PSRR for Linear Regulators

[LoveLaika]

I'm looking at linear regulators to step down a voltage from a switching regulator to a fixed voltage (i.e., turn 8-10 volts into 5 volts). I'm a bit confused about the PSRR. My switching regulator (LM2679 and LM2673) operate at 260 kHz. Most of the linear regulators I've seen have a PSRR at 60 dB at 120 Hz. That figure drops as the frequency increases. So, for example, most data sheets I've seen reach ~35 dB at 200 kHz range. At that level, assuming the formula: dB = 20 * log10(Vin/Vout), the ratio between Vin/Vout for ripple would be 56x; if I'm understanding it right, the regulator would reduce the ripple by 56x (comparing the output ripple to input ripple)? Is that value good considering my input frequency?

[Ian.M]

Dumb question.  You've studied enough linear reg datasheets to come to the conclusion that  ~35 dB PSRR at 200 kHz is fairly average, so you've already got the answer its neither bad nor good.  The smart question is: Is a ~50x reduction in ripple good enough for your application?  Also note that the PSRR is almost certain to drop off further at higher frequencies, so depending on the harmonic content of the ripple, you may not see as great a reduction in peak-to-peak ripple as the simplistic calculation suggests.

[Zero999]

Yes, the PSRR declines, with increasing frequency and you've calculated the correct ratio. If it's not good enough, then you might be able to improve it with an RC pre-filter, but watch out: the PSRR also goes down, when the input voltage is close to the drop-out, do it might not gain you much.

[timenutgoblin]

I'm looking at linear regulators to step down a voltage from a switching regulator to a fixed voltage (i.e., turn 8-10 volts into 5 volts). I'm a bit confused about the PSRR. My switching regulator (LM2679 and LM2673) operate at 260 kHz. Most of the linear regulators I've seen have a PSRR at 60 dB at 120 Hz. That figure drops as the frequency increases. So, for example, most data sheets I've seen reach ~35 dB at 200 kHz range. At that level, assuming the formula: dB = 20 * log10(Vin/Vout), the ratio between Vin/Vout for ripple would be 56x; if I'm understanding it right, the regulator would reduce the ripple by 56x (comparing the output ripple to input ripple)? Is that value good considering my input frequency?

Have you considered using two cascaded linear voltage regulators? If the required output voltage is 5V then you could use a 9V linear voltage regulator to supply the 5V linear voltage regulator. Two linear voltage regulators with 35dB of PSRR each gives a theoretical 70dB of PSRR. Instead of 56x you will have 3162x the ripple rejection (in theory). Having two linear voltage regulators will increase power dissipation. Your switching regulator will need to output a higher voltage (at least 13V including voltage ripple).

Alternatively, have a look at The Signal Path video regarding using a Power Amplifier for noise reduction.

[LoveLaika]

Thanks. At least I'm going in the right direction. I don't think drop-out voltage will be a big issue. I got plenty of wiggle room for input voltage.

Now that you mention it, I do recall seeing power supplies with an input filter before going to a linear regulator. One board that I tore down appeared to have a Pi-LPF-filter (at least, judging based on the traces) before going to a switch, which then went to an input cap of a buck converter (the second C in the filter). So, as per your suggestion, adding something like that would filter the high frequencies before reaching the input of the linear regulator.

[LoveLaika]

Thanks for the idea, though I can't implement it due to limited space. That is something to think about though, something that I never considered, so thank you for that idea.

[radar_macgyver]

There's this video from Jim Williams on the topic, with its associated AN101.


https://www.analog.com/media/en/technical-documentation/application-notes/an101f.pdf

In summary, the switching noise has much higher frequency components that are harmonics of the switching frequency, which linear post regulation will only have a modest effect on, if any. LC filters as suggested by Zero999 should be used with care to avoid ringing. Ferrites are absorptive, so they are much less likely to ring.