"...would it be possible...?" Absolutely!
Is it a good idea? Maybe.
You're off to a good start by powering the brains and brawns separately. But you can still get some induced noise from one to the other, partly because the twists aren't perfect and they're not shielded from each other.
@Pete's comment is also a good idea. Still keep the power supplies separate, but send:
- Constant 24V for the motors. (still has lots of current ripple, but not much voltage ripple like an already-controlled line would)
- Maybe 9V for the MCU's. A local regulator on each board drops that 9V to 5V, right where it's needed.
I wouldn't go much below 9V. A lot of linear regulators require 2V of headroom just to work at all (7V minimum for 5V out; 6V in would produce ~4V out as the regulator saturates), and the extra 2V above that allows for some noise-induced "wiggle".
Local capacitors on both sides of the regulator are also helpful. They take over at high frequencies while the regulator is really meant for DC and "DC-like" low-frequencies.
If an existing standard has some value to you, then you might have a look at PoE. (Power over Ethernet) There are two variations, both of which use a single 48VDC supply, and (usually) a switching power supply (DC-DC converter) at the receiving end as a matter of course. The output of that switcher could be anything you want, including multiple outputs at different voltages.
The two variations are simply which wires that 48V supply is presented on:
- One uses the data lines for 10/100 Ethernet, so that there are still 2 unused pairs in that cable. This is done with center-tapped transformers on both ends (might be buried in the jacks themselves): one pair is 0V (ground) and the other pair is +48V.
This is often used by switches and other networking gear that uses the data lines anyway.
- The other uses those unused pairs. As before, one pair has both wires at 0V (ground) and the other pair has both at +48V.
This is often used by standalone PoE injectors so that they don't have to touch the data lines at all for 10/100. (Gigabit uses all of them, so this reason doesn't work anymore for that.)
A PoE device needs to accept both variations in either polarity (4 possibilities) to be fully standards-compliant, in addition to the signalling that a standards-compliant source requires to enable the power on that port. (so you don't blow up your laptop by plugging it into a PoE switch)
The injector version (putting power on the unused pairs) would work with anything that only requires 2 pairs otherwise. RS485 each direction, for example.
The switch version requires a DC average of 0V on the communication lines, which Ethernet satisfies but RS485 doesn't. You might make it work anyway, with a bit of stuffing at the protocol level, or it might "just work" with sparse communication and a receiver that interprets 0V differential as idle.