communicationdigital-communicationsoptical-fibrephotodiodesampling
In wireless communication, we use a matched filter to suppress out-of-signal-space noise. After this filtering we do sampling. Otherwise, the power of noise is infinite (practically very large).
In short-haul fiber-optic communication systems photo-diodes are used for signal detection. Do we use a similar filter after the photodiode to get rid of the out-of-signal-space noise, before sampling?
There are, no doubt, many reasons why the dark current increased, but to answer the question about whether the diode is broken - badly or otherwise - it's just necessary to measure the current with the diode illuminated. If you're using the diode in reverse-biased mode, measuring the reverse current with various levels of illumination will give you some idea of how its response conforms to the data sheet specifications.
If you're using it in photovoltaic mode, then subjecting it to various levels of illumination and load and plotting its response as a load line will also yield salient data.
In the real world, though, if you've exceeded the manufacturer's maximum temperature spec by what... six or seven times? I'd say you've definitely let the magic smoke out.
I cannot speak to the digital question, but the analog is pretty straightforward. You should use a TIA with a 1000 lux sensitivity, about 8 kohm / full scale volt. Use about a 5 volt bias, and expect a photodiode capacitance of 4 or 5 pF. I'd suggest that you don't need a PGA, rather, use a pair of x10 amplifiers in series and an analog mux to select the active channel. Be aware that selecting the appropriate channel will not be trivial in some cases, especially where there is a lot of AC on the signal. All amplifiers need to be selected with fast overload recovery in mind, or incorporate clamps. Fortunately, your currents are so low that a decent clamp should be straightforward. Log amps are in principle a good idea except that it's hard to make a good log amp that is also fast. The other problem with log amps is that it's very difficult to produce the sort of precision you want over large dynamic range.
ETA In response to a request, here's a TIA using an AD8651 and +/- 3.3 volts -
simulate this circuit – Schematic created using CircuitLab
The AD8651 is available for less than $4 in onesies.
One thing to be careful about is the value of the feedback compensation capacitor. With the values shown, and an assumed total input capacitance of 6 pF (4 pF for the diode, 2 pF for the IC) my simulation gives settling time for a full-scale step (3 volts for 130 uA) of less than 300 nsec. However, this is quite sensitive to the value of the capacitor, and so is sensitive to layout considerations. A very compact layout is a must - which means that breadboards are absolutely verboten. Good decoupling of the +3.3 supply at the op amp is also a must - this is a 50 MHz op amp.
Best Answer
No, we do that to maximize SNR under white additive noise. That's not out-signal-space!
Ah, yes, that's what's said in many textbooks. It's not really what we do, at least for the last 25 years: it's very hard to actually implement sensible matched filters in analog hardware, so we typically anti-alias filter, then sample, then apply frequency correction, then matched filtering and then timing recovery. The timing recovery is what many textbooks confuse with "sampling". That's because what timing recovery actually does is allow the receiver to look at the reconstructed signal at the symbol instant. But, really, I've yet to actually see an RF system that does the matched filtering in the analog domain.
Nah, that's something else: when we assume actually white noise, its power would be infinite. That just goes to show that the assumption of real white noise has to be wrong, but we use it anyways, because we can often assume noise to be white within the bandwidth we're sampling. And thus, it's usually not even "very large" compared to the signal power (but that depends on the design of your system).
Yes! That's Intensity Modulation, Direct Detection (IM-DD), i.e. the diode just actually squares the signal and the system observes the baseband that results.
No. There's no need for that, because noise energy typically is low here; that filter in analog would also be very hard to realize, like in wireless.
Correction: it'd actually be impossible to realize: the transmitting and receiving diodes are a strongly non-linear channel (squaring!), so the assumption that a matched filter maximizes SNR breaks down here.