I have a question about switching supplies, regulators and filtering. I am working on a device powered by a +12V switching supply that has an analog circuit to process decent-quality (objective, I know) line-level audio. The analog section is powered by +10V, and there is another circuit section (also-analog but just LEDs, switches) powered by +5V. I was thinking of using a linear LDO from +12V -> +10V for analog section power, then a switching regulator for +12V -> +5V for efficiency purposes, since noise on that circuit isn’t as critical. The circuits will be separated on the board, but I’m wondering what the best way is to minimize the influence of the switching supply and regulator on the analog section.

My idea was to run the switched supply +12V line directly to the +5V switching regulator with no filter other than the recommended filter caps for the regulator. For the analog audio section, the +12V switched supply would run through an LC or PI filter, then into the LDO regulator down to +10V. In other words there are two separate branches directly from the +12V supply. Is this a reasonable approach? Or does it make sense to filter with an LC directly from the main +12V supply, then branch off into the two separate regulators? I thought maybe some of the +5V switching noise might carry over if I didn’t use LC to isolate the audio section. To be honest I don’t even know if using an LC filter is necessary, but I figure with the right damping ratio it can only help.

Also, any recommendations on corner frequency for the LC filter? Not sure what frequency the switcher is running and the likelihood of harmonics in the audio range, it’ll most likely be a +12V wall wart from CUI or Phihong. Thanks for any input.

EDIT: BTW for anyone coming across this question looking for information on input filters for switching supplies, I'd recommend this 5-part video series on youtube from Kainkalabs. It's the best explanation of switching noise and filtering I've come across yet.

  • On converter choice

Checkout LT and TI's low-noise switcher lineup. They have good parts. Using higher switching frequencies counter-intuitively makes post-filtering easier, as you can use smaller and more practical LC filter values, and ferrite beads are better at rejecting HF than LF.

  • About your 12V-5V buck:

Remember a buck draws high di/dt square wave current from its input, so it generates more noise on the input than on the output! It will then go through the +10V LDO and back into the opamps. If the 12V comes from a cable, this can also turn your cable into an antenna.

Thus, consider adding a simple LC filter on the input of your buck.

  • About your inverting regulator

Input current is smoothed by the inductor, but output current is again a high di/dt square wave. Combined with output capacitor ESL, this means nasty switching spikes in the output. Output filter will be really necessary to keep your opamps happy.

Now read this:

http://cds.linear.com/docs/en/application-note/an139f.pdf http://cds.linear.com/docs/en/design-note/dn95f.pdf

So I would go like this:

Input -> common mode choke and caps -> main +12V_IN.

+12V_IN -> LC filter -> LDO -> +10V

+12V_IN -> LC filter -> +12V_Dirty -> Buck -> +5V

+12V_Dirty -> Inverting DCDC -> LC filter -> -10V

The LC filter prior to +10V LDO can be cost-cut if you want. If this is a hobby build, better spend an extra buck that discover later than performance sucks.

Now, about those LC filters...

Since you're gonna use high-frequency switchers like >500kHz then a small 10µH inductor or ferrite + 10-10µF MLCC provides excellent attenuation. Make sure you layout the caps for minimal ESL (use several vias) and select an inductor with high self-resonance frequency (low parasitic capacitance).

Now, you don't want your filter to ring, so you need some resistance (calculate optimum damping factor). This resistance can be implemented either as a plain resistor in series with the inductor, or the inductor's ESR. A better solution is to use a larger aluminium capacitor in parallel with the ceramic cap (like 100µF).

Higher capacitance lowers the amount of resistance needed for damping, plus it makes a better filter, and the caps' ESR is usually enough to make things well behaved if you select the right one.

Now, notice I didn't put a LDO at the output of the switcher for -10V. This is because we already have a fast DC-DC providing regulation... output voltage will be tightly regulated at low frequencies. A LDO providing better rejection than the LC filter would have to be rather expensive. You can put a TPS7A47 if you really want to overdo it, but it aint gonna be cheap!

  • \$\begingroup\$ Ringing of the filters? For near-optimal (Q = 2), use Rdamp = sqrt(L/C), thus 100nH (4" wire is 100nH) and 100uF needs..... sqrt(100nH/100uF) = sqrt(0.001) = 31 milliOhms. \$\endgroup\$ – analogsystemsrf Jun 17 '17 at 14:27
  • 1
    \$\begingroup\$ Yep. But a ferrite bead or real inductor will have much more L than 100nH, and with a ceramic cap the ringing can be in the audio band, so I'm warning the OP about that... \$\endgroup\$ – peufeu Jun 17 '17 at 14:51
  • \$\begingroup\$ I should mention that rather than +-10V rails, the line-level signal is small enough to use a TLE2426 railsplitter to bias the opamps and still achieve max signal range, so this was my plan. This eliminates the need for the buck converter and extra filter. Based on my understanding of your post, you would use a total of 4 filters in your case, correct? If I were to use a higher roll-off filter at the input (12V), and buffer that +12V_IN from the +5V Buck with an LC, just curious what might still cause the analog section to pickup noise? I don't mind spending the extra money, just curious. \$\endgroup\$ – User7251 Jun 18 '17 at 20:29
  • \$\begingroup\$ Yeah, filter could be BLM41PG102SN1 (ferrite from murata, you can get spice model on their website) + 1µF MLCC // 100µF electrolytic. Now, TLE2426 is crap for audio as its output stage distortion is high, it is class B with very low idle current and rather slow. Just use the same opamp you use for the rest of your circuit, and bias it with a voltage divider filtered by a cap... and you have your virtual ground. One less part in inventory too. \$\endgroup\$ – peufeu Jun 18 '17 at 20:36
  • \$\begingroup\$ Are you referring to the TLE2426 3-pin or 8-pin chip as noisy, or both? The 8-pin has a noise reduction pin which helps significantly. Interestingly I started off with an LF411 follower with resistor voltage divider and bypass cap to bias the other amps before I went to the 2426 and it was pretty dead on for a number of hours, the 2426 does seem to float around a bit, I'll have to re-test. I thought the 2426 would save me some components and board space. The other amps are NE5532's, which have a 30k minimum input resistance ... should I use one for biasing or do you have a better suggestion? \$\endgroup\$ – User7251 Jun 21 '17 at 13:42

An LDO has little or no ability to remove high-frequency trash on its input(unregulated) pin. You must filter out all that trash from the SwitchReg.

enter image description here

Here is the (3 bypass cap) network, with ESL in each bypass cap, that generates those resonances out at 3MHz and 30MHz.

enter image description here

  • \$\begingroup\$ Thanks analogsystemsrf, do you by chance have the component values used in the model for that response? \$\endgroup\$ – User7251 Jun 21 '17 at 13:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.