Electronic – How to isolate processor noise

The solution was to physically separate the power connections for the audio components. While @Dave mentioned in the comments something about a star ground topology, the problem was that while the ground connections were all centralized, the Vcc connections were also, and shouldn't have been.

In the original design, power was supplied to the entire assembly via a small PCB with header pins for each separate component. Two rows of male header pins, one for 12V DC and one for ground. In this arrangement, all of the noise was being dumped onto the audio components each time the servo motor was actuated. Adding capacitors helped slightly, but certainly not enough.

The main two wires supplying the small power distribution PCB are about 10 cm long. I added a second pair, and created a second power distribution PCB. The original one was used for the servos and receiver, while the new one was used for audio and amplifier. The "Y" split in the power cable ultimately had the effect of keeping the power clean on the audio side.

I also removed the ground connection between the RC receiver and Arduino, though this was unnecessary. Its presence did not add noise, nor did its absence cause triggering problems.

An electrical engineer coworker advised me that it was important to have separate power supplies for motors versus digital components. While this project still only uses one power supply, the solution was to split the power as early as practical, and dedicate each to its respective purpose.

The catch with low cost audio chips like this (or even the more expensive LM1875) is that guaranteed PSR is rather low, around 50dB, while the typical one can be much better (90dB promised in the case of LM1875, see below).

The TDA2030 is actually even less promising with only 40dB guaranteed and 50 typical:

And TDA2003 is even worse with 30/36 dB min/typical:

That's definitely going to be audible if the power source is noisy. (Also beware that these numbers aren't quite directly comparable because ST tests at 100Hz noise and TI at 1kHz...) So... if a linear regulator supply is out of the question (which will totally kill any audible noise from that source in my experience [it gives you some guaranteed 90dB with a LM317/LM337 solution]), try a different TDA2030 chip, assuming you're not having fakes or the CXY knockoffs (or try a LM1875 as it's pretty much a drop-in replacement). Luck of the draw with these things.

Your unspecified "cell phone charger" is almost certainly an SMPS with atrocious noise characteristics (better ones exist, but usually not at this price point). See for example what cheap ones have inside: https://www.youtube.com/watch?v=wi-b9k-0KfE

Also, a linear regulator after an SMPS won't be anywhere near as effective as one after a "classic" (filtered) rectifier. The reason is that the typical linear regulator is very good at eliminating mains hum (actually 2x the mains frequency in a bridge circuit), but its performance gets a lot worse starting around 10kHz... and that's exactly where the SMPS noise usually starts. And the audio amp IC has similar curve, e.g. here's the one from the LM1875:

Summary: the guaranteed fix is better (preferably linear regulated) power supply, which you can build yourself. If that's not an option somehow... try the amp chip lottery. However no amp chip will will magically eliminate the horrible noise put out by a super-low cost SMPS.