Switching PSU and residual noise on the rails

Question

Is it ever possible to totally eliminate (at reasonable cost) switching noise on the power rails from either a switching PSU or DC-DC converter? At least, down to the sub-mV level.

I'm measuring it using an oscilloscope with a high impedance probe.

Answer 1

You have to consider the frequency of the noise you want to get rid of. This is very important.

• LF noise/ripple, like below 1 kHz, or 100Hz ripple.

Do you actually need to get rid of it? Most likely your opamps have 100dB PSRR in that frequency range so they don't care. Also most linear regulators have very good PSRR at low frequencies, so they will clean up your supply just fine. You can even use a switcher to get rid of LF ripple, but you get more HF noise in exchange of course.

• Medium frequency, like the 20k to 500kHz sawtooth at the output of your DC-DC or AC-DC switching converter.

As frequency gets higher, most regulators and LDOs begin to suck in the PSRR repartment, and the opamps you power too, so this is the part when you need to look at datasheets in greater detail. Some regulators have pretty good rejection in this range, others are abysmal, especially the micropower ones, which are optimized for low idle current, and thus have rather slow control loops.

• HF noise like a converter's switching glitches.

This will go straight through most regulators and opamps.

The best way to handle the last two is the good old RLC filter.

Usually, in this application, ferrite beads are better than inductors. They use a lossy core material which allows them to turn into resistors at HF. Also they usually have less interwinding capacitance than an inductor (ie, less HF goes straight through). Murata lets you download spice models, so you can play with them in simulation.

So, you select a ferrite bead based on DC current (do not saturate it, or it loses its inductance), its inductance value (for low frequency filtering) and its high frequency impedance (for HF filtering). Or course higher inductance/impedance in the same package lowers the saturation current, so size and price matters.

Then you put a capacitor after it. Its value depends on the frequency of the noise you want to attenuate, of course. This is, after all, a standard LC lowpass filter, so its cutoff frequency should be properly placed...

Noise around 20-100 kHz is a bit problematic, because frequency is high enough to make the PSRR of your regulator suck, but low enough to require annoyingly large L and C values. Good choice of parts is important there.

On the other hand, and this is a bit of a paradox, switching noise at 1MHz is much easier to get rid of, since your LC filter has more frequency to work with. The ferrite bead will have a much higher impedance, and a small ceramic cap works wonders.

For HF noise like switching spikes, it's all about layout and ESL.

Answer 2

Quick answer: no, you can only remove a finite amount of noise/ripple.

Real answer: pretty much, you can remove a lot of noise with cheap(ish) components.

Long answer: if you design the power supply and know all the situations it is running in you can (or should) be able to remove ripple down to pretty much nothing. You'll notice when reading datasheets that ICs will often state the VCC levels to be within some bounds, which effectively shows you the ripple you're allowed to work with. As a general rule, it's often plus-minus a couple of percent of the value. Some devices are very picky, such as some RAM ICs I've worked with in the past that needed to be within 5mV (low power DDR3 I think it was). While other devices can take a huge range (some 555 timers I've used say 3-15V). Obviously the faster, more sensitive, a device is the less ripple you can get away with.

Don't forget: the ripple from the supply switching is only one source of noise/ripple on the power line. Other impacts are the changing current draw from the components, external noise sources and so on.

Back to the question; you remove ripple as much as you need, rather than as much as you can, in order to save cost. Apply enough capacitance and the line will get very smooth, even with large current gulps (which can cause large voltage drops). However, as the power line/plane then goes over the rest of the PCB, it will pick up noise from environment and components, which is one reason to spread capacitors around the PCB layout.

Often you have some fussy (sensitive) ICs needing very precise and stable voltage rails, but there is no need to keep all of the line within tolerance. Therefore you'll often see a power rail isolated with a ferrite bead or a full T or PI filter to give a very stable supply for some very sensitive parts (such as an ADC module).

EDIT: One more thing to add, if you're only worried about the switching ripple, you can measure it quite easily via the use of a simulation tool such as LTSpice. Of course, you need to make sure your models and simulation matches reality. But it is usually worth checking.

Answer 3

Yes, you can significantly reduce noise on your power supplies by following the SMPS/DC-DC convertor with an LDO. There are even integrated solutions such as the LTM8068 or the TI TPS54120. The Linear datasheet shows the typical output of the DC-DC with 200 mV ripple, and about 600 uV after the LDO, and the TI part has an amazing 38 uV noise.

You could create your own discrete DC-DC convertor - LDO design easily to cope with any required DC voltages. This will reduce your overall efficiency of course.

Answer 4

ripple voltage is all about filtering with impedance control.

You have a current switch with Rce or RdsOn and a load cap with ESR.

The ripple voltage is always ΔV=ΔI*ESR at a minimum.

You can consider the Ic=Cdv/dt for pulsed voltages but most SMPS regulators use CCM in high current mode where you have most problems and DCM in low current mode can be resolved with successively smaller C with lower ESR in parallel at some Zc(f).

Therefore you can choose the lowest ESR caps that you can afford or use LC filters with plastic film and low ESR chokes outside the loop to prevent reduced feedback ripple gain reduction in the control loop which affects overshoot from a step load.

From intuition, you should realize that you don't need massive energy storage to filter a current ripple , rather higher frequency switchers with very small low ESR coils and ultralow ESR small tantalum or plastic film caps operating >1MHz. But this requires more skill in the switcher design with EMI controls and better semiconductor switches that match the transition frequency , current and power topology of the design, be it SiC or Si or some other structures like IGBT's.

Ripple measurements should always be done with 200MHz spring probes <=0.5cm on a calibrated 10:1 probe or better with FET buffered Diff probes.