Capacitor-feedback for low noise

I read in sci.electronics.design that Jeroen Belleman wrote (in ) about 'Capacitor-feedback for low noise', on Wed, 31 Aug 2005:

I couldn't even find the web site!

Well, maybe I made a mistake in the URL, or something. I've lost the URL now.

Thanks. I may have typed 'vacumm'. Or 'vaccum' (a neutered cow!)

It's

Jeroen Belleman

Sure, sure. I took the hint.

Regards Jeroen Belleman

You might mention that the mu-r of unobtanium is negative, so coils wound on cores made of it make excellent capacitors, values from 1 kF to

Hi, Jeroen,

I've looked at your webpage and at the paper whose reference you gave - they're both for voltage amplifiers.

I would appreciate your comments on the idea that the best configuration for a transformer-feedback transimpedance amplifier using low-noise op amps would have two stages, the first being a current amplifier with the transformer primary connected between the output and inverting input.

Cheers,

Zigoteau.

Yes, but you want a voltage-amplifier input stage, see below.

OK, let's speak generally. Transimpedance amplifiers are made with high-gain high-impedance opamps, etc., inside them. If you can find an ideal JFET opamp to make your transimpedance amplifier, fine. As I've discussed many times here on s.e.d., there's a straightforward way to analyze the issues involved in selecting parts. Sometimes you're forced to use a kind of "composite opamp," which has several stages in series. The first stage is optimized for low input noise, high input impedance, and provides some voltage gain (note, it's a voltage amplifier), the later stages provide wideband voltage gain.

For the input stage, you're trading off the primary factors of the JFET's gate-leakage current, voltage noise and capacitance, as you select the JFET. Sometimes you're pushed toward an input stage with less voltage noise than is available in commercial JFET opamps (low voltage noise and low capacitance are two very critical aspects for current amplifiers at higher frequencies, where they combine to create a current noise, i_n = 2pi f e_n C).

Now let's speak specifically. Jeroen Belleman has an input stage using one of the low-voltage-noise JFETs I was talking about. The issue we'll address here is voltage noise. When you add a voltage feedback path to the source of the JFET, the lower divider resistor contributes Johnson noise. Keeping the contribution much lower than Jeroen's JFETs would involved resistor values under about 20 ohms. So if you wanted a gain of 10, your feedback resistor would be 180 ohms, an awkwardly-low value, requiring high current at high output levels. Jeroen elegantly solved this vexing problem with a feedback transformer,

When using such JFETs I have solved this issue with a higher gain of 100, which allows using reasonable-value feedback resistors, like 20-ohms and 2k. It also allows DC gain. But my application did not require the high 75MHz bandwidth that Jeroen's did. If he had used a gain of 100, he'd have suffered a reduced bandwidth.

Note, you could create an input stage without any internal voltage feedback to the JFET, getting as much gain as it gives, and thereby avoid a feedback-resistor Johnson-noise problem. But unpredictable gain might create other problems elsewhere in a closed-loop design.

Hi, Win,

Many thanks for your input, Cheers,

Zigoteau.

I must not have been very clear, this is the JFET configuration I was talking about, thanks for the drawing! Except of course, you meant Q2 to be a PNP, as I marked above. To see the problem with this circuit, consider the value for R2, and the 0.65nV spot noise of Jeroen's JFETs. R2's noise adds (sqrt sum of squares) to the JFET noise. If we choose say 0.5nV as the noise budget for R2, to avoid too much noise increase, then R2 has to be no more than 15 ohms! Do you see the problem now? To get a wide 75MHz bandwidth Jeroen specified a gain of 10x, but this would mean R1 would only be 135 ohms! You see the problem now, right? Well, Jeroen's problem anyway; We don't need a 75MHz bandwidth and can use a higher preamp gain. For 100x, R1 = 1.5k, which is fine.

Not necessarily true at the highest frequencies, where the post-JFET stage gain is low, yet full output may be desired. The gains can be setup differently, with less JFET-stage gain and more gain in the output stage at high frequencies, but then you end up with increased high-frequency noise, e.g., as in the LT1028 opamp above 200kHz.

Yes, exactly, but Jeroen's ac-coupled transformer version can't.

Right, that's why I like to run the JFET preamp stage in its own feedback mode, even though it's inside a larger feedback loop.

In article , Winfield Hill wrote: [...]

You can also take the feedback from the output of a second stage. An over simple example:

Note +V must be very quiet.

In real life, the circuit ends up a lot more complicated than this but this drawing shows the concept. The overall feedback of the transimpedance amplifier trys to make the output of this section have zero swing so the output swing of this section only really has the high frequency components on it.

If your input is such that the circuit always stays in the linear mode, the VX of the JFET stage can be servoed to bring the gate voltage to run at exactly zero. This lets you get a DC accurate and low noise input at the same time.

Make that "might" into "is almost certain to at some time". The circuit bandwidth ends up different from unit to unit and depends on temperature.

Yes.

Yes the PNP is the correct transistor in this case. The circuit is also an over simplified version of the final version.

No, not exactly. R2 does have to be 15 Ohms. That is correct. See below:

No, this where I don't see that there must be a problem.

First off, R1 = 135 Ohms does mean that Q2 is running at a large current if we assume something like (oh, lets say) 13.5V across it, we will have

Yes that is a lot of power but that is why Gawd invented the heat sink. I don't remember the power budget being limited. Was it? 1.35W is not that hard to get rid of.

If we assume the JFET is running at an ID of about 10mA it only implies a HFE in Q2 of 10 or more. Q2 doesn't need to be a Darlington to do that.

Does the OPs case require full swing at the high end? I don't remember.

Yes, a prpblem that bit e before LT added information about that to the data sheet.

It may be posible to arrange things so that there is a DC path and an AC path provided the thing doesn't ever go non-linear. Harris used to make some fast op-amps like this. They had two amplifiers inside. The specs looked very nice but never mensioned the overload recovery which was dreadful. We tried them in a charge sensitive amplifier the results were not good :<

I learned the hard way with a pair of 2SK170s leading into an op-amp.

Sorry to have been a bit long responding. I can't be reading usenet all the time...

So, if I correctly interpret your (more complete) description of August 24, the circuit would look like this:

So ignoring the bandwidth limitations due to the shunt inductance of the transformer and its under-unity coupling factor, and also ignoring the limitations of the amplifiers, the transconductance of the overall circuit would be (N A2 R)/(1-A2). Its input impedance (R N^2)/((1-A1)(1-A2)), and its input-referred noise current sqrt((4 k T)/(N^2 R)), ignoring the contributions of the active circuitry.

So the transformer indeed confers a noise advantage: The transconductance scales with N and the noise current with 1/N.

If this had been the classical single-stage transconductance amplifier, with N * R as the feedback resistor to get the same transconductance as in the above two-stage circuit, the noise current would have scaled with 1/sqrt(N).

So, yes, I think this shows promise. Now to turn this into a working circuit...

Best regards, Jeroen Belleman

^^^^^^^^^^^^^^^

transresistance, transresistance...

Sigh! Jeroen Belleman

Hi, Jeroen,

Thanks for getting back to me with your detailed comments. Worth waiting for.

Yes. In fact my initial ideas were slightly confused, and I have realized that this is not classical negative feedback. The more conventional configuration would in fact be:

N +--------UUUU---GND +-----R--+ | |\\ ==== | |\\ | Iin -+--|-> --UUUU---------+--|-> --+--- out |/ 1 |/ A1 A2

I have part-analyzed both circuits, but have not yet decided which is better.

I liked the idea of your high-mu toroidal cores. My experience with ferrite cores is that it is extremely difficult to get more than three orders of magnitude bandwidth.

I have been in touch with Vacuumschmelze, who say that the T60009-E4006-W650 is still on their books. The salesman apologized for the search facility on their website, which does not yet find catalog numbers. There is a minimum order, but they will send me a couple of samples, one slightly bigger than the T60009-E4006-W650. By the way, they say to get in touch and they will buy you a beer to thank you for the recommendation.

I will have to think about trade-offs and such, but initial calculations suggest that, as a result of the high impedance level, the primary may have to consist of 10000 turns or more of very fine wire. I do not fancy the job of winding by hand. To test the concept, I may have to accept a lower 3-dB frequency somewhat higher than the final application requires. The alternative, which I have not yet thought through, is to lower the impedance level at the A2 summing junction, by making up A2 from two op amps, playing tricks to achieve stability, perhaps something like.

Errare humanum est, ignoscere divinum . . .

Cheers,

Zigoteau.