wierd opamps

LTC's LTC6268 competes with Burr-Brown / TI OPA656, and the decompensated LTC6268-10 with the OPA657.

Those are especially useful to make trans-resistance amplifiers, TIA, used with high-speed photo-current signals, etc. See AoE-III Chapter 8, pages 537-552. Or see Phil Hobbs' book, pages 693-714.

The datasheet's current noise plot, which rises with frequency, shows e_n Cin noise: i_n = 2pi f e_n Cin, see AoE-III sect 8.11.3 and E'qn 8.44. The business of voltage-noise turning into a rather nasty current noise surprises people not familiar with the concept.

For example, the amplifier's intrinsic low current- noise density of 7 fA/rt-Hz turns into a high noise of 4000 fA at 50MHz. Their graph should have shown the node capacitance value. We can back-calculate Cin = i_n / 2pi f e_n = 4pA / 314M 4.3nV = 3 pF. This is very low, probably without a PD connected. One thing that makes LTC's new amplifier ICs special are very low input capacitance, which is achieved by bootstrapping the protection diodes and other parts of the circuit to follow the input pins.

This bootstrap scheme also reduces the input leakage current at high temperatures. There are a few other amplifiers that can also do this, see AoE Figure 5.6.

Yup, they've obviously bootstrapped everything in sight. It's a bit of a cheat, really, from a SNR point of view, because the bootstrapped capacitance doesn't load the input but still contributes current noise as if it did.

It helps the loop bandwidth and frequency compensation, for sure. TI keeps pitching its OPA657 as a TIA, but its huge input capacitance makes it a fairly poor choice in reality.

The LTC part is interesting--I'll have to run some numbers on it. The 1 MHz 1/f corner is pretty horrendous--even a pHEMT's isn't that much higher, and they start from a flatband 20 dB lower, which puts their LF noise in the same ballpark.

Thanks for pointing it out.

Cheers

Phil Hobbs

OK, If the current noise is due to e-sub-n-C, in the LT chip. Then why is there nothing similar in the OPA656 data sheet?

I must admit the idea that the bias current is a few fA and the current noise at ~1MHz is several pA strikes me as strange.

George H.

The OPA656 just lets its input capacitance rage, instead of bottling it up. ;) It's still lower than the OPA657's, though, which is weird.

Cheers

Phil Hobbs

No capacitance bootstrapping!

It's a no-free-lunch-ism. An opamp without the bootstrapping can have much lower input current noise, but the higher input capacitance will plop a pole in the TIA feedback loop and make a heap of voltage noise at high frequencies.

OK so It's the bootstrap that folds the voltage noise back in?

Hey what about the opa655... It's only got 1 pF Cin!

George H.

I think the ease-of-use issue is probably the main motivating factor, and getting rid of most of the input capacitance helps. There's still the photodiode capacitance, of course.

I remember, back in the late '80s, being really excited about some new Motorola op amp with high speed and low noise (for the time--probably 30 MHz and 5 nV). Turned out the input capacitance was so horrible that you couldn't use feedback resistors larger than 10k except at really high gain. (I just looked in my 1999 Moto databook, and it wasn't there, so it obviously vanished pretty fast.)

It might be possible to externally bootstrap the part of the input capacitance that goes to the rails (i.e. the CM capacitance) but unfortunately you can't bootstrap the differential input capacitance (between the inputs) without also eliminating the signal! The effects of that on op amp stability might be amusing as well.

Using a single-ended BF862, with an op amp off to the side forcing it to run at V_GS = 0 (i.e. I_DSS) is a reasonable approach. (We've done that together a few times of course--the JL trademarked name is "snooping the input".) You can use the bootstrap as the input to the next stage, which could be an ADA4899 or something like that.

Putting the op amp off to the side as a snooper, isolated by a big resistor, eliminates most of the e_N*C noise, but of course adds the full Johnson noise of the resistor at frequencies within the snooper's bandwidth. To allow the slowest possible snooper, it's good to run the JFET at constant power dissipation, e.g. by DC-coupling the drain bootstrap.

I still miss the LF357 for front ends--10 nV and 2 pF was pretty good for 1973.

Cheers

Phil Hobbs

Yup.

Gone, unfortunately. Nice part.

Cheers

Phil Hobbs

Physicists (excepting you) (see RSI) seem fascinated by differential-pair jfets. Multiply the noise by 40% for no reason!

I think opamps inherently lose out to discretes, when making low noise TIAs. The ESD diodes, the substrate connected to V-, the differential input, stuff like that.

Johnson noise dominates any reasonable design at low frequencies. It's the high frequency noise that's agonizing.

It's amazing how many truly terrible photodiode TIAs are in the literature, and for sale. Bad by factors like 100:1. I've done a few myself, in the ignorance of youth.

Probably they can't afford to upgrade their AoE 1st Ed. ;)

Shot noise, hopefully, at least with photodiodes.

Haven't we all. ;) It's sad when those hard-earned photons disappear under a sea of unnecessary noise. Especially when your thesis is at stake.

Cheers

Phil Hobbs

Hah, the reason the graph has a straight line: check out the draftman's ruler. We give e_n-Cin-noise crossover frequency vs. feedback-resistor Johnson noise, E'qn 8.45, page 539. The equivalent without a feedback resistor is fx = i_n / 2pi e_n Cin. Assuming no external capacitance that's f_x = 1.3fA/rt-Hz / 2pi 7nV/rt-Hz 3.5pF = 8.5kHz. I'm sorry, but that's the reality. Not everyone is aware of this, and the bench measurements aren't that easy.

The explanation is easy enough: the voltage noise on the summing junction causes it to move up and down, and the capacitance on that node needs current to do this, which is supplied by the op-amp's output, and appears as signal.

You should read our extensive discussion of this scene, AoE-III pages 538-552. Figures 8.74 and 8.79 show some actual measurements of this troubling phenomena. Note: Figure 8.74 is for the OPA655, Burr-Brown's precursor to the OPA656. So we're talking apples and apples here.**

The OPA655 datasheet properly shows the i_n = w e_n Cin effect, with a crossover frequency of about 5kHz, above which the current noise rises proportional to frequency.

Sadly, this is with no external real-world capacitance. We show in Figure 8.74 the effect of this capacitance. ** OK, actually Figure 8.74 is with Rf = 1M (and Fig 8.79 is with Rf = 10M).

Ahh... OK nothing to do with the bootstrap then. So If I plug numbers into my fav opa134 i_n= 3fA/rtHz, e_n=8nV/rtHz and C_in = 5 pf(common mode C_in?) I get fx = ~12 kHz...

Hmm.. OK I remember trying to come up with some means to measure the current noise (of "my" opa134). So you are measuring it by loading the inverting input with a big Cap?

I'll have to spend more time with your noise chapter. I'm getting myself a little confused thinking about it. If this HF current noise is really due to the voltage noise, then (it seems to me) the two noises are correlated! (And then I worry about how they should add?)

George H.

Well, ideally they'll be in quadrature at all frequencies, so their dot product will be zero (sin x and cos x are uncorrelated).

Cheers

Phil Hobbs

Ahh, OK. I must admit I find it a bit "mind expanding" to think about the two quadratures of the noise being uncorrelated. I mean (taking shot noise as an example) both signals come from the random generation of one electron.

George H.

They do in the time domain, but the frequency-domain noise is from the whole ensemble. Two electrons arriving at times differing by >> 1/BW make uncorrelated noise too.

Cheers

Phil Hobbs

One solution: use bigger photons. At 4 keV, there's a hundred ion pairs for each X-ray photon. Electronic noise not generally a problem counting those puppies.

Focusing them can be a bit of a problem, though. ;)

(An acquaintance of mine at IBM, Eberhard Spiller, perfected the normal-incidence X-ray mirror, but it isn't for the faint of heart.)

Cheers

Phil Hobbs

On Tue, 05 May 2015 19:46:28 -0400, Phil Hobbs Gave us:

Main feature: First surface was NOT Aluminum, which X-rays pass through like visible light through clear glass.

Probably was Gold.

Den onsdag den 6. maj 2015 kl. 02.20.00 UTC+2 skrev DecadentLinuxUserNumeroUno:

not at 4keV

-Lasse