Page 1 of 1

Versatile preamp for detectors

PostPosted: Sun Sep 11, 2011 6:45 pm
by Doug Coulter
I've been iterating on preamp design lately and have finally come up with something that almost makes me completely happy. Nothing is ever perfect of course. This preamp is designed as a matched filter for signals that have fast rise and slower fall times, that are negative pulses. It's easy to flip this over for the other polarity, and tune it various other ways. I wanted fairly high input impedance, low drain from the supply (presumably a battery for noise reasons), high gain, fast speed, and the ability to run over a wide range of power supply voltages. I got most of that this time. The drain is a little higher than I wish, and it probably cuts off well below a ghz, and the input impedance is only 100k (probably could be raised 2-10x).
Here is the schematic so we can discuss the trade-offs involved in this design -- handy for anyone who might want to make them differently.
Schematic of the new preamp design

To make this operate over a wide supply, and yet bias the output nearly off when there's no signal (so as to get the maximum possible output swing), this version uses a special trick for the bias. Rather than essentially setting the collector voltage of the pnp input, like a normal divider would do, here we use a LED to create what amounts to a current source. The nifty thing about using an LED is that it acts like one Vbe for silicon, plus another voltage that's related to the LED bandgap wavelength. Thus, you get temperature compensation, plus a little voltage over 1 Vbe, so that you can use an emitter resistor in the following transistor to set its collector current (that's the basic idea, due to DC feedback, this gets fancier and biases both stages correctly). The LED gives us around 1.6v, and to bias the input transistor at the desired ~60 uA, we need to "lose" about 1.1v at the emitter some way. At these low current levels, these transistor are getting turned on at a little over .5v Vbe. We get part of that via the 10k in the emitter to ground network. The rest we get when the NPN turns on and draws further current, some of which gets back to the input emitter (the inverting input of this sort-of opamp), so the trick biases both stages. Since we want a very low impedance to ground on that inverting input, we bypass the 10k, and just leave 10 ohms for the AC feedback signal to appear across. This comes from the 1k resistor from the output NPN collector, giving us a theoretical gain of 101 (just like opamp theory) for AC signals.

To increase the ability to drive coax, and to reduce the amount the NPN collector has to pull off ground, it has a 1k to ground, and one as feedback to the PNP emitter, yeidling a net output off ground of about 1.05volts. I'll take within 5% of design center for this! (OK, I'm proud of it, it went from paper to built with zero changes needed, everything right on the bogie.) You want some emitter resistor in the NPN stage, and to allow it to produce some of the overall loop gain, I made that a 47 ohm resistor, which into a more or less 500 ohm collector load, gives that stage a standalone gain of 10, requiring a gain of at least 10 from the input stage. Of course, with a 10 ohm to 10k ratio there - it has much more gain than that for the negative feedback to work with so we get a nice stable closed loop gain of 101 or thereabouts. This gain should extend well into the double digit mhz, but I'll test that. In this case, with the low current draw (70 ua for the input stage, about 1.3 ma overall) that's a trick, and to get that, the board has to be laid out with significant care to parasitic capacity between things, although only the output node actually shows any gain at all over the input signal - so the miller capacity effects are minimal here. That was on purpose. A lot of times the simpler appearing designs actually have a whole lot more going on than first meets the eye -- and this is no exception. For example, had I just used a 500 ohm output resistor on the NPN, straight to the PNP emitter, I'd then have had to use a 5 ohm feedback R, and the cap would have had to be that much better to not make problems. I would also lose the ability to pull the NPN as close to ground as it will now go, as it would never be willing to go below the PNP emitter. So I split up the load R for the output, giving me a little more current with which to discharge the load capacity (that coax) faster. The more you study this circuit, the more things like that pop out. I could cut miller capacity effects further with another bypass cap on the NPN emitter, but that's another part...tradeoffs again. In this case, it wouldn't be the cost of the extra part, but the adding of another highpass, with some other cutoff, that would complicate the phase response a little bit when the loop gain wasn't quite what you'd wish -- and here we have a lot, but none to spare to get the accuracy and linearity we want.

What I don't like about this circuit, but there's only so much you can do at this level of simplicity, is that 47 ohm emitter R in the NPN will potentially allow enough current to pass to fry the thing if the output is shorted, and the bypass capacitor in the NFB loop is too big. With real values and reasonable care, this shouldn't be able to happen, but... The same deal exists with the pnp under conditions of overdrive and too large a bypass cap. I show .1 uf there -- that's far too big, but it was what I could get without crawling out of the cockpit, a real world value might be more like a few hundred pf. You'd like any baseline shift to be able to correct before the next pulse, so this wants to be a fairly fast high pass for most sensors that make pulses. The other thing I don't like was having to use a 10 ohm resistor to be able to use a reasonable feedback resistor for DC biasing and still get a gain of 100 or so. That means that any effective series R in the bypass cap will cause a loss of gain. So, that cap is "special" and wants to be some good type, not an old oiled paper or electrolytic or something else crummy.

If you want to work with the other polarity pulse, just swap the transistor sexes and turn over the led - you still need a snazzy input transistor (low noise, fast) and the output transistor can still be "any good old switching transistor" if it's fast and won't mind some peak currents when driving coax (which will look like a very high quality capacitor to ground -- it can draw real current on a fast pulse).

I went ahead and used a quality high voltage coupling cap on the input, to test with detectors that will need that. A floating anode phototube could connect directly to the input base.
I didn't add a pull resistor to the output coupling cap because this should drive CMOS logic nicely -- with a pullup to it's positive rail for example. Had I put one in (pull up or down), then if you had some different use, you'd have to change it anyway. I'm now looking at adding one more transistor to this to give a TTL compatible pulldown output with gain and threshold -- more or less like a comparator would. Just as soon as I figure out how to deal with tempco, so it'll stay set right once you set it. At that point, we have "all in one" and still with long life for common batteries.

Here's how it looks built. I took some care here to keep the strays down, and the thing compact, given that it wants certain other things paid attention to.
Top view. Blue cap is the input coupling capacitor.

And here's the bottom. Note that this is effectively a single sided PCB layout, and that the major important nodes are tiny islands, with little parasitic C to other things - key to speed with high impedances. Input is at the top in this picture.
Bottom view, note simplicity achieved

Here's a little vid of me demo'ing the thing on the bench, using the bench lamp as the signal source. This should help convince people working with millivolts and high impedances of the need for really tight shielding if nothing else.

Sorry about the audio, I have no idea what was making that heartbeat in the background. This camera seems to have flubby-weird bass response, and it may have been the sound of two LF things beating. I can barely hear the generator, but I have no clue what the other thing might have been, as it seemed real quiet in the room at the time.

Re: Versatile preamp for detectors

PostPosted: Sun Sep 11, 2011 7:23 pm
by Doug Coulter
To add, here's the pulse response. Note the differentiation and resulting slow recovery due to the tiny input capacitor taking a charge during the pulse low time - it has to go somewhere, so you get a positive overshoot as it loses that charge again. Gain isn't to scale here as I added an attenutator in the input to the preamp that the scope probe (blue trace) doesn't see.
As you can see, though there's some time response due to the coupling cap, the rise and eventual fall times are very very fast. Using a larger input cap would make it look prettier for long pulses like this...something to consider. The bypass cap in the feedback loop could be usefully adjusted as well for this particular signal -- but the signal generator doesn't require a preamp anyway.
There will always be a bit of that when tuning for a particular source/signal.
Scope trace of the preamp input and output