Photodiode case

I have always grounded the can directly to chassis resp. the ground plane on the circuit boards. Reverse bias it, to gain a wee bit more BW and avoid rectification of RF noise. I usually select a (very clean!) bias of about 80% of spec sheet limit. If BW is still an issue you can cascode to reduce the capacitive load by the TIA. Phil Hobbs explains this very nicely:

You won't have h-field problems, so any metal box is as good as any other.

I usually ground the case, in my case to the pc board ground plane. Seems fine.

It should.

This was apparently a fairly bad idea, as the

If the pc board that contains the tia is securely grounded to the case, and the pd can is grounded, and the leads from the pd to the tia are short, it should work.

What's the pd capacitance? I'm surprised you need to bootstrap. The bootstrap drive may be causing some of the weirdness you're seeing.

Ground everything!

It would help if you could post a schematic and some pictures.

John

Reverse bias helps a lot. It not only reduces capacitance, mostly in the first few volts, but higher voltages sweep charges out of the junction faster. So use all the volts you can.

RC filter it close to the pd.

His book is worth having if you're doing electro-optical stuff. It's fun to read anyhow.

John

Yes. But I don't dare to come closer than 80% to the abs max. Which can be as low as 5V on modern ones so I usually go for 4V.

Absolutamente. Costs next to nothing.

I've got to buy one. Not having read Phil's book before my last design almost felt like driving without a license.

Wow. So many other people withh nothing better to do on a Saturday ;-) Thanks for the comments, I will head into the lab later this weekend.

I had tried just the cascode and reverse biasing, but I only had available -5V reverse bias. That did not get me the BW I needed. The Cd on the part is about 2.5pF at that bias. I could not resolve this, as I should have had an r_e on the cascode of about

Yep, I would do this, plus use a ferrite. Or maybe a capacitance multiplier.

I have this, it is quite a good book. The circuit is actually the bootstrap plus cascode he covers in the book and the paper, so there's the schematics (different op-amp in the transimpedance stage). I've also corresponded with him several times for clarification (not that he was unclear, I was just being dense). Quite a nice fellow.

Yes, perhaps.

There's sometimes an issue arising in high-gain preamps, with the output stage's coax-driving current return path, and whether any of that couples back to the input signals. For example, I have made wideband high-gain amplifiers (e.g. for channeltrons, etc., with 50MHz bandwidths and G = 5000), and I've found that sometimes a single ground stud from the PCB to the box was a problem, but ringing (or oscillations) stopped once I had two ground studs, one near the input and one near the output.

With two ground connections to the box it's also helpful to split the ground planes, mid-circuit (couple with a differential stage) and use the box for the required dc-path connection. This scheme prevents any of the high coax drive currents from flowing near the input sensor and amplifier.

You can also add an electrostatic shield soldered to the input ground plane, made from thin copper sheet stock, etc.

Another helpful trick you can use when only one ground plane, etc., is available and such modifications would be painful, is to use a floating BNC for the output signal, carefully wiring the ground return right back to the output stage's bypass cap and its ground reference point. There may be a distant ground return, via the AC power ground to a vacuum chamber, or whatever, but the distance may eliminate any feedback problems. In some cases the high-frequency isolation can be assured with a ferrite clamp around the coax cable.

What is the P/N of your diode, and what opamp are you using? My last design only needed 50MHz or so but I got well in excess of 100MHz, sans cascode and with -4V reverse bias.

Pay close attention to the capacitances around your TIA. A stray pF here and there can really ruin things.

Use all mounting holes for grounding to chassis. Plate them correctly, i.e. nickel (do _not_ tin-plate). For long term performance make sure mounting HW and studs are plated compatibly. For proto series unit I usually order that stuff from McMaster.

Split or floating grounds are IMHO a recipe for disaster but it's still taught at universities. OTOH that does provide a nice income for me because I get to reverse all that. The last one was a week ago, but on the way home I got stuck in traffic, big time :-(

Why not? If it's designed and spec'd for X volts, why not run it at X volts? Confession: sometimes we use schottky diodes and phemts at 2X.

That will still not be completely "depleted", namely the capacitance is still on the curvy part of the curve.

John

This gets 180 MHz from a 2.5 pF silicon pin diode at, I seem to recall, -9 volts or some such. I just dumped it into a medium-fast opamp as the tia, no cascode or bootstrapping.

Bootstrapping is likely to cause more troubles than it's worth.

Overkill. Just a cap would be a better hf ground than any active c-multiplier.

John

A photodiode is the ideal floating signal source, so doesn't have these problems.

My little e/o boxes have 4 pems in the bottom of the box, with the pcb ground plane solidly screwed to all. We're doing a custom one now where the board is screwed to several machined metal bars, with lots of screws, and the bar are in turned screwed to the box. These are serious electrical and thermal barriers.

The more you ground things, the happier you'll be.

NEVER split ground planes. NEVER use multiple ground planes.

We seem to be philosophically incompatible.

John

The split ground is _inside_ the small box and is a totally appropriate way to control where the cable-driving currents go, thank you. Without it, even your nickel-plated contact may one day become insufficient, with it, the longterm contact quality is _much_ less important, if at all. You don't think extremely careful control of the RF current pathways is important in high-gain wideband amplifiers? I seriously doubt that.

As for whether it's taught at universities, I wouldn't know - that's across town from me. I'm an engineer, not a professor.

My suggestion for a floating output BNC with ferrite clamp was stated to be for a poor bloke who has to live with what's in front of him, and needs a way to get it working. In any case it's a more reliable technique than you might imagine. Keep in mind it would be employed by the user of the boxes in question.

Well, the client has to pay for another 50 or so photodiodes per system if the PS hicks up ;-)

I've looked at that with a Hamamatsu and a JDS diode and didn't find too much change. IOW we didn't need the last few percent in BW. Initially I thought about making that programmable but then thought about the consequences if someone accidentally steps onto that register.

I've seen it go wrong almost every single time. In theory a split works, of course. In reality stuff has to be connected to the individual sides and suddenly you've got loop antennas. Even in case of opto electronics because the stuff on the PD side of the split must be supplied with juice. Now you could do that with fully isolated forward converters and stuff but with more than one voltage that gets old. Then again I may be a bit biased here because many of my designs must work in rather nasty environments. Defibrillators, diathermia gear, elevators and so on.

IMHO the real education happens at institutes. That's where I learned some of the ropes (RF Institute at the RWTH Aachen in Germany). Seems not to happen much anymore, most of the young grads I interviewed could not even solder (!).

In a pinch a ferrite can "make it work". I do that, too, but only when there is no other option. It's like pain pills, makes the symptoms go away ;-)

The last system with isolated coax connection and ferrites was ultrasound and before throwing out the split grounds the guys were wondering why the image background "lights up" if someone touches the transducer jacket.

But the OP should consult the datasheet first. Some newer PDs have -5V abs max.

Shhht! Not so loud. This is 30% of my revenue source. Just imagine a body shop when nobody crashes their cars anymore.

So, on to politics then?

Not to beat a dead horse, but to be very clear, the split ground plane is for the purpose of eliminating any loops, or more importantly, of preventing two possible paths for the output return current. Remember, both planes are grounded to the insides of the small box (the input is an isolated PD sensor). And I will strongly state, you will NOT ever find it "going wrong." Quite the contrary. Ahem, may I assume we're talking about the same thing here?

OK, I'll take that as a compliment for the Rowland Institute, if I may. (I'll take whatever I can get.)

snip

A new world wide problem has occured, and for a change, it is not the USA causing it

Martin

Maybe we aren't. If you bolt it all to chassis inside the box you don't really have a split ground anymore. Might as well make it one plane then :-)

Yes, institutes can play a major role in teaching young students about the real stuff. They might cause grief and even destroy this, that and the other thing but the value to society you guys are adding by taking them in is huge. AoE is the other huge contribution. I guess I may be responsible for sales of a few dozen copies, telling young grads that no matter what their professors say "you've got to have this book".

Academia alone seems unable to do that hands-on anymore. One client tried out three grads and finally the forth turned out to be a good fit. Imported from Canada...