Can I measure if I need an input capacitor to my switching power supply?

[jnz]

I have a 450Khz PWM smps regulator that suggests a 47uF electrolytic. My circuit only draws 250mA nominal and typically fairly steady minus CAN bus transfers, the power supply is rated to 1A.  Automotive battery source, so I would need a 35-50V rated part. I can't fit it!

I basically can't fit any electrolytic at all in the enclosure. I could struggle all day long to find a 5mm diameter part ... or... I could try and get-by without it. Can I measure this? The output side is only 5V so I can fit a ton of different parts there. The only call for 2x 22uF ceramics, but I'm not sure it would "hurt" to have a niobium or tantalum, or etc, right?

So...

1. How can I measure if I need the input cap? I have the SMPS dev board hooked to a power supply now, they use 2x 22uF ceramics on the 5V output. The RMS on the output a 50uS/div and 12V input is only 850uV with no load. I think I would need some realistic load on this. But, either way, is RMS at high sampling rate a good way to test this? I can hook it up to a car and pull the input cap off, but I don't want to go through all that if my plan of measurement is way wrong.

2. They call for a 22uF to 47uF input inductor. "Depending on application" Can I offset some of me not having an electrolytic input cap by increasing the uH rating? That's KINDA like having a capacitor to fill in gaps right? I know for certain my input at 10V min will always be more than I need for 5V out, but I think the issue is rather that the SMPS might not "keep up" with lots of noise at 450Khz PWM?

3. For the application I have described... How equivalent is a ceramic uF to an electrolytic uF? Like if the application suggests an electrolytic input cap at 22uF and I use a ceramic, am I just not going to be able to fill in those lower frequency gaps, or will the fill in, just not as well?

I tried to research this elsewhere and didn't get very far. Just looking for opinions, thanks!

[Refrigerator]

OK let's get started.

I have a 450Khz PWM smps regulator that suggests a 47uF electrolytic. My circuit only draws 250mA nominal and typically fairly steady minus CAN bus transfers, the power supply is rated to 1A.  Automotive battery source, so I would need a 35-50V rated part. I can't fit it!

At 450kHz you don't need a huge capacitor, 47µ sounds about correct. "Automotive battery" means basically nothing, eg. is it 6V, 12V, 24v?? For a 12V a 16V cap would just about be good enough, 25V if you want to be safe. And at that voltage 47µ electrolytic caps are still quite tiny, should fit anywhere imo.

I basically can't fit any electrolytic at all in the enclosure. I could struggle all day long to find a 5mm diameter part ... or... I could try and get-by without it. Can I measure this? The output side is only 5V so I can fit a ton of different parts there. The only call for 2x 22uF ceramics, but I'm not sure it would "hurt" to have a niobium or tantalum, or etc, right?

So...

A 47µ tantalum would also work.

1. How can I measure if I need the input cap? I have the SMPS dev board hooked to a power supply now, they use 2x 22uF ceramics on the 5V output. The RMS on the output a 50uS/div and 12V input is only 850uV with no load. I think I would need some realistic load on this. But, either way, is RMS at high sampling rate a good way to test this? I can hook it up to a car and pull the input cap off, but I don't want to go through all that if my plan of measurement is way wrong.

No load ripple means nothing unless your PSU is faulty. A 6V tungsten light bulb would be ok as a semi-universal load for such circuits, easy to find too. I use resistors, and for a 5V circuit i would maybe hook up my phone to charge as well, two birds with one stone kind of thing. You could also use a constant current load, they're easy to make, the internet is full of tutorials.

2. They call for a 22uF to 47uF input inductor^[1]. "Depending on application" Can I offset some of me not having an electrolytic input cap by increasing the uH rating? That's KINDA like having a capacitor to fill in gaps right?^[2] I know for certain my input at 10V min will always be more than I need for 5V out, but I think the issue is rather that the SMPS might not "keep up" with lots of noise at 450Khz PWM^[3]?

1. Inductors are not measured in Farads.
2. Not quite, inductors resist change in current, capacitors resist change in voltage.
3. How do you know there's "lots" of noise?

3. For the application I have described... How equivalent is a ceramic uF to an electrolytic uF? Like if the application suggests an electrolytic input cap at 22uF and I use a ceramic, am I just not going to be able to fill in those lower frequency gaps, or will the fill in, just not as well?

I tried to research this elsewhere and didn't get very far. Just looking for opinions, thanks!

1µF = 1µF no matter the type of capacitor. Go read up on capacitor series resistance, series inductance and all the other fun stuff that makes them different.

Basically for an SMPS you want a capacitor as close as possible, large series inductance is a bad thing that will cause transient noise, especially in a switching circuit. This transient noise might be bad enough to damage your SMPS. SMPS circuits like low ESR capacitors. Tantalums and caramics are low ESR by nature, while low ESR electrolytics tend to have gold colored marking on them. Electrolytics are also cheaper and more forgiving. Any capacitor is better than none in this case.

[jnz]

12V (14V) battery and in vehicle. So that's how I know there will be noise. I should have written automotive circuit, not automotive battery, although both are true.

IDK if a tanatlum would work for the input cap. First off, almost all of the 35V tanalum parts would be very expensive, I need ~$1.00 at volume, which there are some, but the better parts seem to be $5.00 at volume. Second, tantalum as the input cap in a car scares the crap out of me. It's not impossible at all to get a spike over 35V in a vehicle. I had a product that "field crystallized" a tant under bad usage scenarios.

Typo. Obviously I meant 22 to 47 uH.

When you say we want the capacitor as close as possible, which one are you talking about? The input? Output? Both?

[srb1954]

12V (14V) battery and in vehicle. So that's how I know there will be noise. I should have written automotive circuit, not automotive battery, although both are true.

IDK if a tanatlum would work for the input cap. First off, almost all of the 35V tanalum parts would be very expensive, I need ~$1.00 at volume, which there are some, but the better parts seem to be $5.00 at volume. Second, tantalum as the input cap in a car scares the crap out of me. It's not impossible at all to get a spike over 35V in a vehicle. I had a product that "field crystallized" a tant under bad usage scenarios.

Typo. Obviously I meant 22 to 47 uH.

When you say we want the capacitor as close as possible, which one are you talking about? The input? Output? Both?

I wouldn't use a 35V tantalum on a power input fed from the automotive supply circuit unless you have a really, really effective overvoltage clamp on the incoming supply. Otherwise, you are just installing a tantalum capacitor based pyrotechnic device.

A low ESR aluminium electrolytic is most appropriate here, preferably rated at 63V to handle expected input voltage transients. With a 450kHz switcher a relatively low input capacitance may be sufficient as long as the capacitor has low ESR and low ESL. A parallel ceramic capacitor, also rated >63V,  will help contain the higher frequency noise that the electrolytic cap doesn't catch.

[Refrigerator]

12V (14V) battery and in vehicle. So that's how I know there will be noise. I should have written automotive circuit, not automotive battery, although both are true.

IDK if a tanatlum would work for the input cap. First off, almost all of the 35V tanalum parts would be very expensive, I need ~$1.00 at volume, which there are some, but the better parts seem to be $5.00 at volume. Second, tantalum as the input cap in a car scares the crap out of me. It's not impossible at all to get a spike over 35V in a vehicle. I had a product that "field crystallized" a tant under bad usage scenarios.

Typo. Obviously I meant 22 to 47 uH.

When you say we want the capacitor as close as possible, which one are you talking about? The input? Output? Both?

Both, because it cuts down on noise and generally improves the performance of your SMPS.

[Siwastaja]

Sorry to say this but you'd have to ignore all the unhelpful and wrong advice here so far. For example, tantalum is definitely no-go (this should be obvious), and yes, you are right that MLCCs lose capacitance over DC bias (this is quite well-known as well).

It seems you are not talking about the input capacitor of the converter because it appears you are using an existing converter module of some sort; they definitely have input capacitors already (for example, a buck converter can't work without).

So are we looking at adding additional input capacitor in parallel with whatever exists. In order to have a meaningful discussion, we would need to know what the original converter is; part number, schematics of its internal construction, etc.

It's well possible it won't need anything additionally; or it may definitely need some. It has to have low-inductance, low-esr bypassing properly laid out already, this most likely means MLCCs on the board. Adding a larger bulk cap with larger ESR, such as an aluminium electrolytic, would be a typical choice to dampen resonances with the input inductance.

Regarding automotive, the big question is are you just using a car battery, or is this thing going to live in a car with alternator running?

[jnz]

Sorry to say this but you'd have to ignore all the unhelpful and wrong advice here so far. For example, tantalum is definitely no-go (this should be obvious), and yes, you are right that MLCCs lose capacitance over DC bias (this is quite well-known as well).

It seems you are not talking about the input capacitor of the converter because it appears you are using an existing converter module of some sort; they definitely have input capacitors already (for example, a buck converter can't work without).

So are we looking at adding additional input capacitor in parallel with whatever exists. In order to have a meaningful discussion, we would need to know what the original converter is; part number, schematics of its internal construction, etc.

It's well possible it won't need anything additionally; or it may definitely need some. It has to have low-inductance, low-esr bypassing properly laid out already, this most likely means MLCCs on the board. Adding a larger bulk cap with larger ESR, such as an aluminium electrolytic, would be a typical choice to dampen resonances with the input inductance.

Regarding automotive, the big question is are you just using a car battery, or is this thing going to live in a car with alternator running?

Yea, I agree with you and srb1954 above, a tantalum on an automotive circuit, yes, hooked to the running alternator is a bad idea. I'm not sure I could protect for that enough. I think they recommend you need 400mS of 40V+ but that means a 50V tant which is $$, or try and limit down under 35V which might be an option I guess.

I do have a regulator device, it's an Infineon DCDC system basis chip. So, it's not a "module" per se, but it's not a discrete circuit either. All models show their example circuits with C1 right after a protection diode but before the input inductor and say "C1 47 μF ± 20% Electrolytic, Buffering capacitor to cut off battery spikes, depending on the application. The voltage rating depends on the application." so... I don't think this is as required parts as the others "C2 100 nF ± 20%, 50V Ceramic, Input filter battery capacitor for optimum EMC behavior" or "C3 1..10 μF ± 20%, 50 VCeramic, Input Buck capacitor. Low ESR."... To me... This suggests C1 is recommended for filling in caps on the input, but not as required as the obviously not-optional description of C3. They aren't really clear on the intention.

The cap they use on their dev board is a 47uF .30ohm ESR, so yes, pretty low. They also use an input inductor after the cap.

This is the part: https://www.infineon.com/dgdl/Infineon-TLE9271QX-DataSheet-v01_50-EN.pdf?fileId=5546d46268554f4a016885465c676856  PAGE 118 for diagram similar to how I'm using. PAGE 119for the recommended components list.

I also don't get how "required" the filter / input inductor is. At least the cap I get, to buffer out battery gaps, I just can't tell if I need it.

[Siwastaja]

Oh, that's one strange datasheet; you'll get totally lost into the complexity of that chip, forgetting you are also designing a DC/DC converter! If you look at the datasheets of DC/DC controller chips, without all that CAN bus system on chip stuff, they do have clear schematics, layout examples, and what's best, sections discussing and calculating component values such as input caps, output caps, and inductors. Sometimes these application notes are very good, sometimes unusable, but you'll at least have an idea what you are doing.

I'd suggest you Google at some appnotes, for example using keywords like "buck converter input capacitor".

In a buck, the input cap is one of the most crucial components, much more critical than the output cap. Layout is equally important. This is because the switches pull hard, square-wavish current spikes from said capacitor. Theoretically, you'd aim for zero ESR and zero inductance; in real life, a combination of very low-esr, very low-Z MLCC, and a damping higher-ESR elcap is a good idea; the damping is especially important if there is any chance of this being hot-plugged.

For capacitor value, you'd want to choose maximum input voltage ripple seen at the cap (say, 0.1V for example), and solve C from I = C*dV/dt, knowing the switching period and switch current. This is a rough assumption ignoring the circuit feeding the C, but nevertheless works for the first-order design; you can always choose to double that capacitance, for example.

Using different sizes of MLCC in parallel isn't a good idea because they may resonate with the parasitic inductances present; but using a large size elcap isn't a problem because it has a lot of ESR.

With elcap in automotive environment, use known-good brands, with high temperature ratings, long operating hour ratings, and derate ripple current spec, for long life.

[jnz]

Oh, that's one strange datasheet; you'll get totally lost into the complexity of that chip, forgetting you are also designing a DC/DC converter! If you look at the datasheets of DC/DC controller chips, without all that CAN bus system on chip stuff, they do have clear schematics, layout examples, and what's best, sections discussing and calculating component values such as input caps, output caps, and inductors. Sometimes these application notes are very good, sometimes unusable, but you'll at least have an idea what you are doing.

I'd suggest you Google at some appnotes, for example using keywords like "buck converter input capacitor".

In a buck, the input cap is one of the most crucial components, much more critical than the output cap. Layout is equally important. This is because the switches pull hard, square-wavish current spikes from said capacitor. Theoretically, you'd aim for zero ESR and zero inductance; in real life, a combination of very low-esr, very low-Z MLCC, and a damping higher-ESR elcap is a good idea; the damping is especially important if there is any chance of this being hot-plugged.

For capacitor value, you'd want to choose maximum input voltage ripple seen at the cap (say, 0.1V for example), and solve C from I = C*dV/dt, knowing the switching period and switch current. This is a rough assumption ignoring the circuit feeding the C, but nevertheless works for the first-order design; you can always choose to double that capacitance, for example.

Using different sizes of MLCC in parallel isn't a good idea because they may resonate with the parasitic inductances present; but using a large size elcap isn't a problem because it has a lot of ESR.

With elcap in automotive environment, use known-good brands, with high temperature ratings, long operating hour ratings, and derate ripple current spec, for long life.

IT IS a complex datasheet. Not a lot of time to explain capacitor options when they have 40 or 50 SPI registers to also discuss.

I'll look for some appnotes, but I'm not sure they will be directly comparable to this part. Other chips we were looking at recommended no input capacitor at all besides a 10uF MLCC specifically called out. Others called the electrolytic as specifically optional.

I'll go measure a vehicles and see if I can a feel for ripple. I wonder if I'll have to put the scope in the car and use the stock inverter to power the scope while driving around. This is probably something I should have done years ago anyhow.

I think the lede got buried here. It doesn't really matter which brand of EC I'm looking at because I can't fit any of them. I can fit some MLCCs, it's roughly 4mm max height that is the issue.

So that comes back to the base issue. Other than not being AS GOOD at filling in the lower frequency gaps, how comparable is a similar uF MLCC to an EC on the input here?

AND what is the appropriate way to test this? I can hook a scope up to look at input and output at the same time look at the difference (hmm, no diff probes but that's fine). I can pop the EC off, hook up to a vehicle and look at the ripple, dips, etc in the output with some expected restive load. Is that appropriate?

[Siwastaja]

In absence of more sophisticated test&measurement gear, you can always measure peak-to-peak ripple voltage at the power source end of the cable feeding your converter, for differential-mode noise. This gives you some idea how well the capacitor is doing.

Most switching converter, and definitely buck converter, absolutely require the input capacitor. The idea of the cap is to bypass the power source inductance, so that the high-current, high-frequency switching current only goes through the small loop formed by the cap and the switches. For this, low inductance is #1; this includes layout. A low-ESL cap does nothing if it's laid out too far. MLCCs basically only have inductance as defined by the package size. Using smaller packages in parallel helps, you can then spray them around the loop area; or think about within the loop. I try to use 0805 as maximum size, also reducing risk of cracking.

High-ESR capacitance, where C_hi_esr >> C_low_esr, say at least 2-3 times as much, is highly recommended; otherwise, you'll likely see parasitic oscillations, high inrush peak voltage during hotplugging, possibly enough to kill the switches. Say if you have 10uF of low-ESR (say 10mOhm) MLCC, parallel that with bog standard 47uF elcap (say around 1ohm) and you are done. See if the elcap heats up; that would be a sign that you don't have enough of the MLCC, and significant part of switching current is going through the elcap.

You don't have to use elcap for damping, though; MLCC + explicit series resistor works, too. A TVS can also be used to clip the inrush voltage peak, but doesn't help if you have smaller-amplitude ringing during operation; having a stable power filtration network is more preferable.

About the DC bias effect, note that when a manufacturer recommends a 10uF MLCC, they already take this into account. What they do not tell you, is the actual capacitance they require. It may be 5uF. Avoid the worst types like Y5V, and avoid the combination of highest ratings in the smallest possible packages, they may go as low as 20% of the rated C at 80% rated voltage; 10uF nominal could be only 2uF actual! At 12V DC bias, a 2.2uF in 0805 package or 4.7uF in 1206 package is usually OK rated X7R, or maybe X5R if you only run it at room temperatures.

[jnz]

In absence of more sophisticated test&measurement gear, you can always measure peak-to-peak ripple voltage at the power source end of the cable feeding your converter, for differential-mode noise. This gives you some idea how well the capacitor is doing.

Most switching converter, and definitely buck converter, absolutely require the input capacitor. The idea of the cap is to bypass the power source inductance, so that the high-current, high-frequency switching current only goes through the small loop formed by the cap and the switches. For this, low inductance is #1; this includes layout. A low-ESL cap does nothing if it's laid out too far. MLCCs basically only have inductance as defined by the package size. Using smaller packages in parallel helps, you can then spray them around the loop area; or think about within the loop. I try to use 0805 as maximum size, also reducing risk of cracking.

High-ESR capacitance, where C_hi_esr >> C_low_esr, say at least 2-3 times as much, is highly recommended; otherwise, you'll likely see parasitic oscillations, high inrush peak voltage during hotplugging, possibly enough to kill the switches. Say if you have 10uF of low-ESR (say 10mOhm) MLCC, parallel that with bog standard 47uF elcap (say around 1ohm) and you are done. See if the elcap heats up; that would be a sign that you don't have enough of the MLCC, and significant part of switching current is going through the elcap.

You don't have to use elcap for damping, though; MLCC + explicit series resistor works, too. A TVS can also be used to clip the inrush voltage peak, but doesn't help if you have smaller-amplitude ringing during operation; having a stable power filtration network is more preferable.

About the DC bias effect, note that when a manufacturer recommends a 10uF MLCC, they already take this into account. What they do not tell you, is the actual capacitance they require. It may be 5uF. Avoid the worst types like Y5V, and avoid the combination of highest ratings in the smallest possible packages, they may go as low as 20% of the rated C at 80% rated voltage; 10uF nominal could be only 2uF actual! At 12V DC bias, a 2.2uF in 0805 package or 4.7uF in 1206 package is usually OK rated X7R, or maybe X5R if you only run it at room temperatures.

Well....

The input capacitor is 1-10uF ceramic. I BELIEVE - They are recommending/suggesting/stating that an electrolytic is a good idea to fill in input dips. It DOES NOT appear to me to be required. The way you keep writing, it seems like you are not considering this, and suggesting it's a mandatory component of the design.

I hooked up the scope to a dev board, and tested with an without the input cap. Not much of a difference at all, esp in lower frequency waves. When looking at the 20nS scale, it's about 9mV output frequency standard deviation with it, and about 13mV without it. I'm just not seeing a big first or second order wave anywhere. The noise definitely increased on the output without the buffering cap, but damn near unnoticeable in a running vehicle. What's more, big changes like idle to 5000 rpm resulted in no major difference in noise, although a tiny difference in overall voltage. Still, we're talking mostly 5.000 to 4.995V

So... I'm grateful for your contributions here. But I don't think you are and I on on the same page, that it seems like the buffering capacitor is not actually part of the circuit, but "a good idea" suggested part.

OR - I am wrong, and this is a super required part, and by removing it and not seeing a difference, I just wasn't testing correctly?

SEE PHOTO. The yellow is 12V Battery of a running vehicle. The lower reference and blue active channel are with and without electrolytic cap. It's a difference, just not one I'm sure I care much about. The load on the 5V was static, so maybe when I can test a digital circuit I might change my mind. That giant spike was throwing me, which is why I took the pic... It's the vehicle's inverter. I needed it to power the scope. It's a 60Hz spike all over the vehicle and I checked downstream of a factory ECU and saw a pretty darn similar spike. Apparently they just don't see to care how noisy it is on their 12V line. Who knew!?

[Siwastaja]

Yes, you don't necessarily need a larger "buffer" capacitor. Some decades ago, when largest MLCCs on the market were some hundeds of nF, maybe 1uF, and f_sw was in range of say 100kHz, capacitance requirement for low enough ripple often necessitated the combination of elcap and MLCC. Today, you get enough capacitance with MLCCs alone. With todays switching frequencies and edge rates, the amount of C is less important, and minimized amount of L (so package size and PCB layout) is paramount.

The main reason for the elcap now is for damping the LC resonances. In this case, high ESR is necessary (note that in this context, a "low esr electrolytic" is high enough ESR; but beware of polymer electrolytics, they may actually be too low on ESR!), and C must be significantly larger than the total C of the low-ESR caps. Given enough MLCC, taking the majority of switching current, the elcap is stressed less, and easier to design in.

Measure the voltage while hotplugging a live DC input with worst case wiring length, possibly with wiring running a larger loop instead of neatly side-by-side. This shows if you have a stability problem caused by low ESR MLCC solution. If yes, then adding an electrolytic is a good idea.

See https://www.analog.com/media/en/technical-documentation/application-notes/an88f.pdf, if in hurry, look at fig.4 & table 3 for results.

I almost always add the electrolytics for damping because I don't like to do assumptions about input wiring length and shape which I would then need to verify and enforce.

[jnz]

... Hmmm.... As to wire length. This test I was running was with an extra 18 FEET of wire (18 AWG) than the real circuit will use. I had it already spooled out, so I just threw it all in the vehicle. Probably should have considered that, but that just means this is an extra double worst case I've already tested, yea?

[opampsmoker]

You can have any input capacitor you want...as long as..

1...It doesn't resonate with the input inductor at the switching frequency.
2...It's big enough that the ripple volts at the input is less than vin/10
3.....It doesnt mean your transient response is too slow.
4....Your input filter should have an impedance of at least 10 times less than the SMPS dynamic input impedance up to the crossover frequency.....that's the inpedance of the input filter as seen looking back from the smps input...see the "dcm application note" by vicor which makes this easy to understand
The SMPS dynamic Zin is  (Vin^2)/Pin.
5......You dont draw sharp current waveforms over the supply cables or PCB tracks.......eg for a buck you shouldnt see a perfect sharp trapezoid current...you should put enough caps to filter that....preferably enough so that the input current looks like a sine wave (ish), or flat constant DC if you can afford a lot of caps.
6...you pass EMC  test.
7....Beware vin overvoltage due to input filter ringing when vin is applied as a step.....put an inrush limiter or similar to avoid this..or just use caps with enough voltage rating to not be affected.
8...You dont over ripple current the capacitor.

BTW, all SMPS need an input capacitor of some description.......its just that in say a battery fed smps, the battery can act as the input cap sometimes.

[jnz]

Yea... OK.

I'm just going to have to test it when I get the demo board soldered up. I'm pretty sure I'll have another question on that later. Off the top of my head I think the waveform generator with a noise patter will be a good test when I can place different caps on the input as I test it.