Chopper Amp Noise

[Samuel Groner]

Hi

Not directly audio related, but I hope to get some expertise here nonetheless.

While searching for low noise chopper amplifiers I came across a design by Jim Williams which claims 40 nVpp (0.1-10 Hz). There are at least four different implementations published but let's focus on this one: LTMag-V16N01-11-FETAmp-Williams.pdf.

I'm puzzled about how this very low noise figure is achieved. The text notes that This low noise is attributed to the input JFET's die size and current density. Or in other words, attributed to the low voltage noise of the LSK389. But--there is a ~10k feedback network (at the output of A1) right in series with the JFET. Doesn't this entirely spoil any attempt for low noise..? And then there's a 3k in series with the input. My calculator says 22.1 nVrms for a 9.9 Hz ENBW, so easily enough to prevent 40 nVpp overall performance.

Let's say the 3k is just a typo (other implementations show anything from zero to 300 Ohm). But what about the feedback network? I realise that a chopper amp has special properties regarding 1/f noise, but if the 10k feedback resistor doesn't contribute noise why should the JFETs? Any help?

BTW, any pointers to other interesting (discrete) chopper schematics welcome!

Thanks,
Samuel

 

[Svart]

The text notes that This low noise is attributed to the input JFET's die size and current density. Or in other words, attributed to the low voltage noise of the LSK389

My only guess is that they are counting on this having low current noise, look how light the loads are.

Take a look at the feedback from A1.  It looks as if they intend to negate any FET noise by using a secondary fet in the feedback to the NI pin.

Anyway, I'm just as confused about this as you are.  Sometimes I see app notes that make me scratch my head and ask myself if they've actually tried their circuit.

They don't even specify what types of resistors are used in the circuit beyond a few choice ones.  I would use thin film or metal film SMD types for lowest flicker noise.

 

[PRR]

> there is a ~10k feedback network (at the output of A1) right in series with the JFET.

No, no, no. There is 10K plus 10 ohms divider.

Noise of 10K gets charged to the output. Noise of 10 ohms is charged to the input.

Same as a Greene mike preamp. Two 10K resistors from output, a shunt resistor which for high gains is under 100 ohms. The <100 ohm is less than mike resistance, so noise is small.

The 3K is odd and I don't know about it.

> scratch my head and ask myself if they've actually tried their circuit.

But this is Jim Williams, who is just about smart enough to know how dumb he is. He left an EE curriculum and was chained to a service bench for years. You can't B.S. a sick 'scope, you have to know what is really going on. Williams is actually smart enough to make-up answers which are 99.44% perfect, but his background means he won't write-it-up until he has seen it in action.

And also there is a photo. Pictures don't lie. (?)

Typos by editors and draftsmen are possible.

 

[Samuel Groner]

Thanks for your response.

My only guess is that they are counting on this having low current noise.

Current noise surely is a consideration, but the LSK389 is not a smart choice at several hundred pA gate leakage. I can easily make up a dual JFET with two orders of magnitude lower leakage/current noise.

It looks as if they intend to negate any FET noise by using a secondary fet in the feedback to the NI pin.

You can't cancel noise that way, the contributions of the two halfs of the LSK389 add up (as square-root sum). The LSK389 plus A1 is simply a simple hybride opamp.

No, no, no. There is 10K plus 10 ohms divider.

No, no, no. Look again--it is 10M plus 10k!

Pictures don't lie.

As do statistics ;D.

See also AN-93 (page 9ff) and Ultra low noise amplifiers.pdf (page 4ff). These implementations make sense to me. All series noise sources (switches, AC amplifier, feedback networks) are low impedance/noise.

Samuel

 

[Samuel Groner]

Wait, I think I've found the trick: the second reference linked in the previous post states that the bandwidth of the amplifier is just 0.2 Hz; this of course means that thermal noise is pushed down to a region where 40 nVpp is possible. And it also questions how much sense it makes to measure noise of such an amplifier with a 0.1-10 Hz bandpass. The 0.1 Hz choppes off half the bandwidth...

In any case I still wonder how one can use such low noise JFETs and then slap thermal noise from 13 kOhm in series with it; the reference mentioned above shows that it is easily possible to do much better.

Samuel

 

[xmvlk]

But--there is a ~10k feedback network (at the output of A1) right in series with the JFET.

Exactly. This is dominant input noise resistor. Try to remove FETS to anything and
look how noise figure does not change.
I do not understand why they do not chop TL072, maybe they get similar performance
if hey wants ONLY DC zero point stability. They may spare components and power with
near the same "performance".
Maybe I would change that 10 k resistor by short (1u directly to gate) or {simplier topology} realise chopper error amplifier unsymetrically with one fet. It may decrease the noise by 3 db.