Ferrite bead/ Capacitor filter causes ringing in output?

[Sjoertdb]

I am testing the output voltage on a 3V3-buck regulator (TPS62175). On the output i see around 18 mV peak to peak ripple, which is completely fine.
However, once the signal passes a ferrite bead/ capacitor filter, the signal is distorted with alot of ringing. Could this be a resonant filter?
I've read a small article on Q-factors that might be too high. If someone has experience with this, could you share this?
Thanks

Sjoert
Design Engineer

[Siwastaja]

Easily. Very typical.

Either an AC lossy inductor, or an AC lossy capacitor, or both, tend to help. For the inductor, being lossy means equivalent parallel resistance; for the capacitor, it's equivalent series resistance. This means turning some part of high frequency components to heat, instead of storing and releasing as reactive energy.

Ferrite beads are supposed to be lossy, but they always aren't, and even those that are, stop being lossy once saturated, and the saturation current tends to be unspecified (a lot less than the thermal max current).

So make sure the bead doesn't saturate; and if the ringing is still an issue, consider using a resistive capacitor. If it's an MLCC, add a series resistor, or use a tantalum (with proper derating & peak current analysis) or aluminium electrolytic (with proper lifetime derating/analysis).

[Sjoertdb]

Thank you very much for your reply Siwastaja 
I will replace the ferrite for a type that does not saturate. I've spent days looking for this problem

Thumbs up for you 

[Sjoertdb]

I've measured the Q-factor using an LCR-meter.
In the oscilloscope plot there clearly is a periodic pattern of 13 kHz visible.

When the LCR meter is set to 13 kHz, the Q factor is 1.74 and the inductance is 10.9 uH.

The filter consists of a 100 nF -> bead (1000R at 100 MHz) -> 1 nF capacitor MLCC.

Should this configuration be sensitive for ringing?

[ConKbot]

A bead that is 1k ohm impedance at 100 MHz (if it's normal 0805-0402 size) is more meant for signal lines, though still perfectly usable for a few mA of power. Look at the datasheet for it. It should have curves for resistance, inductance and combined impedance vs frequency.  Down at 13khz, almost no ferrite beads are going to be very resistive, so you're working with an inductor that isn't very lossy at that frequency.

Parallel a resistor,  in the 50-500 ohm range, across the ferrite bead. It will damp the LC ringing, at the expense of less attenuation from the bead. 

[Siwastaja]

And, if your current is more than a few mA, shop for inductors sold as "inductors"; for some strange reason, they list saturation current in the datasheet, and even in the supplier's parametric search! 

Then, you can add the parallel resistor to introduce the AC loss.

[T3sl4co1l]

13kHz suggests reg instability, or something about your load (is it varying at 13kHz?), not LC resonance:

1nF is 12kohms at 13kHz, but 11uH is 0.9 ohms.

You would need 12uF for that to be the case; the resonant impedance would then be around an ohm, so if this is done with a lossy (tantalum or electrolytic) type with ESR = 1 ohm, the network will be reasonably well damped.  I would use more, on both sides (say 0.5 ESR, 47uF), to further lower it and keep the reg happy.

If there is more load -- and bypass -- after the ferrite bead, that is part of the system as well, and you need to include that in the capacitance figure.

Follow the reg datasheet's recommendations on capacitor size and ESR.

Tim

[Sjoertdb]

Thanks for the great input guys! It's greatly appreciated. 
i should add the note that this ferrite filter configuration is added 3 times, so it forms a 3 stage filter.
The picture i added is a scope plot behind the 3 stages. At the regulator's side, the output voltage is completely quiet.

The problem arises when the signal passes these filters. When i remove the ferrites for a resistor of 33 ohms, the signal has an even worse periodic pattern.
The load is an MPR121 capacitive touch controller. This IC has a maximum current draw of 29 uA according to it's datasheet.

[Sjoertdb]

And if i remove the beads and replace them for a 33R resistor to form a LPF at 4.82 MHz.
I see the following on the scope.

The signal also crosses trough a ribbon cable.

[Siwastaja]

How do you bypass the supply of MPR121 after the filters? Which kind of capacitors? They may participate in the ringing/noise. - or, the supply is just poorly decoupled, and the noise you are measuring is just caused by the peaky power consumption of MPR121 itself, and your filter works great to prevent it being visible on the regulator end .

Why do you think MPR121 requires such power filtering, or is it just a test load to experiment with the LC filters in general?

[Siwastaja]

I'm seeing dips at almost exactly 1.0ms intervals - these can't be anything else but the chip drawing current in pulses, like digital ICs tend to do. Maybe your "sample interval" is configured to 1ms?

Note that the current consumption is DC rating - average current. A digital IC can easily eat peaks 1000-10000x that. The most important aspect is to bypass the highest frequency components by providing a tiny bit of very low-inductance capacitance directly to the power pins - your typical 100nF cap. This tends to take care of CMOS switching, so nanosecond level stuff - but it does not filter for "bursts" of processing, your >>microsecond range. So you see the IC supply voltage fluctuating. The small bypass caps should take care so that the fluctuations are slow enough (dV/dt and dI/dt) not to cause noise coupling (EMI), but they can't "regulate" the DC bus on the mid-frequency range (tens to hundreds of kilohertz) where your typical 100n cap can't do anything, but your input power supply regulator cannot catch up, either. For this, you need a bigger capacitance (if you really need to filter it, anyway).

Now, it's usually acceptable to see some noise on the digital supply lines. Is there any specific reason why you are trying to filter it?

In any case, you need a steady source of charge for the chip. If you want to filter such a low-frequency bursts, the typical 100nF bypass cap isn't enough. You need way more. Try adding a large 4.7u or 10u MLCC and see what happens. (Now, adding this MLCC probably creates real conditions for the LC ringing to happen, so you may want to consider using a cap with some ~1-2ohms of ESR instead.)

[Sjoertdb]

Thanks Siwastaja 
I've had a look in the software I2C registers of the MPR121. Indeed, as you assumed, the sample interval was set to 1 ms. 

The reason why the MPR121 needs a clean supply, is because the product is not properly reacting to touches. After adding a different supply and an LDO, the issue was gone.
The reason of faillure was a noisy supply.

I'll try adding a capacitor with some series resistor soldered into place. I'll try a 10uF one to see if the supply gets better.

I really appreciate your help! I feel i'm close to the final solution.

[Sjoertdb]

I've tried a bit of modification with a 10 uF MLCC and a 2 ohm series resistor. The scope has the following plot when the voltage is measured.
It shows a delta of 146 mV, which is a little bit too high. What could be done to reduce this under a 100 mV?

I've tried a bulk capacitor of 220 uF, and it reduces down to 76 mV. Is there another solution instead of such a bulky capacitor?

[Siwastaja]

You can replace the large bulk capacitor with a regulator with a fast control loop - meaning you don't need to store as much energy, the regulator takes care of keeping the output correct, and can do it by having extra input voltage available (which is allowed to dip more). In practice, this may mean modern fast LDOs, that can easily respond in megahertz range. Most switchers can't, because the switcher's response is inherently limited by its inductor.

Although, a modern high-frequency switcher should be able to do it as well. I mean, the cheapest, most integrated, modern one-chip buck running at, say, 1.5-2MHz and with a tiny inductor - and hence, relatively small capacitors as well. You shouldn't need 220uF...

This single positive peak with fairly slow rising edge (100us) is bugging me. It doesn't look like inductive ringing to me, but like the device suddenly stops consuming current for a while.

Can you post a schematic, or even better, an actual photo or layout picture that shows the full regulator, it's output caps, any filters you have currently in place, and finally, the bypass capacitors for the load?

[Sjoertdb]

The schematic of the 3V3 is as follows (in the attachment). I also have a 5V line coming in, i should be able to add an LDO to that line with a very fast response time.
Or accept the 200 mV drop when i put it to the 3V3 line.

[Siwastaja]

One thing you need to understand is the timescale of your measured peaks (~100us). The 100n let alone 10n caps can't do anything on this timescale. They are useful in your filter if you have an EMI issue in tens-hundreds of MHz range.

Local regulation using a fast regulator is the key. If you look at the typical transient response curves of your TPS62175, it should react in a few µs, even in pulse-skipping power save mode. And, if you are indeed measuring stable output voltage from the regulator, but have dips at the IC power pins, the problem has to be between the regulator and the IC. Let me suggest: the filter is the problem. Possibly too much DC resistance, even! Try connecting the load directly to the regulator output, with a few inches of wire max, and see what happens?

[Sjoertdb]

Thanks for the input! I'll try adding a regulator with a LE33 from ST. This one has a drop of 200 mV.
I'll try to remove all the possible dc resistances. I'll send an update later.

Thanks! Your information is very valuable to me! 

[Sjoertdb]

I've tried tying an LDO to the 3V3 output and the 5V output.
The one on the left is the output of the LE33 when 3V3 is used as input, (probably LDO stops regulating by now)
The one on the right is an LDO added to 5V, now the issue is completely gone.