Capacitor-feedback for low noise

Hello Zigoteau,

Be careful. Depending on what type of core they have that might not always be the case.

Regards, Joerg

That would be nice.

Or you could charge pump: zot in one small unit of charge just after you digitize each sample, gated by the software when it's in the mood. It's then pretty simple to compute the amount of charge very precisely (just average a lot of shots) and the charge injection will be very low-noise (if you do it right.) One of those old Jim Williams/Bob Pease v/f converter tricks should work.

Hell, that might even be practical.

John

Hello John,

That reminds me of a strip chart recorder from the 80's. It's reset button literally shorted +5V (!) to GND. Bzzzt.... worked every single time until the switch contacts gave out.

Regards, Joerg

I read in sci.electronics.design that Joerg wrote (in ) about 'Capacitor-feedback for low noise', on Tue, 23 Aug 2005:

That should allay your anxiety.(;-)

Hello John,

It does but it's tough to get small qties for a prototype build. Same for Euro parts, just got a mighty black eye there. Took me hours on the phone to get the parts.

Regards, Joerg

Hi Jeroen,

Many thanks for the references.

I've been doing a bit of thinking - quite painful when it comes to noise analysis of multinode circuits. While your examples are invaluable, I will have to understand the theory to optimize the design for my particular application. Am I on the right track?

My requirement is for a transimpedance amplifier, whereas your example has a high-impedance input. I think that my first stage must be configured as a current amplifier, with conversion to voltage occuring first in the second transimpedance stage. Compared to your circuit, my bandwidth requirement is much more modest, and I plan to use low-current-noise op amps for both stages.

Using op amps, which are inherently voltage in-voltage out, rather than discrete components, I have to deviate significantly from your configuration, but I think I have come up with a viable one, where the transformer primary is between the input and output of the stage 1 op amp, and its secondary, with much fewer turns, feeds directly the input of the stage 2 op amp with feedback resistor Rf2. Seen from the stage 1 op amp, the feedback circuit is essentially the magnetizing inductance M of the transformer primary, in parallel with Rf2*N/A2, where A2 is the DC voltage gain A2 of the second op amp and N is the turns ratio of the transformer.

The low-frequency roll-off happens at the break frequency of Rf2*N/A2/2/pi/M, and there are no stability problems at that frequency. The stability problems occur near the frequency 1/2/pi/sqrt(M*Ci1), where Ci1 is the input capacitance of op amp 1. This gives a resonance in the feedback fraction, which must be damped by a resistance somewhere. I think the only way you can provide this resistance without contributing to the input node noise is via the apparent resistance Rf2*N/A2.

I think that the noise generated by this apparent resistance Rf2*N/A2 is much less than the Johnson noise from a real resistor of this value. It will essentially be the total noise current at the input-node of op amp 2, divided by N. Do you agree?

The resonant frequency 1/2/pi/sqrt(M*Ci1) provides a strict upper limit to the bandwidth of the overall amplifier, and it might be hard to get within an order of magnitude of it. This puts an upper limit on the magnetizing inductance M of the transformer.

Thanks again for your excellent suggestion.

Best regards,

Ziggy.

In article , Zigoteau wrote: [...]

Insert the word "ideal" in front of both of those. Many capacitors and inductors make nice little microphones.

All the high K ceramic capacitors have this problem. Stick with NPO/COG in low noise circuits, or don't try to go surface mount.

Inductors make noise due to stress on the cores and also pick up external fields.

The resistance of an inductor is noisy like all resistances.

In article , John Larkin wrote: [...]

Someone, I think it was IBM, made a device some years back that could gate through a single electron at a time. You need one of them and a very jitter free clock.

BTW, which Vitrovac core did you use in that amplifier?

Sure, here is one:

J. Lepaisant, M. Lam Chok Sing, D. Bloyet, "Low-noise preamplifier with input and feedback transformers for low source resistance sensors" Rev. Sci Instrum. 63(3), March 1992, p2089

This is basically a differential amplifier with series transformer feedback in the input. Its bandwidth is 5Hz-100kHz. It has 65pV/rtHz and 1.5pA/rtHz noise, so you'll want to change the transformers to trade current noise against voltage noise.

In a project of mine, I used a different scheme, a folded cascode with transformer feedback to the sources of the input FETs. Here, the bandwidth is 10kHz-75MHz and the noise voltage is 650pV/rtHz. I did not measure the noise current, but it should be in the 30fA/rtHz ballpark. You can see short description at:

There is quite a bit of flexibility trading off voltage against current noise, and there's the bandwidth to play with too.

Jeroen Belleman

Thanks, Jeroen.

What were your selection criteria, size and permeability? Is that an obsolete part now? I didn't see any that small.

I used the T60009-E4006-W650. That's a 6.5x3.5x2.1 mm (od x id x h) tape-wound toroid core with Al=13uH/turn^2.

Jeroen Belleman

Hola now! I have no time now to work out if your line of reasoning is correct, alas. Can we pick this up again next Wednesday?

Cheers, Jeroen Belleman

A thousand? Sheesh, what are you going to do with a thousand cores? Give some away? :>)

Size and permeability, exactly. Despite the very high permeability, you can make transformers that work up to a few hundred MHz with these cores.

I don't think the cores are obsolete, but it's true I've been unable to find them on their web site. I last ordered a thousand in June, from Sekels GmbH, Dieselstrasse 6, 61239 Ober-Morlen, Germany. They dropped on my desk just today. I paid 2.72 Euros/pc. They used to be cheaper.

Regards, Jeroen Belleman

Sheesh, John, I found the website and I must be 5x further away from it than you!

Well, could you spare me a dozen?

Well, they work so well I tend to end up using them in many applications where I used ferrite before.

Perhaps you remember the thread some years ago, where I reported on my attempts to build a four decade bandwidth hybrid transformer? Using these cores, four decades was a snap. In fact, I got 5kHz-500MHz (-3dB, 50 Ohm). I had to use two network analysers to cover the bandwidth :-). Another neat device was the 7 decade bandwidth 1:4 transformer.

I'm in the process of building 150 amplifiers that contain two of these cores each. I also use them to make more hybrid transformers and there's another 150 going in there. I could go on.

In fact, I believed that the thousand I bought in 2002 were a lifetime supply, but we've been going through them faster than I thought.

Regards, Jeroen Belleman