Switchmode PSU capacitor selection advice needed

[Electr0nicus]

Hi

For a switchmode PSU design I need to define the output capacitor values. As in a switchmode design the main selection criteria of a Output/Input cap is not primarily the capacity, but the maximum ripple current and ESR. If you find a capacitor with appropriate values for ESR and ripple current, the capacity of this cap is mostly sufficient for filtering.
Technology of caps ranges from tantal to ceramic and electrolytic types. For my design i want to use only ceramic and electrolytic.
So I've calculated the ripple current and ESR specs the capacitor has to meet and also found a appropriate cap 560uF with a ESR of 20mOhm and a max. ripple current of 2A (max. Output ripple current = 1.1A).

http://at.farnell.com/panasonic/eeufr1v561/kondensator-radial-35v-560uf/dp/1800667?Ntt=EEUFR1V561.

This will result in approximately 14mVpp ripple for the 5V 3.5A output and 10mVpp ripple for 3.3V 2.5A, because I use 2 of this caps in parallel (simulated with LTSpice). So this output ripple is quite within spec.
Now I looked up the reference design schematic on linear technology's website

http://cds.linear.com/docs/Demo%20Board%20Schematic/1425asch.pdf .
There I discovered that they used a electrolytic capacitor in combination with a few ceramic caps. So I know from a post on this forum that ceramics have huge tolerances if you use them near their rated voltage. Then there are only 20% of the rated capacity left  .

https://www.eevblog.com/forum/projects-designs-and-technical-stuff/esr-of-ceramics-%28mlcc%29/

So they are very bad in that respect but, have really low ESRs.
So do I need additional ceramic capacitors in parallel if a "only electrolytic" solution will just work fine. Are there any advantages/disadvantages in load step response, regulator stability and so on?

Thanks in advance, Gregor

[Kevin.D]

For higher frequencies large electrolytic's  perform poorly because of self inductance etc .
That's why you often see them in combination with small ceramics which will filter  at the higher frequencies .
If you look on the forum theres a really good post  somewhere about it that shows
you how the capacitance of different type of caps effectively varies with frequency.
I think it was  a part of a post  about a review of an lcr meter or something .

[Electr0nicus]

I actually found a very interesting PDF.
http://www.ultracad.com/mentor/esr%20and%20bypass%20caps.pdf

[Electr0nicus]

Hi

If I'm not wrong actually the only thing which spoils the performance of electrolytic caps is the ESL (equivalent series inductance).
To simulate a real capacitor in LTSpice I'll need a ESL value. Unfortunately it isn't listed in the capacitor datasheets, and I've looked at many.
Is there a way to calculate the ESL, or are there standard values which I can use for the simulation. Because it's senseless to simulate
a circuit with ideal components, when the reality looks different.

Cheers Gregor

[Psi]

If I'm not wrong actually the only thing which spoils the performance of electrolytic caps is the ESL (equivalent series inductance).
To simulate a real capacitor in LTSpice I'll need a ESL value. Unfortunately it isn't listed in the capacitor datasheets

If the datasheet doesn't give the ESR value then the capacitor isn't intended for applications where ESR is important.
Look for a capacitor that's labeled as low-esr or intended for switching supplies as they will supply the ESR value.

[vk6zgo]

Hi

If I'm not wrong actually the only thing which spoils the performance of electrolytic caps is the ESL (equivalent series inductance).
To simulate a real capacitor in LTSpice I'll need a ESL value. Unfortunately it isn't listed in the capacitor datasheets, and I've looked at many.
Is there a way to calculate the ESL, or are there standard values which I can use for the simulation. Because it's senseless to simulate
a circuit with ideal components, when the reality looks different.*
Cheers Gregor

*You are learning,grasshopper!

If you can find some values of ESLs quoted on the 'Net by someone who has measured them,you can assume it's similar for your caps,& add the inductance in series with your caps,just like real inductors.

Or you could measure the phase of an AC current through the capacitor w.r.t. a test voltage.
The amount it differs from the pure capacitive case shows how much inductance (& resistance) is present.

I think in most cases,people just make a guess,chuck some ceramics in,and see what results they get,or copy someone else's design!

[Kevin.D]

Here's a thread with some  posts of graphs of different caps at various frequencies look for post's by  "The Electrician".
https://www.eevblog.com/forum/product-reviews-photos-and-discussion/mastech-ms5308-lcr-meter-with-esr-measurement-on-discount-at-the-moment/30/

here's another good thread
https://www.eevblog.com/forum/beginners/bk-precision-lcr-879b-meter-readings-wrong-or-user-error/msg48674/#msg48674

[free_electron]

Serious cap manifacturers like TDK, Murata and others actually have a piece of free software where you select the capacitor and they gove you all that information. I have posted the links on this forum already. Search for it.

[Electr0nicus]

So, I finally downloaded the SEAT design tool from the TDK website and played around with it for some time. But now I have a new question:

How do I select a appropriate capacitor. Should I select one, so that it's resonance frequency matches the frequency of the signal which
should be smoothed out, in my case 350kHz.  The problem is, that beyond that
point the capacitor behaves inductive. The advantage is clearly that you get the lowest impedance at the resonance frequency, more precisely
only the ESR.  Or should I select a cap so that my desired frequency lays on the linear part of the Cs Cp curve?

F.e. a TDK - C4532X7R1C336M 33uF 16V capacitor

[jahonen]

It doesn't matter if the impedance of your capacitor is capacitive or inductive as long it is low enough for the purpose. If you overlap impedance curves various capacitors with same physical size, you'll notice that high frequency behaviour is identical (as long as inductance is identical) but resonance point shifts to lower frequencies when capacitance increases. To put it simply, above self-resonant frequency, it is the inductance which dominates the impedance. Also note that inductance of your PCB vias and other interconnect will have significant impact. To get the impedance from the cap you payed for, you would need a zero external inductance which is of course impossible.

Usually, most troublesome noise generated by the switcher are at much higher frequencies, where components start to resonate, typically between 50-200 MHz. For this reason, you want lowest possible inductance. Putting all this together, you'll want highest capacitance in smallest possible package (lowest inductance). Beware of Y5V, Z5U etc. dielectrics, they have very significant modulation of the capacitance along with the DC bias voltage.

However, don't go by that general fallacy of putting several different-sized ceramic capacitors in parallel (especially smaller ones), it only tends to form parallel resonances which will cause the situation to become worse for wide ranges of frequencies. Perhaps if you carefully measure and test your power system for troublesome frequencies it is ok, but I certainly would not recommend that blindly.

See page 96 onwards: http://koti.mbnet.fi/jahonen/Electronics/Stuff/AE102_SI_EMC_XT.pdf

Regards,
Janne

[Electr0nicus]

It doesn't matter if the impedance of your capacitor is capacitive or inductive as long it is low enough for the purpose. If you overlap impedance curves various capacitors with same physical size, you'll notice that high frequency behaviour is identical (as long as inductance is identical) but resonance point shifts to lower frequencies when capacitance increases. To put it simply, above self-resonant frequency, it is the inductance which dominates the impedance. Also note that inductance of your PCB vias and other interconnect will have significant impact. To get the impedance from the cap you payed for, you would need a zero external inductance which is of course impossible.

Of course you are right. That's why it's not recommended write postings in technical forums when it's 3 o'clock in the morning and you're very tired It's clear that beyond the resonance frequency the reactance of the capacitor is inductive, but the capacitance and inductance stay the same. Also that the inducance of similar sized capacitors is nearly equal. So I will be able to calculate the series inducance with the formula OMEGA = sqrt(1/(L*C)), because OMEGA and C are given. OMEGA is the resonance frequency * 2 *PI and C would be the rated capacity of the cap. So for the above mentioned TDK cap I'll get a inductance of 1.9nH. This values I then can use for simulation.

Usually, most troublesome noise generated by the switcher are at much higher frequencies, where components start to resonate, typically between 50-200 MHz. For this reason, you want lowest possible inductance. Putting all this together, you'll want highest capacitance in smallest possible package (lowest inductance). Beware of Y5V, Z5U etc. dielectrics, they have very significant modulation of the capacitance along with the DC bias voltage.

Yes I'm aware of that. Looking at the fourier analysis of a rectangular waveform there are only odd harmonics present, like f, 3*f, 5*f and so on. I've looked at the DC-Bias voltage vs. capacitance graph of the abovementioned TDK cap, and it says that there is a reduction of 7% of the rated capacity @ 5V and nearly 1.5% @ 3.3V, which is pretty acceptable.

[Electr0nicus]

Hi,
Still got some questions.

First: Is it crucial to select a capacitor, whose resonance frequency is just the same as the switching frequency of the smps,
or in other words: does it have a large effect, when the resonance frequency of the capacitor is not equal to the switching frequency of the psu? I for my self think that it is no problem, unless you accept the larger impedance and thus lower filtering capabilities.

Second: Concerns the dependency of the capacity on the DC Bias voltage.
If I have let's say a 5V output voltage and on that point the capacity is only 75% of the rated, am I right in thinking, that the resonance frequency also shifts upwards, because the inductance stays the same?

And third: If I parallel capacitors of different values,
is it a good idea to select the values, so the bumps of |Z| at the resonance frequencies of the capacitors match with the frequencies of the harmonics, 3*f, 5*f, 7*f and so on? Building a parallel resonant circuit should be near to impossible in that case, or I'm wrong?

Cheers Gregor

[vk6zgo]

I honestly believe you are over-thinking this--It's a switchmode PSU.not a Radiotelescope!

OK,Electrolytics do look inductive at higher frequencies,due to their construction,so it is something to take into account,but basically,it is their capacitance you are interested,as if they are in the output filter circuit of a PSU, a major part of their function is to help with the regulation of the supply.
You could build a supply with silver Mica capacitors with really short leads,& they  would have very low inductance at the switching rate of your supply & above,but your supply wouldn't work!

Electrolytics with paralleled mica or ceramic caps have been used successfully in High frequency Engineering for many years.

 

[jahonen]

Targeting the impedance minimum at the switching frequency is a bit futile attempt, since there are significant amount of energy elsewhere in the spectrum. Same goes for trying to tune the capacitors to the harmonic frequencies. My advice is that do not bother with that. If you still need improvement after putting some capacitors, add something which turns lossy at higher frequencies (like a ferrite in suitable place) and you'll reduce likelihood to develop a severe parasitic resonance considerably.

Yes, your resonance frequency will go up when bias voltage drops the capacitance. I have in fact used that property to estimate the amount of drop of capacitance in the real world.

Also, odd harmonics occur only if you have strictly symmetrical waveform with 50% duty cycle. This is usually not the case so you'll generally get all the harmonics and roll-off depends also on the duty cycle and symmetry. Likewise, mr. Fourier fails to predict the parasitic resonances lurking in your circuit, which often dominate the final cleanliness of your output voltage. It usually looks like this (a 300 kHz SMPS):



Regards,
Janne

[SeanB]

Add ferrite beads on the diode leads, and add some ferrite beads between each capacitor, to make a lossy filter to the high frequency switching noise. The beads are lossy and will reduce noise. 2 stages of LC filtering will reduce the noise considerably, and as a bonus the PSU can use lower value capacitors to improve the transient response of the control loop. Adding too much capacitance can make some SMPS controllers oscillate at a low frequency.

[Electr0nicus]

Tanks jahonen for the spectrum image. At least I now know how the reality looks like  Yes I have to admit that I sometimes look at things in a too complicated manner. Well  I think I will use a electrolytic cap in parallel with a ceramic MLCC cap.
So I will use a Panasonic 220uF 50V electrolytic Cs=220uF Rs=30mOhm Ls=9nH
and a TDK C3225X7R1C226K 22uF 16V ceramic. Cs=22uF Rs=2.09mOhm Ls=1.5nH
If I plot the impedance using LTspice, I get a reasonable curve. See attached screenshot. The impedance is relatively low in the areas where the first few harmonics occur, so they should be damped in some way.
So I will start designing a test PCB now. I will add some ferrite bead footprints, so that I can add them if necessary.
 

[vk6zgo]

I don't really think there is a lot of point in using a quite large ceramic  in parallel with your  Electrolytic cap.
The more common usage is of a much smaller ceramic/mica/polyester/polycarbonate,which will hardly contribute at all  to your storage capacitance,but will have a low impedance at those higher frequencies where the inductance of the electrolytic becomes substantial.

Think on this:

Thousands of Engineers,Technicians,Ham Radio Enthusiasts,etc,have designed tens of thousands of switch mode PSUs over the last 50 years or thereabouts,so it isn't Brain Surgery.
Well before switch modes became common,the technique of paralleling Electrolytics with much smaller,less lossy capacitors was used extensively in High Power RF Amplifiers.

Using the series resonance of a capacitor is a fairly esoteric technique,which is sometimes used at VHF/UHF to resonate mica or ceramic capacitors with their lead inductance.
SMD caps have no leads,so this is seldom necessary these days.

[SeanB]

The only application I have seen of using the resonance as a technique is in GSM mobiles, where they are a specified unit that has a controlled resonance that is designed to fall on the different GSM bands.