Electronic – USB 5v power filtering

The current transients mentioned, depending on magnitude and how well the amplifier circuit is decoupled, may cause dips in V_supply as described.

Some points to address for a solution:

• Most important: Use a freewheeling diode, connected reverse biased, across the speaker leads. The on-to-off transition of the speaker causes a significant reverse spike from an inductive load such as a speaker.
  • Use a power Shottky diode if the current involved is big, else any fast diode with a current rating of at least twice the current expected through the speaker coils in on state would work.
• Use a lot of bulk capacitance at the source of the transients, i.e. across the amplifier circuit, to reduce supply noise propagation: One always tries to eliminate noise at source if possible.
  • Depending on the transient magnitudes, a few hundred to a few thousand microfarads are common in power audio amplifiers. Voltage rating of better than twice the supply voltage would be best.
• Separate out the power and ground planes for the amplifier portion and the microcontroller portion if feasible, interconnecting grounds at a single point.
• Coupling the power plane using a choke coil as mentioned in the question is a good additional measure, but ensure this is a very low resistance high inductance coil, preferably with a core. Tricks with PCB traces will not provide the high inductance needed, an actual inductor would be best.
  • Values from a few hundred microHenries to a milliHenry or so should be good to effectively throttle spike propagation. Use an inductor with the least DC resistance that you can afford. Personal preference is something well under 250 milliOhms if possible.

I get basically the exact same issue with my car stereo if I charge my phone/MP3 player while having it connected to the line-in of the stereo. The ground loop is allowing common-mode noise in (in this case, the ignition system of the car is producing huge amounts of noise, and it's getting in through common-mode coupling).

Since this is common-mode, galvanic isolation will not necessarily fix the issue, since the critical measurement is common-mode coupling capacitance, which can be fairly high even on brand-name hardware (basically: efficiency, cost, and common mode isolation. Choose any 2).

What you need is a "hum eliminator", e.g. an audio isolation transformer.

They seem to run about $10 in general:

It may seem counterintuitive that the solution to your issue of noise when you connect the USB line is to put a filter in the audio line, but that is indeed the proper solution.

The noise is a function of the loop created by the two cables. Basically, both the USB and the audio cable are grounded together in the gopro. As such, if there is a voltage differential across between the audio-connection in the plane, and the power connection, it will cause current to flow through the audio cable, through the gopro, and out the usb cable (or vice versa).

As such, breaking the connection in either one will solve the problem. Audio isolation transformers are much cheaper and easily available, which is why I suggest it over a isolated power supply.