Hello! Thanks for sharing solution! How it works after time?
Plus of LED strip should be connected after filter?
The filter is between the power supply and Shelly.
RGB(W) strip then goes normally to Shelly.
I can confirm after 2 years of frequent operation that the solution is permanent and works.
It could sure use a nicer form factor, the box is almost as big as a power supply 
But it works and it is hidden behind a bed so I kinda forgot about it.
Okay, new update here.
I managed to solve the issue. The problem is now completely gone and the PSU is dead silent. Since I went a little overboard with the specs of the caps, this takes a solid half a second to power down when unplugged since the caps are so huge
I contacted my friend and he charted up a simple low pass filter with a inductor (coil) and couple of caps. He told me that to be extra sure it will work I can add some big-ass caps as well, so I went a little overboard with the design. The huge C1 caps are probably not needed The inductor however must be rated for whatever your power supply can deliver, in this case I am underrating it a little because it is a 4A PSU on 3A coil - I simply couldn't get a bigger one. But in real-life scenario, the current will never be over 3A, just for a brief period of time before the caps charge when powering it on.
Anyway, here is the schematic:
[Blockierte Grafik: https://andy.nech.me/shelly/scheme.jpg]
And here is how the final product looks like in the random box I had at home. Excuse the crude implementation, but it works and it's going to be hidden under a bed so I don't mind.
[Blockierte Grafik: https://andy.nech.me/shelly/lpf.jpg]
I also checked the thermals with thermal imaging camera after 2 hours of operation, all cold. The hottest is the RGBW2, for which 38°C is normal temp.
[Blockierte Grafik: https://andy.nech.me/shelly/thermal.png]
Hi. I have the same problem, tried two PSUs so far that didn't have any problem with my previous dimmer. I am not a (professional) electrician either, but let's say I have some overlap with my hobbies.
The noise is indeed resonance of the dimming frequency, which is 1Khz square wave. You can see this on an oscilloscope connected to any of the R, G or B lane. Furthermore, they are not exactly square but PWM, square would be PWM with exactly 50% duty cycle. As duty cycle changes, you introduce different harmonics to the fundamental (integer multiples of the base 1khz) which mix together. The PWM is also not clean on any single channel by itself, as observed with the scope on just one color. You can see this dirty PWM on this picture:
[Blockierte Grafik: https://andy.nech.me/shelly/dirty_pwm.jpg]
This wouldn't be a problem by itself as the LED strips or the RGBW2 itself isn't making the noise - it's the PSU.
The reason why this happens is that the power intake is not smoothed, the load on the actual PSU changes rapidly with some sum of the individual waveforms of each color channel.
See this next picture, this is the scope of the DC power supply 12V line (zoomed in the scope to make the point). This should be a flat line, but it is oscillating by over a volt!
[Blockierte Grafik: https://andy.nech.me/shelly/dirty_dc.jpg]
Fortunately, this looks like a job for a smoothing capacitor, that, when added over the + and - lines, should make things fine by absorbing these irregularities. I don't have the knowledge to calculate such capacitor parameters, but maybe there are websites for that. I am sure I will have to take into account some variables that I need to figure out, while other I know (power intake of the LEDs, frequency and voltage). We are looking for a low-pass cap here - a cap that will not let any frequencies above a certain treshold through. We can have this treshold very low, something like 100Hz so that by 1kHz it is surely gone, because lowpass never works "instantly" on its desired frequency - there is a roll-off.
Additionally, 1khz frequency is very unfortunate sellection, this will cause a lot of interference with rolling shutter cameras. You can hear me talk about this topic on this video. Of course higher frequency switching requires more precise components that are more expensive, so I understand why they went for 1khz - it is standard in the field.
