# Can the noise of any switching PSU be attenuated if I put linear regulator before the output?

A friend told me that noise of any switching PSU can be attenuated if I put linear regulator before output. Is that true?

For example, if I want to power a +-12 V op-amp for an amplifier, I can use a switching-mode power supply (SMPS), say, with a noisy 15 V output and then from the SMPS output feed an LM7812 and a LM7912.

Will the output from LM7812 and LM7912 now have very very low noise compared to their inputs?

If this is true, this amazing as there is no need to use a transformer anymore.

Is it really correct that a heavy PSU using a transformer for Class A and B amplifiers is no longer needed?

Yes, it is true that adding a linear regulator after a SMPS (switch mode power supply) will reduce noise, but care is still needed. Results can be very good, but the result may not be as good as if a mains powered transformer plus linear regulator had been used.

Consider a common LM7805 5V regulatorfrom Fairchild. This has a "ripple rejection" specification of 62 dB minimum. "Ripple" is input noise but usually related to the twice mains frequency variations from the rectified and smoothed mains input. This is a reduction in noise of 10^(dB_noise_rejection/20) = 10^3.1 ~= 1250:1 That is, if there was 1 Volt of "ripple" at the input this would be reduced to 1 mV at the output. However this is specified as being at 120 Hz = twice USA mains frequency, and no specification or graph is given for noise reduction at higher frequencies.

The functionally identical LM340 5V regulator from NatSemi has a slightly better specification (68 dB minimum, 80 dB typical = 2500:1 to 10,000:1) at 120 Hz.
But NatSemi kindly also provide a graph of typical performance at higher frequencies (bottom left corner of page 8).

.

It can be seen that for 5V output ripple rejection is down to 48dB at 100 kHz (=250:1). It can also be seen that it is falling about linearly at about 12 dB per decade (60 dB at 10 kHz, 48 dB at 100 kHz) . Extrapolating this to 1 MHz gives 36 dB noise rejection at 1 Mhz (~= 60:1 noise reduction.) There is no guarantee that this extension to 1 MHz is realistic but the real result will not be letter than this and should (probably) not be much worse.

As most (but not all) smps supplies operate in the 100 kHz to 1 MHz range one can guestimate that noise rejection will be in the order of 50:1 to 250:1 in the 100-1000 kHz range for fundamental noise frequencies. However, smps will have output at other than their fundamental switching frequency, often much higher. Very thin fast rising spikes which may occur on switching edges due o leakage inductance in transformers and similar will be less attenuated than lower frequency noise.

If you were using a smps by itself you would usually expect to provide some form of output filtering and using passive LC filters with a linear "post regulator will add to its performance.

You can get linear regulators with both better and worse ripple rejection than the LM340 - and the above shows you that two functionally identical ICs can have somewhat different specifications.

Noise elimination from smps will be greatly helped by good design. The subjct is too complex than to do more than mention it here but there is much good on this subject on the internet (and in past stack exchange replies). Factors include proper use of ground planes, separation, minimising area in current loops, not breaking current return paths, identifying high current flow paths and keeping them short and away from noise sensitive parts of the circuit (and much more).

So - yes, a linear regulator can help reduce smps output noise and it may be good enough to allow you to power audio ampliers directly this way (and may many designs do just that) but a linear regulator is not a "magic bullet" in this application and good design is still vital.

• But you're going to use a low-pass filter to get rid of the MHz stuff before it gets to the linear regulator in the first place Aug 12, 2011 at 14:46
• It's common to see post-linear regulation in power supplies that have extremely low noise requirements, like medical units for ultrasound. It's noise in the kilohertz that the linear regulator will excel at attenuating (switching ripple, any beat frequencies from multiple converters, etc.) Aug 12, 2011 at 20:44
• Would having two of those LM340s in series then provide 96dB ripple rejection? Jun 10, 2013 at 22:58
• @us2012 - 96 dB using 2 x LM340 is series? = Conceptually yes, in practice = maybe. 96 dB = 63,095:1. Or about 16uV of noise out per Volt of ripple in. If your SMPS noise was around 100 mW to start you are asking for about 2 uV in the output. It is very Very VERY easy to "find" 2 uV of noise along the way. eg if your output was 5V at 100 mA Rload =~ 50 Ohms. At 100 kHz you'd need about 0.5 pico-Farad of capacitance into a 50 Ohm load from a 100 mV low Z signal source. ... Jun 11, 2013 at 12:09
• @us2012 ... Add all the other non-idealities, couplings, unexpected leakage paths and more and you require leading edge performance and a vast practical experience base and understanding of the subject to achieve that sort of result. Jun 11, 2013 at 12:09

A linear regulator has a limited bandwidth on which it can regulate. High frequencies are passed through. How good a regulator dampes frequencies is found in the ripple rejection. Look up a LM317 datasheet and search for graphs of ripple rejection ratio's against frequency:

It depends on load current, input and output voltages and apparently also if you put a capacitor on the Adj pin. Moreover it drops rapidly at frequency. Most specifications are made at low frequency so it works perfectly after a transformer (which is likely to be a 100 Hz or 120 Hz ripple).

If you get a typical SMPS of these days, it may switch at several hundred kHz. Apparently a LM317 with a 10 uF capacitor on the adjust pin only manages 40 dB at 100 kHz and 20 dB at 1 MHz. A 1 MHz 1 Vpp ripple would still pass through as 0.1 Vpp ripple out. At higher frequencies it will only get worse and drop down to 0 dB, which is no amplification nor damping.

This is a cheap LM317 regulator, there are better ones on the market. LDO's are typically not as good in ripple rejection because of their nature of being a bit less stable.

Alternatively you could use a LC filter to damp the high frequency stuff. Note however that a LC filter has a resonance frequency, which may attenuate a certain frequency tens of times instead!

I can't see (unless your regulator is oscillating) a linear regulator would amplify noise instead. Sure, it will always add broad spectrum noise (temperature noise, flicker noise, etc.), but so will transistors, resistors, opamps, diodes, etc.

However, because you're talking about audio, I would like to add to that specific situation:

• An op-amp has its own PSRR (power supply rejection ratio) as well. Some components don't have graphs for this figure, however this adds to your linear regulator as well. A AD8622 precision op-amp has about 20 dB - 40 dB damping at 100 kHz. (Positive supplies are typically better damped than negative supplies).
• If a SMPS switches above 400 kHz, would you mind/hear the noise?

Like Hans says, a linear regulator won't stop the HF noise from an SMPS. You can filter if with passives like capacitors and coils. Because the frequencies involved are a lot higher than the 100Hz ripple you have to get rid of in a classic supply you won't need that big electrolytics. (These electrolytics have to be big because they're most often the only way of "regulating" the rectified voltage.)
So passive decoupling is the word. If you really want to use a linear regulator you can use an LDO since its input voltage won't vary.

BTW, you still need a transformer in your SMPS of course, otherwise your amp may be a shocking experience. But you can make it a lot smaller than the classic ones.

The main thing you need to do is route your traces properly. If you connect your audio signal to the ground right next to the SMPS, and then have a linear regulator after this, it won't do you any good. You need to "pipeline" the ground traces from one stage to the next, and connect your audio circuitry to the ground at the output cap of the linear regulator.

Wires are not perfect conductors, and a noisy current passing through a ground node will cause the voltage to fluctuate. Using a fluctuating ground as your audio reference means the fluctuations become part of the signal.

Torroidal chokes & Low ESR caps also reduce ripple which may be simpler to reduce 40 db or more & eliminates need for LDO regulator.

here is some more information supporting a few of the options Russell already explained in excellent detail.

Page (9) of the artical I've attached is definitely worth noting, as the characteristics curves of ferrite beads are another excellent consideration for high frequency damping, but very seldom are used.

Again no magic bullet, and ferrite has a smaller window of useful application then a common LC or RC circuit because its effect aren't as drastic, but the big take away is its effect on impedance without the common side effects associated with the other two options, and used in the right place, ferrite can have an exceptional effect on stability.

As Peter asked earlier, regarding audible noise, it is very true that filtering within an audible frequency band, say 20hz-20khz; can be a quick way to make a power supply very usable. We see this in RC filters in guitar amps all the time. In my experience especially in audio instrument amplifiers, this only becomes more true when the end>>engineer, is in fact a traditional output transformer which has a cutoff freq generally between 20khz-10khz, which then couples to a traditional metal frame speaker, and as is the case for guitar, these speakers are commonly attenuated to have a cutoff of around 8Khz.

So we start to raise the eyebrow even at 100khz noise, not worth the effort.

But in practice it is a different story, because as we know the fundamental frequency of interest tends to do no one any favors and naturally creates harmonics of itself, extending all the way down into the audible range. If the fundamental frequency is inherently noise, this becomes an elusive control measure, because it so often times includes more then one fundamental frequency, and using both RC and LC filters can have off putting effects by changing the "tone" of the noise more then treating it. Thus you can see how easy these effects can create a run around on paper.

So to accommodate for this, getting into the right ballpark can sometimes be as easy knowing the characteristics of the Ic we choose, or any inherent characteristics of the power supply design we choose. After that point, making sure to approach noise with equal considerations both in the audible frequency, and the high order frequencies can give profound results.