Switching Noise DC-DC Converter

Is the switching noise what makes people not use dc-dc converters in sensible audio circuits such as DACs?
dc-dc converters tend to be far more layout sensitive that linear regulators and so harder for many DIY to keep noise free.

Yes they can work perfectly in audio and lots of professionally manufactured audio gear use them.
I have used then in many commercial audio designs of mine over the years however to be successful the design of the switcher and the physical layout of all circuity is pretty unforgiving.
Makes it hard for the one off DIY project to hit it 100% right the first time out.


I also do not agree that in a DAC 50mV of supply noise is of no issue. This will depend greatly on the power supply rejection of the DAC circuits.

Consider the DAC with a 3.3V rail and a 50mV ripple. That is a 1.5% error with 0dB power supply rejection. 20dB power supply rejection is still .15% error.
Any power supply ripple in a DAC reference supply will also add jitter as each DAC sample value beats with the switcher noise. Not good.
Of you are looking for -90dB noise or better with low jitter in your DAC then your DAC power supply noise best be in the sub mV noise level.

In my experience the way to get there was with a dedicated to the DAC alone fast linear regulator right on the PCB at the DAC IC.
The dedicated fast linear regulator removed all system power supply noise for the sentive DAC circuit.
 
Last edited:
  • Like
Reactions: 1 user
My audio home made rig has an (also) home designed SMPS: a 50KHz flyback converter current mode (UC3842 + STP7N60 + PC817 +BYW98 + TL431) an I never noticed such an effect. It disturbs a little AM reception but not at a level to injury me.

What kind of DAC do you run off that supply, and what kind of other circuitry?
 
Last edited:
Which regulator did you use? I can't think of a linear regulator that can suppress these noises, they are very fast.
The way I have designed complex systems that need good audio performance is as follows.

Distribute the main power source across the PCB usually as a 5V rail typically generated by a off board SMPS running between 80Khz to 200Khz.

For logic sections create local rails of 3.3, 2.5 or 1.8V as needed by using high speed, ultra small switchers.
These today come in very small packages and run at 200Khz to 4Mhz speed with high efficiency.

For local analog sections create the rail locally with high speed, low noise linear regulators like the TI tps7a20.
These are very small and very fast with excellent PSRR of about 45dB @ 1Mhz , 75dB @100Khz and 95dB at 1Khz.
The noise floor of these regulators is also very low @ 7uV.

Each local regulator is preceded by a Pi filter (CLC) to attenuate incoming high frequency noise on the global rail to below a frequency the local fast regulator can reject.
The Pi filter also prevents re-injection of high frequency noise from the local regulators and circuits back into the global rail.

This system of Pi filter isolated local regulators also prevents locally generated high frequency noise from circulating across the PCB grounds as all the power supply return ground currents of each (regulator and circuit function) island is kept local.

The recovered audio from the DAC is then passed to the output buffers at the PCB's edge as a differential signal to keep any remaining ground bounce out of the audio path.

I have found this strategy even with a very a busy board with many processors, radios and DACs can keep the analog noise floor close to the design limits of the DAC being used.
 

Attachments

  • tps7a20.pdf
    8.4 MB · Views: 4
Last edited:
  • Like
Reactions: 1 user
E.g. LT3042/45 or TPS7A47 have over 50dB PSRR at 10MHz.
Well, that's together with the capacitor at the output, that takes over, reducing the source impedance at the higher frequencies. However, the spikes in the opening post are unlikely to occur if such a good bypassing were already used by the OP ... It is a fundamental design / measurement error that I suspect, especially because the OP doesn't talk about it anymore.