A cautionary tale on the use of SMPS

[scintillation]

I recently decided to experiment with the use of some switch mode power supplies (SMPSs) for an individually isolated 9 V DC power supplies for guitar effects.

The net result was 10 KHz noise present on my signal (I think) from down mixing of each SMPS operating at different switching frequencies.

I think a lot is written about the use of SMPS in audio applications, and for a lot of applications they are appropriate, you just need to use different techniques to linear power supplies. I think in my case I went for the cheapest SMPS which I don't think were shielded, so maybe wrong component rather than wrong idea?

More detail here:

http://www.experimentalnotes.co.uk/blog/acoustics/switch-mode-power-supplies-in-audio-applications/

[gjcamann]

Don't give up. I'm sure there's ways to tame this beast.
Is this on a circuit board or breadboard?
I think your filtering could be improved. To use a PI filter, you will want a capacitor before and after the inductor.
See bottom. http://www.electronicproducts.com/Passive_Components/Magnetics_Inductors_Transformers/Fundamentals_Inductors_101.aspx
If you put a 100uF cap before the inductor you should get a cutoff of around 2K, and that should clean it up pretty good - in theory.
http://circuitcalculator.com/lcfilter.htm

Maybe you need a pi filter in addition to your current filter? 

[teemuk]

And in addition to filtering you probably need to think noding with much, much more care because power supply lines (including ground returns) will be infested with noise. Layouts and noding practices adequate with generic linear supplies might not be with SMPS. Also, you might find a need to filter the audio signal paths with "beads" and other EMI suppression filters.

Just for the record, it can be done and has been done without considerable noise, whether in signal lines or that which is just radiated around. Yes, it just needs different designing approach.

[R.G.]

My day job involves tech support for a 9Vdc switching power supply for pedals, so I've lived with the issues involved in this for some years.

As noted, running switching power supplies at high frequencies is entirely different beast from audio electronics. The slowest frequencies involved are in the radio frequency range, and the harmonics reach much, much higher. At these frequencies, the length, width, and even number of turns in direction of a PCB trace make big differences in whether the signal stays in the PCB trace or radiates out of the trace. An inductor, for instance, can be a 90 degree turn. Capacitors are not so much capacitors as little boxes of side effects.

And that's all before you start with the problems of how multiple such power supplies interact. You've described classical heterodyning, where two radio signals interact and produce two signals, at the sum and difference of the two frequencies. You're hearing the difference signal. We have run into this before when a pedal uses an internal switching power supply - like some digital pedals converting 9V down to 5V or 3.3V for the logic - and the internal switcher is not well thought out in terms of being "polite" to its own power supply.

So don't get discouraged and mark off switching power supplies for pedals or audio in general. But do recognize that the frequencies and techniques used are much more technically demanding than getting some germanium transistors and making a fuzz face. It can be done - but if it was easy, all pedals would already do it.

[teemuk]

I'm not well-versed with high frequency switching  thingies, whether class-D amps or SMPS, but one thing I remember is that you preferably should use the same switching frequency for all of such circuitry. One of biggest obstacles of self-oscillating class-D amps, for instance, is that they can't be sync'd to other switching circuits in the amp, whether they are other amp stages or switchmode supplies.

I'm not really sure what the underlying principle is behind the need to use the same switching frequency throughout but it definitely has to do with something R.G. mentioned.

[R.G.]

You'd pick it up fast enough.

The principle is that any nonlinear process on two signals produces a sum and difference signal - literally what we call intermodulation distortion in audio.

If the two signals are locked, the sum is 2x and the difference is 0. For RF, 2x frequency is even further from the desired passband, and zero - that's DC, not in the audio band at all.

The problems come when the difference in the signals is within the audio band. When you're dealing with 100kHz up to 1MHz, a sneeze can make one of them be 100Hz-10kHz different from the other, and you hear it as audio.

The inductive nature of traces makes this worse, because a copper trace that's  a milliohm at DC can be many ohms or several K at RF.