adcanalogldooperational-amplifierpower
I have very tough requirements on noise (sub mV, no 50Hz allowed) and LDO seem to fail in that. The required current is like 20-40mA, so i am thinking about using opamp with reference voltage.
Is there any showstopper for that? Maybe, somebody did it before?
While I do understand the reason for limiting the bandwidth, I've seen a lot of schematics where there is no input filter.
The inherent noise from an op-amp, as a first approximation (e\$_{NOISE}\$), cannot be reduced by using an input filter. It can be reduced by an output filter or filtering in the feedback loop to reduce the noise bandwidth of the circuit but it cannot be removed by input filters.
Taking this to a more accurate approximation, some inherent op-amp input related noises can be reduced by input filtering such as i\$_{NOISE}\$. These for many circuits will be less than the e\$_{NOISE}\$ and are usually reduced by choosing smaller value resistors around the input stages of a circuit. This also reduces inherent thermal noise from resistors because the resistor values are much smaller.
It's sometimes good practise to limit the frequency response of an op-amp circuit to that needed for performance/spec reasons but there are many instances when this won't be the case such as when the op-amp is inherently limiting the frequency response i.e. you are pushing a design to its limits at high frequencies.
I presume that your 12V is also supplying the E-P transducer power (and perhaps other stuff) and that's why the regulator is getting hot. You should heat sink it so that the case temperature is reasonable if you want reliable operation, or use a switching regulator. The exact numbers depend on your ambient range and desired reliability, but "wicked hot" sounds like it exceeds the 60-70°C range that I'd like to see for general purpose applications.
The 'fluctuation' in op-amp output could be caused by oscillations. The LM324 is fairly robust with small capacitive loads (such as could be caused by cable to the transducer), especially for gain > 1, but you may still be having problems. I suggest this kind of output circuit:
simulate this circuit – Schematic created using CircuitLab
The input filter R4/C2 keeps high frequencies out of the input and matches the impedances looking out of the op-amp inputs so as to minimize the effect of the relatively large bias current on the LM324. The resistor R1 isolates the capacitive loading on the output from the internal output impedance of the op-amp. C1 forms a tight loop for AC to stabilize the op-amp. The feedback (R2/R3) is taken directly from the output so that R1 does not affect the output voltage when sourcing current (note that when sinking current the op-amp cannot get close to the negative rail when sinking more than 50uA or so, so the 0V output can't be approached even without R1). In most cases, the input of your transducer looks like a resistor to 0V so it's not an issue.
D1 and D2 conduct any transients to the power supply, saving U1 from damage due to ESD. R4/C2 and D1/D2 are optional as far as your immediate problem goes, R1 and C1 do the work. I strongly suggest you get that regulator temperature under control as well.
Best Answer
The main gotcha is that op-amp stability with capacitive load is often dubious at best- if you do this you can add extra compensation or use a special op-amp that is guaranteed stable with nF of capacitive load. Connecting an ordinary op-amp voltage follower to a reference will typically result in oscillation when the bypass capacitors are added to the output, and op-amps have pretty high output resistance at high frequencies (tens of ohms) so you usually need bypass capacitors.
We never have a problem with 50Hz noise here, but 60Hz (or 120Hz) can be a possibility.
Personally, I don't think you should do that, there are LDOs available with a few uV of noise (eg. LP5907). If you don't like the 90dB PSRR at 100Hz, add another regulator in front of it (doesn't have to be a great one) and it should be undetectable, if your layout is good. If uV of noise is too much, then add some passive filtering afterwards or, as GH says, a capacitive multiplier.