What ground loops and noise? Why are you designing your circuit so it picks up these things?
I think the disconnect is, those who teach "star grounding", stop just short of a general truth. It works for DC, sure, but real circuits are anything but DC.
Actually for that matter, I expect a lot of examples of "star grounding" aren't. As in, they don't actually accomplish the thing they set out to do in the first place, and just make things altogether harder, while inviting problems like oscillation and susceptibility.
The generalization, that works up to RF, is to use a star topology
of transmission lines (TLs). Common mode voltage ("ground loop") is allowed between regions of the circuit, because there are no connections crossing between those regions -- except along transmission lines, which are routed through common nodes. At RF, common mode arises as induction along a ground path; since the TLs are well coupled to ground, minimal mode conversion occurs and immunity is high.
To be more specific, define a node as any subcircuit, group or local region, where the voltage drop or phase shift across that region can reasonably be considered as zero. That is, it's point-like, zero-dimensional; a local star point.
Define connections to nodes as ports. Ports are also locations in the circuit which can be treated point-like. Kirchoff holds in the usual way, and all that. Note that connections to transmission lines obey a modified Kirchoff's, where the signal current flows in one terminal of the port, and immediately out the other (image current). Whereas, along a given wire in a transmission line, the current is delayed by propagation, so Kirchoff does not hold in general.
The topology preferably shall be loop-free. If signals need to be looped back between nodes, they should be routed along existing connections (following the unique path between those nodes). The topology really doesn't matter much as long as shielding is good, but also common mode chokes or isolation transformers can be used to improve response.
Overall, the circuit can be understood as a network of transmission lines between nodes, and the crosstalk / shielding of those connections is what determines the performance of the circuit, and what if any remediation shall be used to deal with it (e.g. keeping distance between transmission lines, adding CMCs, etc.).
At AF, we don't have the luxury of well coupled grounds: copper foil rolls off (i.e., is around a skin depth) in the 10-50kHz range, below which it's more or less just another resistive conductor.
Note this is no excuse: we must still follow RF practices, as we might be using devices that operate into the 100s of MHz. If we follow the RF methodology exactly, we run into the problem that our transmission line grounds carry some DC supply current from the circuits they're routing between; and this manifests as an error added to our signal at one end or the other -- mode conversion, classical ground loop. What to do?
It's convenient, then, that we built our circuit with transmission lines. We can separate their grounds from circuit ground, lifting one end at DC, so that we can sense the DC/LF error, and subtract it from the signal. At high frequencies, the shielding remains good (or, it can be, at least) and we don't need any special precautions; therefore we can bypass the DC-lifted end to GND. We get what looks like an ordinary transmission line at AC, but a differential pair at DC.
We could go one step further and make the situation symmetrical: route a full differential trace pair between locations, and use a differential receiver to pick up the signal. This can improve CMRR (particularly around the crossover frequency), but has the downside of requiring an amp with enough CMRR bandwidth to cancel out the noise.
Applied to traditional circuits, full differential design isn't the simplest. If you've got op-amps handy, not really a problem, but for single discrete devices (tubes, transistors), you have the restriction that power ground is also signal ground. This can be addressed to some extent by separating them with bypass caps (e.g. bypassing the cathode to signal GND, leaving the resistor going to PGND alone). You're still stuck by problems like PSRR, where a fraction (triode) or whole (pentode) of the B+/VCC supply's ripple is added onto the output signal. This can be solved in precisely the same way: prefer topologies that are differential (so the supply noise can be canceled out), or ground-referenced (e.g. CCS or gyrator load, with shunt feedback to set gain and reference output to common).
Note that amplifiers don't need anything fancy, as far as ground topology. The main load is the output, so connect the rectifier (and filter) to that. Route power and ground back to the driving stage, and so on, in a linear route, to the input stage. No ground loop to be found, no hum, no crosstalk, no susceptibility. The only interesting part comes when considering stereo, from the same power supply: now a U topology is had, with the power supply being common. Ah, but then the inputs must be split, and in general, they'll just be coming from a common ground source, won't they? Indeed, so we must consider some differential treatment at the input. The input grounds don't need to be hard-grounded, they only need to be referenced to their respective input amplifiers. I can think of some ways to apply a common-cathode stage for this purpose, or there's always the long-tailed pair.
Tim