First, you need the grapes and the barefoot girls...nah, you just need good active devices that have nice characteristics, and a design that doesn't waste them. For truly fine design, these parts should be operating as they were designed to, and be readily available ones that have already stood the test of time - and are still used enough to keep them in the distribution channels. The only concessions to "new" made here are the use of a super good MLCC power bypass capacitor, and surface mount design to keep all the strays low. The rest is classic, and actually does work fine built up with through hole parts on perfboard, just not quite as perfectly as this. Here is the schematic we'll be discussing: (as usual, click the pic to get a big version)
Errata: The last transistor is shown as a 2n3906, which is a pnp, but drawn wrong here - look down-thread for a more accurate schematic drawn by Joe Jarski.
What the heck - here it is inline too. Oops, drew that emitter the wrong way, and yup, it matters.
/////////////////////////////////
I've come up with a preamp design for nuclear detectors that I'm pretty happy with, even proud, so I thought I'd share it here. For those less up on the arcane details of analog design and interactions, the explanation of how it all works separately and together should be helpful. I have used variations on this topology for a lot of things over the decades, this is specialized for the kind of signals we encounter in our fusion work, but is pretty good general purpose too – with different optimizations. All of those are done by simply selecting different parts and values, so a single PCB layout handles pretty much all uses of this.
The goals here are fairly simple to list:
1. Wide supply voltage range – this shouldn't be a picky eater, and should work over a battery lifetime if desired.
2.Millivolts in make volts out, a good loud signal that won't be bothered by noise around a high voltage device. If possible, enough output swing to directly drive TTL or CMOS with no comparator and variable reference required - a schmitt trigger input should handle our output fine and simplify your system overall.
3.Gain easy to adjust to make the above work well.
4.We want our response to be matched to the input signal, which in this case is a low level negative going pulse with fast risetime and slow decay. A matched filter will get us the best signal to noise possible with our known signal characteristics.
5. Low noise.
6. Tiny, self shielding, convenient.
I want to do a stage by stage analysis first, then show what the feedback does to alter that as a separate issue.
First is the input protection and DC bias. I show here two 1n4148 diodes set up to clamp any input signal to the power rails, plus or minus 1 Vf for a diode. This is to protect the input transistor with a cheaper and easier to replace part. While I personally use phototubes with a negative supply and floating anode that can go “straight in” to the input base, many like the positive 2 wire supply and coupling cap or transformer to a preamp, so I made provisions for either. The issue when you have a coupling capacitor is that it might hold substantial charge versus what a base-emitter junction can handle, and hot plugging such things does happen by accident. Even a tiny capacitor can deliver amperes to a sudden short, and the fairly fat 1n4148's can take that better than the low noise, low power preamp input transistor. They do add a little capacity, even reverse biased, but not much. In a hard wired and carefully built detector design, you could just leave them out if desired.
Next, I wanted this preamp to be mostly DC coupled, and only add various time constants where I wanted them. You don't want too many capacitors making for undershoot and long recovery times from overloads if you can avoid that. This requires a bias source that is relatively independent of the power supply voltage – one goal is to have this work over a fairly wide range. We also require that this voltage have the same tempco as a transistor Vbe, so the output level won't drift around too much with temperature. It turns out that a LED has a one Vbe drop, plus a drop related to the quantum energy of the light frequency it would emit, along with some parastic loss due to series resistance – nothing's perfect. But that nice fixed voltage on top of a Vbe is just what the doctor ordered here to just barely turn on the input transistor. We can fine tune the bias voltage by led selection for coarse adjustments, and series resistor to control the LED current. In this case, we just want to barely bias all the transistors on, so the full supply voltage is available for the output pulse swing. The LED provides just the right temperature-variable voltage to accomplish this – we may or may not put enough current into it to make it visibly light up – that's not the point here.
Now for the first gain stage. This is a low noise at high impedance bipolar transistor. I chose a PNP for this slot because transistors turn on faster than they turn off for a variety of reasons, and that's the match for our signal. For fast but not super fast speeds, the audio standby 2n5087 was chosen as at low collector current it has almost theoretic noise figure at the medium impedance level we have here for phototubes and gas proportional tubes. For higher impedances, a FET would be slightly better, but much harder to bias stably – it would have to be tuned for each sample of the FET used. Looking at the gain from base to collector, due to the resistors shown we have a DC gain of a little less than 1 (inverting), but an AC gain of roughly 47, due to the ratio of 4.7k to 100 ohms. The bypassed emitter resistor is one way to futher fine tune biasing in this design, and we have a little high pass from the reactance of the bypass capacitor here in parallel with the 4.7k. I chose the capacitor to be large in comparison with the expected signal pulse width, but other choices might be favored in other applications. Note that the Hfe of the PNP is generally in the range of 300 or thereabouts at the collector current we want to run. We want to produce just about 0.7 volts across the output 4.7k resistor, which works out to 148 uA. This is about where we want it for least noise with some tens of K input impedance. The 100 ohm AC emitter impedance will be boosted up by the Hfe to about 30k – which is less than the 100k bias resistor and input load we show – but we've not covered yet what feedback does for us.
Moving to the second stage, I flipped transistor sex to an NPN, again, because as the first stage turns on, so will this – on that fast risetime or onset of our signal. This stage is set to have a gain of about 21, due to the resistor ratio. Without feedback, this gives us a total gain of 21 x 47 or 987 to work with.
This stage is all DC coupled as you can see. It will bias with a gain of 21 off whatever change occurs to the voltage out of the first stage, which is close to the practical limit without feedback for stability. More gain than that becomes really hard to get stable in DC with various non-theoretical parasitics in the parts (series resistances and such) and even small power supply variations. We want to burn more collector current in this stage for speed, but not so much we get into power issues, so we do – here we burn perhaps a milliamp or a little more. We also want the base of this transistor to not have to wiggle much – here we move 1/21th of the output signal. The reason for this is that the first stage transistor will have some base-collector capacity. The more we let it's output move in voltage, the more the input signal has to be diverted to charging this capacitor, reducing gain and speed. If you think of this as a charge sensitive preamp, that's the main capacity being worked on by input charge, and we don't want it to be multiplied by the miller effect.
The final stage is just a simple emitter follower to reduce the output impedance further and directly drive coax if desired. We therefore use a relatively smaller load resistance, and provide a series match to coax. The other purpose of that series resistor is to prevent creating a one port oscillator with the follower – an esoteric but real troublemaker in some situations, as a piece of coax can look like a tuned circuit under some circumstances – in this case, the series R limits the Q. AC coupling is provided as well as places for pull up or down resistors on the output. When it's all tuned up, the emitter of this resistor wants to be sitting just below the postive supply rail. This is to make the full supply voltage available for output swing, and to reduce quiescent current draw.
OK, now lets look at what the feedback (that 15k resistor) does for us. We had a little too low input impedance. With the feedback (assuming it was coming from infinite gain, but our 987 is “close”) we now move the input emitter to track its own base input. This eliminates the effect of any emitter-base capacity. It also bootstraps the input impedance by roughly the ratio of open loop to closed loop gain, in this case that would be our 987/150 or 6.58 to one. That kind of handles taking our original 30k in parallel with 100k to a point of letting the 100k be the main determining factor of input impedance, sweet! It's now in parallel with a 197k effective transistor input impedance – so we wind up with about 66k input impedance net. Not so bad.
We also get some DC feedback, and in this case the model is a little different, as the input stage 4.7k emitter resistor is also in play. So we're only trying to close the loop on the DC gain of the second stage here. This cuts that DC gain down from about 21 to about 3 – the ratio of the 15k and the 4.7k, so the thing becomes more stable to small DC variations caused by temperature and parts variations.
The resulting circuit satisfies all the goals, hands down. One can tune all this by using different parts or values. For example, to get more speed, at a cost of lower input impedance, one could reduce the first stage resistor values and burn more current there (which helps discharge capacities quicker and turn off the transistor faster). You could, at the limit, start looking for a faster PNP to put there too, within the noise and current gain limits you decide to tolerate – low noise high gain PNP RF transistors don't exactly hang out on street corners! Ditto the other stages – you could reduce the resistor values, keeping ratios about the same, throughout. This will cost you some noise, but mainly a lot more supply current – instant death to batteries! But you might not care about that if you have the need for speed.
To reduce the drain, the bias is probably the place to fiddle. You could in theory go all the way to the point the output transistor is just plain off – or even the second gain stage. This would create an implied (but real) threshold on the input signal, but if you don't go too far that way, you'd get free “baseline clipping” from that as well as a reduction in power use. In general you'd not want to go much below about half the collector current on the first stage shown here – the characteristics of this transistor don't scale forever. You would get a little higher input impedance for least noise figure doing that, but only to a point. In truth, noise from other sources starts to be the big problem here, with this high impedance and high gain, which is why the PCB has one side all ground – point that side at any source of noise, and the top right at the signal producing device!
To change the gain, just change that 15k resistor. If you change it a lot, you might want to redo the bias as well via either changing the LED current, or the 4.7k emitter resistor in the input PNP.
As shown, the values are more or less optimized for a photo tube with floating anode, direct into the input base, to produce TTL level outputs and with speed to match fairly slow scintillators – which would include the ZnS:Ag in a hornyak detector, NaI:Tl, BGO's longer time constant (it has two), or a gas proportional tube, for which this is faster than needed as is.
I'll post up some scope traces of putting the thing through its paces when the boards get here and I get to build one up all surface mount. This board design takes care of not having excess stray capacity where it will hurt, and in just plain having very little parasitic C or L anywhere, so it should squeeze the best possible performance out of the parts used. The bottom all copper, and a ring around the top should help with noise pickup issues as well. I included a minor RC network to take the real bad fuzz off the power input, but not enough that you should run 10 foot long unshielded power wires by a high voltage arc on the way in – do be careful there.
/////////////////////////////////
Here's what the board looks like as laid out by Joe Jarski. And yes, we'll be making these available either built or as kits at DJ's real soon now. Although there will of course be a labor charge for building and testing these - unless you can do SMD pretty well, you should probably spend the money and get the built version. These won't be real easy to make, but worth it to get the size and performance.