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?
Do we use a similar filter after the photodiode to get rid of the out-of-signal-space noise, before sampling?
Generally, we don't have a separate filter device or circuit. We'd rather connect the photodiode directly to a trans-impedance amplifier (TIA) chip to avoid losses due to impedance matching (the photodiode produces a current signal so we'd rather it drive a low-impedance input to the chip than a middle-impedance filter circuit).
But the other circuit elements are themselves bandwidth limited and this typically helps to reject out-of-band noise.
The photodiode has an associated parasitic capacitance that limits its bandwidth.
The TIA has limited bandwidth, which rejects higher-frequency noise.
In wireless communication, we use a matched filter to suppress out-of-signal-space noise.
No, we do that to maximize SNR under white additive noise. That's not out-signal-space!
After this filtering we do sampling.
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.
Otherwise, the power of noise is infinite (practically very large).
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).
In short-haul fiber-optic communication systems photo-diodes are used for signal detection.
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.
Do we use a similar filter after the photodiode to get rid of the out-of-signal-space noise, before sampling?
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.
