pair would be much greater than the noise of the opamps.

Does anyone know the mechanism by which the discrete differential pair reduces the overall noise of the circuit? A simple 3-opamp inamp would be much simpler to implement.

Simply, you can't get op-amps that good (maybe you can now, but when the THATcorp devices came out, no op-amp could get as low as 940pV/√Hz).
The biggest contibutor to the noise is the first stage, because the noise at its output is the voltage noise multiplied by the gain.
If the gain of the first stage is 1000, then the noise its output will be 950nV/√Hz, which when added to the 4.5nV/√Hz from the 4570 in the next stage results in a total output noise of 950nV/√Hz (because we add the squares and take the square root with uncorrelated noise)
Obviously, if we use the 4570 as the first stage, the output noise is then 4.5μV/√Hz, 13.5dB higher.

This is the original paper
https://leonaudio.com.au/double.balanced.mic.amp.notes.pdf
and it used NatSemi's LM394 devices which were the precursors of the Precision Monolithics' SSM2210, which were the precursors of the ThatCorp devices.

 

If you want to try out the circuit, instead of the ThatCorp devices, get some 2N4403s. They are famous for having really low noise, and a fraction of the price.

 

Simply, you can't get op-amps that good (maybe you can now, but when the THATcorp devices came out, no op-amp could get as low as 940pV/√Hz).
The biggest contibutor to the noise is the first stage, because the noise at its output is the voltage noise multiplied by the gain.
If the gain of the first stage is 1000, then the noise its output will be 950nV/√Hz, which when added to the 4.5nV/√Hz from the 4570 in the next stage results in a total output noise of 950nV/√Hz (because we add the squares and take the square root with uncorrelated noise)
Obviously, if we use the 4570 as the first stage, the output noise is then 4.5μV/√Hz, 13.5dB higher.

This is the original paper
https://leonaudio.com.au/double.balanced.mic.amp.notes.pdf
and it used NatSemi's LM394 devices which were the precursors of the Precision Monolithics' SSM2210, which were the precursors of the ThatCorp devices.

OK I had to go refresh myself on opamp noise gain to understand why the opamp noise would not be amplified in these circuits. At first I thought that because the opamps are part of the gain that noise must be amplified.

But the noise gain of opamps is described in terms of feedback and gain resistors in common configurations. Would it be correct to say this mic preamp doesn't amplify the opamp noise as much as the signal because of the feedback configuration?

Thanks for the link!

On a side note, how did you make that square root symbol? I tried using the LaTeX tag but that put the symbol on it's own line and centered.

 

The noise of the first stage gets multiplied by the total gain. The noise of the second stage only gets multiplied by the gain of the second stage.

That means that you get the best results by using something REALLY quiet for the first stage, and then getting as much gain as you possibly can from the first stage.

The square root symbol (and many other handy symbols) is obtained by clicking on the icon that looks like the Parthenon in the top margin. Don't ask me why it looks like the Parthenon (maybe because a lot of the symbols are from the Greek alphabet). Maybe the moderators know.

 

The lowest noise devices that I am familiar with are made for magnetic tape playback heads. BUT EVERY year has brought lower noise devices. I would suggest checking with"ANALOG DEVICES"company to see what they offer. Certainly they do have detailed specifications of products that are truthful. Other semiconductor manufacturers also, but I am not as familiar with them.

 

There are now several op amps that can get that low, e.g.:

TI OPA891/892 950pV√Hz
TI OPA855 980pV√Hz
Analog LT1028/1128 850pV√Hz
Analog LTC6228/6229 880pV√Hz
Analog ADA4898/4899 900pV√Hz

and other manufacturers...

but you do have to be careful to compare like-for-like in terms of the noise bandwidth those figures apply to 1, 10 or 100kHz...

 

The thing that it often overlooked is the current noise. These devices are only low noise for a low source impedance.
Total noise V=√ ((vn^2) +(in.Z)^2) where V = total noise density, vn=voltage noise density, in=current noise density. Z=input impedance.
Also you need to compare the 1/f noise corner frequency. Below the 1/f noise corner, noise increases at 6dB/octave.

 

True, most of those have current noise around 2.5pA √Hz which suggests an input impedance <200ohm for a total noise figure < 1nV √Hz .

 

" Microphone Pre-amps" covers a rather broad spectrum of applications. If the TS will supply some details about the application it is much more likely that they can get some useful advice. Otherwise it will be random suggestions.

 

True, most of those have current noise around 2.5pA √Hz which suggests an input impedance <200ohm for a total noise figure < 1nV √Hz .

The good old NE5534 takes some beating on noise, because it has low current noise (400fA/√Hz) for a low voltage noise bipolar part. The dual NE5532 is slightly noisier than the single.

 

How do those compare with the "classic " National LM381 tape head playback amplifier??

 

How do those compare with the "classic " National LM381 tape head playback amplifier??

The datasheet says 500nV equivalent input noise in a 10kHz bandwidth, for a 60Ω source impedance. Assuming it's all voltage noise, that's 5nV/√Hz. I didn't remember it being that good!