4
\$\begingroup\$

We have a switched power supply with a sw frequency at around 70 kHz.

In our DAC PCB, we use an LDO with a PSRR of about 60 dB reduction at 70 kHz.

We are thinking of adding a LPF (LC or PI) before the LDO to attenuate even further at least 30 dB at that frequency (all this because we want to clean the power supply thus having better sound at output).

  1. Is it a good practice to add LPF (LC or pi) before the LDO?
  2. Will the PSRR add of LDO and LPF (the LPF and LDO are in series)?
  3. What LPF filter is recommended (pi, LC) or does it only depend on how many dB of attenuation we want to achieve?
\$\endgroup\$
1
  • \$\begingroup\$ TI AN2162 is a good starting point. \$\endgroup\$
    – Matt Young
    Commented Nov 18, 2015 at 15:54

1 Answer 1

2
\$\begingroup\$

A pi filter could mean anything but if you mean one that is an inductor with a capacitor to ground either side then my answer is exactly the same because, the input capacitor i.e. the one receiving power from "upstream" is, in effect, in parallel with the upstream bulk capacitor in the "switched power supply".

So, an LC can do a lot of good in reducing ripple fed to the input of any linear (or switching) regulator but it can do serious harm if not properly designed. For instance if the load (down stream of the LDO regulator) took pulses of current, the back emf due to that current changing in the inductor can be enough to "double" the input voltage to the regulator it feeds. It won't stay double for long because it's a transient event but, this transient event is enough to kill silicon if not sufficiently voltage rated at it's input.

This is hard to prevent so my advice is make sure that your regulator (fed from the LC) can handle (as an input voltage) at least twice the normal running voltage. The C in the LC will alleviate this situation considerably but I've fallen foul of this recently so I am speaking from experience!

You could choose to put a zener diode at the input to the regulator to catch large excursions - this is an accepted way of dealing with things but the zener needs to be rated a volt or so higher than the worst case supply voltage under extreme quiescent conditions.

Assuming you have got this all sorted out you still have to be aware that an LC filter has the potential to exacerbate ripple at the "wrong" frequency and this could make the regulator output even noisier. What I tend to do (under these circumstances) is design the LC so that it isn't under-damped: -

enter image description here

If \$\zeta\$ is small, any input ripple frequencies occurring at the point of resonance will produce output ripple that could be several dBs higher i.e. putting the LC in actually makes things worse. So, if you can drive a simulator, engineer a solution with a small value resistor in series with the inductor - it can't be a big value because it will heat up and degrade the level of DC voltage feeding the regulator.

Other than the above it is worthwhile especially as the load is a class D amp because class D amps are notoriously bad for power supply rejection.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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