Help designing a low-noise TIA

ll. The original Tolman-Stewart used this rotating coil and braking; they re-did their experiment about a decade later with a constantly rotating sys tem and then everything moved to ac and was easier. The recent papers from the 80s used the ac system. The problem is, the physics is harder, and we want this to be for undergrads. So we like the braking experiment better.

onitors and running at night. Ideally, we?d like to be able to run this as part of our undergraduate advanced lab, so making it an extremely low-noise environment is not ideal. But we can try. We?ve actuall y tried mu-metal, and it is not great at keeping out the magnetic fields. We?re currently considering just adding a 60 Hz notch filter.

t now the noise swamps the TIA, and brings it to the rails.

d a volt or so. We?re working with low-noise op-amps, and so they are not trimmable. We were using the LT1792 and that one we could trim and that worked well. We can and have tested the circuit with an actual hones t-to-goodness current source, and with two coils, where we send a pulse int o one coil and pick it up in the second coil. The first experiment works w ell, the second is inconclusive so far.

rotary feedthrough that one of my students found on the web:

formatting link
rcotac.com/html/110.html . The stationary coil, we just bring the contacts out and onto the bench. None of the electronics is attached to the coils, it is all on the bench. We connect our two coils in series when the wires come to the breadboard.

Mercury? Gallium if you heat it a little, I wonder if there is a temperatur e spec.?

Oh dear, don't throw another monkey wrench at it. Maybe the thermal times are long enough so that he can look at differences between the two spin directions.

And no one's mentioned 1/f noise. :^)

George H.

ed to integrate the current pulse. Our thought was that integrating would allow us to integrate over the noise and be left with only the slower 0.5 s pulse. But in order to get a large current, the input impedance of the me asurement system has to be small. This is why the TIA, with its virtual gr ound at the negative input is nice, acts as zero input impedance for the cu rrent pulse. What would be the input impedance of your transistor circuit be?

ribed, but without the switch. So to prevent pinning the op-amp we added a feedback resistor also. The system we had was very unstable, even for sho rt periods of time. But we did not try it with a switch to keep the capaci tor for charging. You are right that the math is easy if we can integrate the current to get the charge. Then we only need to know the starting spee d and the total charge.

Can those be built to reject the 60 Hz noise?

Reply to
George Herold
Loading thread data ...

ed for the op-amp drift.

on their website. We moved to a liquid metal rotary feedthrough (the liqu id metal is mercury) to remove problems with noise that comes from brushes. I don?t know what the noise specs are. We are set up to run it a ll using wire that twist up, but that is hard to run tests like that. You get one try, and then have to replace the wires.

current by a factor of ten or more. So I am not sure your design will work . We?d drop our expected current down to 100 pA, or less.

as hard to maintain and did very little for the static (Earth?s) fi eld that we also have to get rid of. (If you don?t get rid of the Earth?s field, when the coil is spinning the centrifugal force make s the coil slightly bigger, and then when you brake it it gets smaller, lea ding to a signal that looks like the one we are trying to see.) So we aban doned that. But we haven?t tried electrostatically shielding the e ntire setup. We put our breadboard into a metal box to reduce noise ? ? are you suggesting we box up the whole thing into an electrostatic box? We had a real one of those when I was in grad school, a screened room. I s that necessary for the coils as well?

actual V vs t of the pulse on the oscilloscope screen for the students to b e able to see it before digitizing it. But if I have to digitize and then play with the signal, I will.

So how would you sense it? Say using opamps* on the front end?

One issue with his compensating coil is that it has the same R and L as the spinning coil (I assume)... I was going to say that cuts the voltage signal in 1/2.. but that might not be true.

George H.

*as a fellow physicist, discrete front end design is not our forte. (Excluding Phil H. :^)

Reply to
George Herold

"mercotac" is a subtle hint.

One more hardly makes a difference.

I have, several times.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

AD8429. I'd spin it and bring out a big signal.

It won't. It would cut the current in half if you use a TIA and put it in *series*.

Process the comp coil separately.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

On a sunny day (Fri, 07 Jun 2019 06:52:55 -0700) it happened John Larkin wrote in :

He wrote: >Jan Panteltje, several kOhm of input impedance will reduce our expected current

He says 100 pA, dunno why u use volts.

Really?

Did you try that?

Bad specimen? O/C?

Reply to
Jan Panteltje

On a sunny day (Fri, 07 Jun 2019 15:49:26 +0200) it happened Jeroen Belleman wrote in :

Yes Zout is not infinete, but depending on the tranny not as low as 10 k.

Reply to
Jan Panteltje

On a sunny day (Fri, 07 Jun 2019 07:09:14 -0700) it happened John Larkin wrote in :

What spice?

Reply to
Jan Panteltje

I'm not sure how to calculate the input impedance.. (I know, look in AoE.) But isn't the problem the emitter resistance. r_e = 25 mV/Ic At 1 uA of Ic, r_e = 25 k ohm and the gain is R_c/ r_e = 400.

Jan, Do you know the Ebers-Moll model of the transistor?

George H.

Reply to
George Herold

LT Spice. It's free and fairly easy to learn and use. It includes a lot of transistor and mosfet and diode models.

I rarely do much of the old classic EE math any more... just guess then Spice it. I even Spice voltage dividers and RC timers and such.

I just did an RLC to go between an opamp and a 40 Ms/s ADC input. The ADC data sheet wants an RC, but the time constant was too slow for my purposes. A little inductance snaps it right up.

And I had another case: we're redesigning a laser controller, for IC lithography, that we did in 2002. We used (at the customer's request) a Maxim tapped silicon delay line to generate some asynchronous delays. The part is naturally long gone, so I'll use a triggered oscillator and some FPGA logic to do the same function.

This took minutes to simulate:

formatting link

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

Dynamic resistance Re is 25K and Rb is beta times that.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

Oh, thanks. That's fairly obvious, now that you've told me. :^) GH

Reply to
George Herold

That means doing an UNCONTROLLED experiment, and sorting through lots of data to compensate. The demonstrate-to-the-audience value is nil.

When a single 'run' takes twenty seconds and produces one number (or two, one from the moving coil and one from a nearby stationary coil by way of nullifying interference), signal-averaging just means making a histogram of a few dozen trials. We all know what the interference is, don't we? Fluorescent lights have AC voltages on the ionized gas in the tube, all the wiring in the walls makes E and B fields because that's how the energy is transferred, and big metal moving through Earth's magnetic field makes ripples in the distance. So, turn the lights off, ground some chickenwire around the apparatus, don't let vibration act on any ferromagnetic bits (you can get bronze nails and form lots of mechanism in wood and copper).

Again, it is sometimes effective to compensate/correct/calculate, but the deceleration-rate is easy to reject, it complicates the result, so why NOT reject it? You can easily brake from multiple different initial rotation rates to show the pulse is proportional, it does NOT take a microsecond-by-microsecond breakdown to do that.

If you do zero-velocity trial spin, you get a zero voltage at the opamp output. You would certainly accept that. And for noise, the histogram of zero-velocity trials should give a zero-centered histogram of outputs, while high-initial-velocity trials would not. You can also reverse the spin direction.

Friction isn't noise-free, repeatable, easily modeled, so what does 'properly shaped pulse' even mean?

Reply to
whit3rd

If you put a battery and op amp into the spinning disk, the rotating contacts can have the buffered integrator-capacitor voltage on them, so a few microvolts won't matter. The nanovolt signals don't need to pass through the mystery metals.

Noise only makes a single run uncertain, multiple runs will solve that. PMI's old designs used SiN passivation with SiO2 overcoat, and had very low 1/f noise; those OP27s, and other offerings, are available from AD nowadays. 80nV in 0.1 to 10 Hz bandwidth, and you only really care about 0.1 to 1 Hz if you look at the brake-pulse integration. It might hurt the take-a-thousand-readings schemes, however.

Reply to
whit3rd

It means that it agrees with the physics, namely that the coil voltage is proportional to acceleration.

Other experimenters instrument the angular position of the coil to demonstrate that exact relationship. It's a good crosscheck in a system that typically has a lot of deceptive things going on.

If the signal from the coil is *not* a reasonable pulse that corresponds to the acceleration that the electrons see, something is wrong. A simple integral measurement hides all that.

The pulse would look cool, too.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

And in a voltage amplifier, if you double beta, Rb doubles, so you get half the base signal current, times twice the collector current, which cancel. So voltage gain is independent of beta.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

The noise-impaired coil voltage won't be a clear indication.

Now you aren't thinking like a scientist. 'Something is wrong' means you can test a new hypothesis, but it doesn't tell you WHAT new hypothesis. There's no value in a 'something is wrong' determination unless another hypothesis can be put forward.

Testing for 'unknown' is less than productive, and demonstrates... nothing in particular.

A simple integral measurement is a controlled experiment; such simplicity is the reason we have controlled laboratory conditions. We make simplicity, it's productive to make simplicity, and counterproductive to destroy it.

You aren't welcome in my lab!

Reply to
whit3rd

It will be a clear indication of the noise. If, say, the opamp is railing, or there's a big DC offset, or a lot of hum, the windowed integral is less useful.

I wouldn't submit a paper, or teach a class, if the phenom involved is an artifact.

Seeing the coil voltage would be useful in understanding what's going on. Hiding it as a single integrated number is keeping your head in the sand.

Isn't that what science does, discover things?

I poke around with scope probes and discover all sorts of things.

Do you have a lab? Do you have an oscilloscope?

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

On Jun 7, 2019, whit3rd wrote (in article):

For gaussian noise and the like, averaging does help. For 1/f noise,

One large reason for going from the current "DC" scheme to an "AC" scheme is that 1/f noise falls off at higher frequencies.

The classic scheme would be some kind of torsionally vibrating coil feeding an AC coupled amplifier and a synchronous detector. This approach has already been mentioned and dismissed as being too complex for teaching, but has the advantage of being workable in exactly such a lab..

It occurs to me that a possible arrangement is two torsionally oscillating coils that are coaxial and almost touch, but vibrate opposite to one another. The coils would be wired such that the desired electron acceleration signals would add, while variation due to changes in coil radial size due to centrifugal effect interacting with the local magnetic field would largely cancel.

.

Lower 1/f noise is always helpful.

Joe Gwinn

Reply to
Joseph Gwinn

On Jun 7, 2019, John Larkin wrote (in article):

While we may not be able to put a battery in the rotating coil, it occurs to me that we could put a zero-bias JFET preamp in the coil, power from outside via a twisted pair that also carries the amplified signal. Feed from a constant voltage source and sense the current draft. This way, capacitance in the twisted pair has little effect. And the twisted pair cancels magnetic fields from powering the JFET.

Joe Gwinn

Reply to
Joseph Gwinn

I'd expect that a couple of button-cell batteries or a supercap could ride on the coil. Put the diffamp in the exact center to minimize forces on it.

Heck, put everything along the rotational axis, on a long skinny PC board.

Short the coil and spin it to make sure the amp is not reacting to the acceleration, Hall effect or something. You can make a lot of offset by pressing on the top of an opamp.

--
John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  
 Click to see the full signature
Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.