Input buffer and belamini powered from external power supply

giuliomoro

I have seen more than one issue with grounding on these SBCs, but I had not come across your specific one before.

Usually, battery power or transformer isolation on the audio inputs work fine, as you found out. Also, powering via ferrite bead had helped before. Now,cutting the trace connecting the analog and digital grounds on the Bela Mini cape worked very well in some cases, although this requires that an external ground connection is provided at all times. Careful with your sharp knife there so you don't damage the board by cutting nearby traces. For anyone else reading this: do this only if you are good at it, as it may void your warranty if this damages other parts of the PCB. Also, keep in mind that in order to "undo" this, you'll need to scrape off silkscreen and soldermask and join the traces back together with a bit of solder (well, or run another wire between exposed ground pads).

alt text

picture is for BelaMini Rev C1. Check design files for other versions: https://github.com/BelaPlatform/bela-hardware/tree/master/belamini.

quintosardo

giuliomoro Looks like this cut trace thing helped. I currently have the trace cut, the custom dual supply, the two stages filter on P1.1 voltage input and a ferrite bead there.

The custom dual supply works better if not left floating.

Looks like P1.1 is quite delicate and noisy and needs careful treatment, not just a wire going into it from a breadboard. There is a lot of radiated noise while using it (the breadboard has long wires, etc.)

quintosardo

Thanks Giulio! I'm testing this.
Currently, I obtained quite good results with this:
- Custom isolated supply for audio input buffer.
- Two stages LC filter at V+ input (quite big values, like two 2200uF capacitors).
Works quite well but looks overkill and still leaves the feeling of putting something on top of something wrong.
I don't think I had any result with a ferrite bead at V+ input. Looks like the noise is at high frequencies but also going down into audio because of modulations.
Testing the trace cut thing!

May be useful: there is no way of accessing the differential audio inputs of the codec instead of the single-ended ones, right?

Oh, I also saw this: Bela is not visible through USB (no IDE) while it is being powered like above.
(USB cable only between belamini and Mac, 9Vdc Boss supply feeding the breadboard, switching +5V regulator feeding Bela, with ground common to the 9V supply it is fed from, floating custom +/-9V supply, fed by the 9V supply, feeding the audio input buffer, guitar connected to buffer input and to floating ground).

giuliomoro

quintosardo Currently, I obtained quite good results with this:
- Custom isolated supply for audio input buffer.
- Two stages LC filter at V+ input (quite big values, like two 2200uF capacitors).

Are both or these needed together? Does the LC filter make a difference while using the isolated power supply on the input?

quintosardo May be useful: there is no way of accessing the differential audio inputs of the codec instead of the single-ended ones, right?

No, they are not broken out. Hard to fit everything on a tiny design like this.

However, if you are using only one input channel, you could feed a differential signal to left and right inputs and subtract those in software (assuming they have the same gain). While not exactly a fully balanced input, I have seen snr improve drastically with this trick.

quintosardo Bela is not visible through USB (no IDE) while it is being powered like above.

Hmm that's weird. USB it should be working all the time unless you are messing with the USB pins on P1.

quintosardo

giuliomoro Yes, I only use the isolated supply for the audio buffer, now. The two stages LC filter is needed. The most annoying part of the noise is a steady freq like 5kHz. One stage is not enough, the noise is still hearable.

Great, I'll try with the 3.3V part.

USB and IDE: no, just the normal connection.
Like if the grounds are moving too much?

giuliomoro

Btw see if you have any luck placing caps in parallel on the 3.3V line instead .

giuliomoro

Is the 5khz the switching frequency of your power supply?

quintosardo

giuliomoro No, it is 200KHz.
I also tried all this with the voltages all coming from a desktop power supply.
This whistling has always been there, since I connected to P1.1 for the first time, in many different configurations.

If anybody reading is interested to try, it is just an inverting opamp with two resistors.

quintosardo

alt text
Current situation.

quintosardo

The first problem in this thread looks like this?

https://forum.bela.io/d/1244-digital-noise-on-audio-out

... but there's no follow up by the user...

giuliomoro

With the trace cut, can you go back to having a common ground for the 5V and ±9V rails ? I checked a custom design we made in the past and that had the equivalent of that cut in the ground track, then a common ground for the 5V power supply for the PB (fed through a ferrite bead) and for the ±5V power supply for the THAT1570 preamp. The results were pretty good; both ferrite bead and track cut were needed, but we didn't need additional LC filtering. The 5V regulator for the PB was the Recom R-785.0-1.0, while a LM317/LM337 pair was used for the bipolar rail (we had an 18V AC input). All of this on a dedicated PCB. Did you have a chance to test the noise performance of your power supply?
Also, are you currently using the analog inputs for anything? If you have pots there that are used to e.g.: control the volume or other parameters in your test code, then any noise on that ADC would be coupled into the audio path by the software without appropriate treatment in the DSP.

Btw, why are you going for ±9V in the first place when the input into the codec is about 1.8Vpp?

quintosardo

giuliomoro All the grounds are together (including the floating one, not floating anymore).
I don't think that power supply quality is an issue as everything was tested with all the sources from a bench supply, too.
The +5V regulator is Recom R-785.0-1.0 (but I also tested it simpler, no regulator, 5V coming from the bench supply).
The analog inputs are not used, not connected to anything.

+/-9V are there because we want to prevent overload on input and to have good voltage on balanced output.

What sends me crazy is that without a buffer, everything is straightforward: connect guitar to audio input, power supply to bela mini, everything is clean.
As soon as I put the audio buffer in the middle, everything goes crazy (not simply noise but very strong noise).
This happens in all simplest configurations, including this one:

bela mini besides a breadboard, an op-amp with two resistors on the breadboard, guitar to the op-amp. Nothing else. Guitar to bela: perfect. Guitar to buffer (unity gain): disaster.

Same situation, guitar to buffer, bela mini powered only from the USB connection: everything fine.

External supply to bela mini and audio buffer can't go together.

giuliomoro

You could also try an isolation transformer on your input. Assuming no pathological condition with your 5V supply, that has solved similar issues, according to other reports on the forum.

giuliomoro

did you try an isolation transformer (see https://forum.bela.io/d/2330-uart-usb/9 ) ?

quintosardo

giuliomoro Uhm no, as the issues where there even without anything connected to the audio output (using the headphones, instead).

quintosardo

Several burnt units later...

Looks like the answer is:

a fourth order LC lowpass filter (two Ls, two Cs) on external power supply input.
another instance of the filter on Bela's power input.

The filters are relatively heavy (20mH inductors, 2000 uF capacitors), to go down to a 25Hz cutoff frequency.

The weird part of this is that all the literature talks about "radio frequencies" but discontinuous high frequencies can go down to audio.
All examples about switching supplies etc. show filters with high cutoff freqs (well beyond audio range).

I have a beautiful background silence, now, but still wondering if I'm missing something right.

giuliomoro

I meant to suggest an isolation transformer on the audio input, not output.

quintosardo

giuliomoro Oh, no, I didn't try as there's a guitar passive pickup there, so completely floating...

giuliomoro

I mean between the preamp and the BelaMini input

quintosardo

giuliomoro Right, good one, even though they share the ground and I imagine a transformer there to prevent connecting different grounds together.

giuliomoro

And then see if you can get away with a simpler power supply arrangement