Hello, I suppose there are more than a few variables, but as general thoughts on:
1. XLR to input transformer vs
2. Input transformer to amp vs
3. Amp output to output transformer vs
4. Output transformer to XLR?
Like which node may be most susceptible to noise pickup?
A lot has already been said.
I would like to give more general analysis guidance.
The first part is consider the sources and mechanisms of interference and the sensitivity to interference.
Sensitivity is both basic and complex. Given the same amount of interference, the greater the signal level, the better the SNR. So on the most simple level, the lower signal levels, the greater the sensitivity to noise.
HOWEVER, noise interference (level) is NOT fixed, even if the interference source is fixed, the amount of coupling of the noise depends on mechanism and coupling efficiency.
One part is electromagnetic, magnetic fields are not shielded by the shield around conductors, instead they are best rejected by tightly twisted conductors. Coaxial cables also work well.
This extends for example to AC connections to and from mains transformers.
Magnetic fields fall off by the square of distance, so distance is the best mitigation strategy.
If magnetic interference must be shielded, usually mu metal foils, sheets or cases are needed. They are expensive and still limited. In one case my shielding solution for a low level amplifier was mild steel on the outside, then a layer of copper foil and then a layer of Mu-Metal.
Electromagnetic coupling is always b!tchy, often it can come from outside the actual unit once in a rack.
Doughnut shaped transformers (Toroid, C-Core and R-Core) are notorious for their strong vertical field when placed flat, the internal magnetic steel screw in toroid's acts as kind of a focusing device increasing field strength and making it more narrow.
Shorted turns around the outside of transformers (not inside the winding etc.) can help limit the escape of magnetic fields, by acting as a shorted turn "transformer" for the magnetic field outside the transformer. understanding of how transformers work and where magnetic fields escape containment is needed.
Magnetic field coupling gets worse with low impedances, because it's down to current flow. With infinite impedance, there is no current flow and magnetic coupling is defeated.
Of course magnetic items (such as nickel plated smd part terminations or magnetic steel wires on parts or indeed transformer cores) will act as especially sensitive receiver for magnetic fields. In this case high impedance does not help and twisting or other measures for cancellation is usually not possible.
As many passive components are magnetic, it is possible to pull out hair trying to trace the source of hum until you replace a generic part with an "audiophile" one which incidentally is also non-magnetic.
So first rule, no matter what the impedance, no matter if you have a plastic or metal case, twist all wires, consider transformer shapes, placement and magnetic fields (both in terms as source for other gear and sourced from other gear).
Second rule, place sensitive circuits in places where they are not only far from the device's power transformer, but also consider typical placement of power transformers in other gear in the same rack and don't place it where there will be a good chance of a power transformer below...
Distance, distance, distance. If required, adjust PCB layout to run hot/cold or signa/ground lines from connectors in side by side or top/bottom of PCB pairs (even with a ground plane in-between, it's transparent to magnetic fields)
Other interference is electrostatic. It is literally capacitive coupling. It is shielded by electrostatic screens. Because it is effectively "capacitor coupling", in presence of a zero impedance it is completely defeated, high source impedance and even higher termination impedance maximise this source of interference.
As the source for such coupling is usually 120V/60Hz or 230V/50Hz (nominal) the interference source is VERY HIGH VOLTAGE.
So if we have (say) 120mV signal with 100kOhm circuit loop impedance and we want 94dB SNR the coupling must be low enough to have no more 2.4uV coupled from (say) 230V. So the 230V noise needs attenuating by 160dB!
So 230V & 100kOhm @ 120mV we need to reduce coupling capacitance to 0.016pF (16 femtofarad) if we do not shield the circuit. Loose metal not grounded can act as antennae and the couple noise into the circuit.
The next thing we observe is we are back to basics, the higher signal levels, the higher SNR.
Low impedances defeat capacitive coupling but are exposed to magnetic coupling. High impedance's defeat magnetic coupling (into conductors) but maximise capacitive coupling.
Combining twisted wires and shielding will guard against both sources of interference. Starquad is even better than simple twisted pairs.
In PCB layout minimising loop areas, observing symmetric and parallel/layered running of hot and cold signal lines (remember, ground isn't) and doing so all the more where signal levels are low and shielding high impedance lines (buried on an inner layer inside a multi layer PCB) all can help making a design inherently resistant to interference.
If all else fails, self adhesive copper foil allows improvised shielding of traces, even in production, if magnetic interference in PCB traces bothers, cut them away and run twisted (shielded?) pairs of wire instead.
Thor