Transimpedance amplifier

Are you using the pmt in current mode? Or are you trying to see the pulses? (60k ohm and 4.7pF has a ~0.3 us time constant.)

George H

I am using the PMT in analog / current mode - no pulse counting required. The output will be sampled at 500 kHz; I tried to put the corner frequency at roughly that value to optimise signal to noise. I can lower the 4.7pF though; it needs about 1pF for stability.

That corner frequency may be right. It may be sadly wrong. It depends on the problem at hand.

I suggest that instead of randomly picking a corner frequency based on sampling rate, you do so intelligently, based on the sampling rate and your expected signal spectrum:

The circuit itself looks correct to me. C29 and C30 look to be a bit of overkill unless they're at board level.

If you really DO feel you need that much supply decoupling then you may want to consider small resistors in series with the power leads before you get to the decoupling caps, or local regulation, or other such "oddball" stuff.

OK, 100uA sounds like a lot of current for a pmt... I would have guessed

George H.

Thanks for the white paper on sampling, it is very detailed and contains a practical no-nonsense approach to sampling. I like it.

The section on time-shape response (p.19-23) sampling is applicable here. The setup will be used in a scanning microsopy application. If the frequency content is too high we can lower the scan velocity and thereby lower the maximum frequency content. The 4.7pF was chosen as a compromise between bandwidth and noise - the exact number is a bit arbitrary.

Right, 100uA is too high. The 100uA has to come out of the last node in the divider chain, without changing its voltage. Bypass caps on the last few nodes helps to allow for higher pulsed currents.

The PMT is a Hamamatsu H10720. The 100uA max. signal current comes from their specifications, see [1].

Looks not awful. Tim's comment about choosing the bandwidth carefully is relevant. Once you've sampled, all the out-of-band noise will be aliased down into your signal bandwidth.

You should try to minimize the cable length, and watch out for ground loops. Not only does the cable leak (so that it'll collect interfering signals), but multiple grounds will cause all sorts of nasty hum in your data.

A more subtle problem with cable length is that the cable capacitance greatly increases the high frequency noise. The cable capacitance differentiating the voltage noise of the op amp contributes

i_Nc = 2 pi f C_cable e_N.

With a metre of 50-ohm cable, C_cable ~ 100 pF, so at with 4 nV/sqrt(Hz) at 500 kHz you'll have an extra current noise of 1.3 pA/sqrt(Hz). That dwarfs the thermal noise of the feedback resistor, but is probably much less than the amplified shot noise of the primary photocurrent. If you're running with a gain G, the amplified shot noise at the anode is

i_Npmt = G * sqrt(2 e I_primary) = sqrt(G * 2 e I_anode)

With G = 100k and I_anode = 100 uA (which I agree is a lot), you get

i_Npmt = 1.8 nA/sqrt(Hz) .

So the noise peak is not a SNR worry, but it may be a stability worry.

I might put in some protection diodes to the supplies from the inverting input, and a bit of series resistance (maybe 1k, depending on your cable capacitance, or maybe a couple of LND150s in series) from there to the PMT. That'll help the amp survive things like ion events, or some moron^H^H^H^H^H grad student opening the light shield with power applied.

Cheers

Phil Hobbs

The PMT will be mounted directly on the PCB with the transimpedance amplifier directly connected to it. I expect no issues with cable capacitance.

Thanks for the hint on protection. I'll include a BAV99 to the power rails and then a 1k series resistor.

It'll shorten the tube life fairly drastically, though, especially in such a small size (8 mm diameter). If this is going in a product, be sure to characterize that. PMTs (and especially microchannel plates) have two aging mechanisms: diffusion of gas and caesium, which happens even in the dark, and dynode wearout, which puts a limit on the total integrated charge you can get from the anode over the life of the tube. It's typically a few hundred coulombs per square centimetre of photocathode, so that little 8-mm tube run continuously at 100 uA will die on a timescale of

t_death ~ 200 C * pi * (0.4 cm)**2/1e-4 ~ 400 hours.

Given what Hamamatsu charges, those are pretty expensive photons.

The Hamamatsu modules use Cockroft_Walton generators, which produce a lot of switching noise, but have the advantage that the lower taps (near the anode) have much lower impedance than the upper ones (near the photocathode). That makes them a good match for the PMT application.

Cheers

Phil Hobbs

On a sunny day (23 Mar 2015 10:27:31 -0700) it happened Winfield Hill wrote in :

There is not always a 'divider chain', for example I use several taps on a voltage multiplier, exactly to maintain constant voltage:

But 100 uA? I do not use constant current either.

OTOH 100 uA this way at say 1 kV is only 100 mW.

This is data for the XP2342 PMT I have: Limit Hard to tell without seeing the PMT specs.

Anyways, resistor dividers suck, use power, and change voltage under load.

Maybe voltages can be made more ripple free that way, but I am not sure. You could also add a LPF at each dynode.

Maybe Phil Hobbs should answer that...

Nice little PMT, can you share with us the cost? (in 1's to 10's)

George H.

They are available from Edmund Optics starting from $750, cost depending on the type of photocathode material. If you can buy from Hamamatsu directly you can get it a bit cheaper (~10%).

Thanks, (I'm guessing Edmund's mark-up is more than 10%.) I'll have to send Hamamatsu an RFQ.

Re: 100uA, If it was me, I'd consider scaling back the current, or pick another detector. (What are you doing?) If you can afford the loss in BW it would be easy to limit the current with more gain in the TIA.

George H.

The cost I mentioned was for a single unit. Probably could get -50% at large volumes.

It is for a scanning microscopy setup (confocal); each sampling point is a (gray scale value of a) pixel. Some pixels might have a instanteneous current of 100uA but most of them will be well below that. I can adjust the gain of the PMT to set the average output current, which will be about 5-10uA.

I'll probably leave the gain as-is, build it and get some real application data. If I need more gain I can easily adjust the circuit.