My PSU design ripple and noise with picture measurements

[VEGETA]

Hello

I have done extensive tests as required and here are the pictures: https://slow.pics/c/tXFqHF3r

please notice picture's name under it, which signifies measurement parameters.

I have added some capacitors directly at 3.3v rail output pin to gnd pin as indicated by pictures, values like 4.7n, 47n, 2x47n, 22p, and 10u elec, as well as 33u elec. It doesn't seem to have done a noticeable effect.

what do you think?

[T3sl4co1l]

so I just hook it up as usual then zoom in to 200us or so range?

Why guessing?  I gave a number earlier?

~200us/div may show insight into control loop compensation, or the like.  Not that you're doing a proper load step test, it's just whatever the console draws, which... who knows, it could be doing anything?  Wouldn't it be nice to know what it's doing?

it is quite difficult to measure the current, i mean i can not break the circuit to insert the current measurement. and since there is nothing to see for current then maybe no need to do it.

I didn't say it was going to be easy to measure -- or necessarily that you have to measure it at all, but if you're looking for cause and effect, y'know... wouldn't that be a good thing to know?  Worth cutting a wire and inserting a probe, current shunt, sensor, whatever?  Or testing with an independent load of known behavior?

looking at the images above, how do you compare between the 3 PSUs?

Well, on what basis should they be compared?  Evidently they all work; is that sufficient?  What things would your potential customers want to see before buying?  What things can you test?

If I were considering buying such a thing, I would want to know that it's followed good practices.  Now, I don't expect to see a product description claim that; it should be safe to assume.  Customers have that expectation, and they're relying on your good will that it does what it says, and doesn't blow up their hardware or whatever.  It's a lot harder to test for good will, or not blowing up, but you can invest in preserving your reputation by doing electrical tests that exercise the extremes of electrical and other conditions, which should have a strong correlation with also not blowing up.

For my part, I would do at least:
- Supply voltage range test: make sure it meets or exceeds ratings, over the full input range;
- Protection, if applicable: for example if the device has reverse polarity protection, check that it works down to the rated reversal.  Similarly with transient or overvoltage protection, to the extent I need to/care to/can test it.
- Load current range test: meets or exceeds ratings, while at supply min/nominal/max; characterize the current limit / foldback / hiccup / etc. behavior;
- Load step test: output ripple voltage is within tolerance for typical load conditions e.g. 50% base, step to 100%, step back to 50% of rated output;
- EMI/RFI emissions: set up the device in a testing jig, to measure RF emissions from input and output ports.  This is particularly important for things connected to long wires/cables, AC mains, or low-noise systems.  If it's just some internal part, maybe it doesn't matter; this isn't a required step for any particular module or subcircuit (like a DC-DC converter, say), but in that case, the things that do reach those paths (long cables etc.) should have their own filtering to deal with it.  Or it can be dealt with in a case-by-case basis, perhaps the module is too noisy for the filtering in one particular system and does need to be improved, etc.

I happen to have test jigs (or know how to make my own if needed) for that last one, so it's feasible for me; conversely, I don't happen to have the equipment (signal generators, RF amplifiers) to do RF immunity testing, or transients (EFT, surge, etc.).  So for those, I follow good design practice, and hope that it works out.  (Which, I have the benefit of calibrating that experience by testing commercial products in the lab; that would be hard for you to do on this project, unfortunately.)

The next best thing, that you can do presently, for that item, is at least just looking at it on the oscilloscope.  It has the frequency response required (well, some of it anyway).  It doesn't read signals the way an EMI receiver does, but still, it's something.  Even without a test jig, the fact that when you're clipped onto it, you're seeing tiny (~us) squiggles, is proof that something's going on in there.  How many mV tall are those squiggles?  Would it be enough to interfere with a nearby radio?  (Mind, you can only test for radiointerference when you have A. a radio that operates at some frequency, and B. a radio station at that same frequency, to potentially be interfered with.  Or at best you pick up the pilot tone or buzz of the SMPS, on a dead channel, not needing B.  Trouble is, maybe you can pick up the AM and FM commercial broadcast bands; maybe you even have a SW receiver too.  But beyond that?  I mean, it's still something, but it's very spotty coverage, not representative of the full 1 to 200+ MHz range we expect an SMPS to produce.)

And no matter how hard you squint at it, you're not going to tell what's going on at those frequencies, looking at a 50ms/div trace.

Tim

[BrianHG]

Those pulses are approximately 20ns wide, or in the 50MHz region.  The right series ferrite/rf choke would prevent the propagation of that switcher EMI interference as we see the GND input trace is clean.

[VEGETA]

well thanks for your feedback, here is what I think I should do:

label it as "high quality low noise power supply replacement for dreamcast". it is of high quality for the job as it uses good parts and good design practice, went on many revisions..etc and low noise since it is noticeably less noisy than the best alternative (about x2 times) and a better than original PSU as well. I also tested it for long play hours and it went fine with very clean picture which is the ultimate purpose. not to mention that it does have a lot lower temperature than stock PSU, and it has 12v rail properly loaded in case of removing gd-drive.

as of protection, this is a duty for the used 12v wall adapter not this device. this is the case for all designs as well. also short circuit protection the same. I could design a "pro" version of it later on which takes ac directly using recom or meanwell 12v power modules but this will be very expensive and not the scope of this board. such power modules have everything inside and all protections.

I am interested into the potential issue inside the switchings... aren't they just the switcher switching itself? and since this was tested to work properly without problem we can assume it is not going to be one.

I maybe can add one ferrite bead right before the output pin which is the same type used here... but isn't adding 10nf caps better? like adding them before and after the inductor, plus the ferrite bead, and at the load pin itself?

I am open to do any change which is easy to implement in the upcoming 20 boards production batch, which won't do any problem... meaning if it didn't enhance the system, it won't break it. i believe adding ferrite bead and 10nf caps as mentioned won't make the circuit broken or unstable if it didn't enhance it.

the project should be easy and specs are simple, it works and produces low noise and low ripple when using normal 12v power adapter. that is it for me... so that people can buy it knowing it is the best choice as replacement for the original one.

[VEGETA]

hello,

kindly check the attached picture, it has small changes which can be done without fearing of circuit getting broken.

it is just replacing 47uf input filtering for each switcher with 1000uf, plus adding small 10nF in 3 places for extra filtering for higher speed ripple if necessary.

I can relocate stuff to fit these easily, on expense of removing some silkscreen.

i figured that ripple is coming from input side rather than the switchers themselves since they are about 2mhz while ripple is slower. so, having a significantly larger input cap (while keeping some 22uF ceramic input caps) should enhance the signal.

right now i can modify the circuit i have to add these and test them next 1-2 days since i have a lot of 1000uf smd caps, it doesn't have a footprint for sure but can be installed using very short wires and so on.

i just like to hear your opinion.

EDIT: hmm i could replace 4.7uH inductor which is the input to the 2 switchers with 10uH one (not shown in this pic), also the other 4.7uH which is the filter of 12v output rail. i found one with same footprint. however, switchers inductors should still be 4.7uH

[VEGETA]

just a side note, i've found some small footprint 1uH inductors with high enough current.

Would it benefit if I put one of 1uH L on the output of the rail, just before one 22uF ceramic and 47uF elec. cap? plus one at the input of each switcher to act as a small pi filter with either the currently installed 47uF or the suggested 1000uF?

also, adding either 4.7uH or 10uH inductor right after the CMC and before the global 1000uF elec. cap?

oh fk! i am fking redesigning it again despite it worked fine and delivered better results than the others!!! am I ill?

[BrianHG]

Do not change you output circuit.  You board is working fine.  It is the 25-50MHz band switching pulses you want to filter out.
For this, you do not want to add inductors to the middle of your circuit or even add caps.  What you want is a series ferrite bead choke right at the output of your PCB before it feeds the Dreamcast.  You want a ferrite bead with as low as possible impedance to make sue you do not loose any voltage or regulation at the output, plus, a ferrite which has a significant -db signal choking capability between 25-50MHz.  This one component, I'm guessing 2 of them on the +3.3 & +5v out will prevent the migration of those tiny pulses from your PCB to the Dreamcast.

The source 12v power supply would not be the generator of these spikes as they are too high in frequency/narrow to make it down the long power cable in tact as we see in your scope shot.

[VEGETA]

So not even an extra 22uF ceramic cap at the output?

The ferrite beads I am using is Würth Elektronik 7427922808, it is already there and inside the feedback loop itself (the feature of the switching IC). how does it compare here?

from its datasheet I see it is about 8-10 ohms @ 100 MHz, but only 5 ohms at 50 mhz. it was recommended by the switcher datasheet itself.

the other recommended ferrites are: BLE18PS080SN1 and 74279221100.

which parameter in datasheet should I look for in ferrite beads to know the suitable part for 20-50 mhz attenuation? and what about my current one's capabilities?

EDIT:

I searched and found these alternatives:

Z0603C280APWST -> very cheap (30% of the one used currently). about 28 ohms @ 100 MHz, estimated of 21~26 ohms from 20~50 Mhz.

BLM18KG300TH1D -> very cheap as well, labelled as 30 ohms @ 100 Mhz. website shows 15-22 ohms from 20-50 mhz.

FBMJ2125HS420-T -> probably cheapest. 42 ohms @ 100mhz. 32~38 ohms at 20-50mhz. is this the best out of these suggestions? however this is 0805 which requires some modifications to fit it... original one was 0603

if these are accepted, can they replace the one already used inside switching regulator's loop? this will enhance the price and looks like it will also enhance performance.

[T3sl4co1l]

None of these specify DC bias, so it's not obvious if they would be suitable.

At such frequencies, layout is critical.  It's very difficult to say offhand how much is present by inspection alone.  Why do you refuse to measure it?

Tim

[VEGETA]

None of these specify DC bias, so it's not obvious if they would be suitable.

At such frequencies, layout is critical.  It's very difficult to say offhand how much is present by inspection alone.  Why do you refuse to measure it?

Tim

can you show me where is the dc bias in another one if you know?

how can i measure it? didn't i take all measurements?

[T3sl4co1l]

Apologies, I had checked your last link after your last reply after it but it seems either it wasn't updated by then, or my cache didn't update, in any case I only saw the 50ms/div measurements.

So, the thing about the ~10ns pulses.  Do they show similar magnitude when probing ground to ground?  Any ground, any connector?  If so, it's more common mode than the differential mode you think it is.  For which, differential filtering will accomplish nothing, and the question of ferrite bead is moot.

At a guess, the input won't have much of this noise (at least when probed by itself), due to the CMC.  The question is then, which connectors/grounds is the noise between, and should that be solved by cutting ground, rerouting traces, adding CMCs, etc.?

Or it may still be differential, suggesting fairly poor performance of the inductors I think, as the layout isn't that bad.  Maybe you just need better inductors, or another LC stage with a smaller value like 0.47uH then some 0.1uF's, say.


Bias curves are shown in the datasheet, or mfg supplementary (characteristic) curves, if available.  Most do not provide them.  Laird is the most reliable, having bias curves on almost all beads in their catalog.  Wurth sometimes does, as was in this case:
https://www.mouser.com/datasheet/2/445/74279221100-1720641.pdf

Tim

[VEGETA]

so I should go and do the same probing but try to probe another ground points? like the negative of ch2 is on output ground pin while positive of ch2 is on other ground pins of the circuit?

you can see all pictures here: https://slow.pics/c/tXFqHF3r > there is a 500ns/div picture and others, please re-check. name of each picture is under it.

inductors are the maximum size for the switchers, 4.7uH. suggested inductors in datasheet are very very expensive.

I can add an extra stage as suggested earlier, will be 1uH inductor + 22uF + the 47uF elec cap since it should be the final one before the connector. or add another ferrite bead just before the connector. if 22uF is not suitable, I have 1uF ceramic used elsewhere, so I can use it here. such components are very cheap and can be added easily. but why very small inductor and cap size for extra LC stage?

I still want to think about making input filtering to the switchers a bit bigger... or add an inductor right after the main CMC, or making main inductor filter for the switchers input 10uH instead of 4.7uH...etc. what do you think? a better way is to remove 2 of 22uF ceramics and replace them with 47uF elec. this is half the price and better performance since input voltage is 12v and the 22uF 1206 ceramics will surely suffer from dc bias. then add another 47uF to be the total input filtering per 1 switcher is 3x 47uF elec + one 22uF ceramic + 10nF (bypass at pin). instead of 3x 22uF ceramics + 1x 47uF elec. + 10nF. how about that

we already used a CMC at the input and it did great to suppress much of the earlier noise and it helped eliminate 50hz stuff. adding another one will be expensive unlike the extra LC stage.

so maybe it can be like this: switcher -> 4.7uH (main L) -> 4x 22uF -> main ferrite bead -> 3x 22uF ->1uH (extra L) -> 1uF -> 47uF elec. -> 10nF (already exists). with an optional ferrite after this.

still, what do you think about the suggested ferrite replacements? they are cheaper and still have more Z at frequencies.

[VEGETA]

the funny thing is that TPS62913 is out of stock and will be in stock in September!!

[VEGETA]

I would like to point you to this image: https://i.slow.pics/31fiYciR.png (taken from all photos here: https://slow.pics/c/tXFqHF3r)

it is 500nS time base, and shows about 2 MHz pulses of about 17 mV p-p. this is clearly the switcher's switching frequency which is about 2.2Mhz.

hmm can we choose a ferrite bead which has a lot of impedance at 2 mhz to solve this?

this one (https://www.digikey.com/en/products/detail/w%C3%BCrth-elektronik/78279221281/14669530) seems nice with lots of impedance but the dc bias makes it about 1 ohms @ 2 mhz frequency when current is about 3 amps. so it becomes useless?

[BrianHG]

I would like to point you to this image: https://i.slow.pics/31fiYciR.png

it is 500nS time base, and shows about 2 MHz pulses of about 17 mV p-p. this is clearly the switcher's switching frequency which is about 2.2Mhz.

Those pulses are in the ~50MHz frequency range.  They may be happening at a period of the switcher's 2.2MHz, but as you can see they are tiny single cycle ~50MHz oscillation happening twice every 500ns, a single pulse going up, and a single pulse going down.

If it was a 2MHz signal, you would see a single fat sine wave going up and down, where each pulse would be a fat wide 250ns high and 250ns low.  This is not what we are seeing.  This is why choosing the right filter is important here and optimiziog your output filter for 2MHz might end up filtering nothing letting that higher 50MHz pulse go right through.  You want a filter where the '50MHz' component will be something like 100 or more ohms series resistance so any 0.1uf output cap will divide that component by a factor of 100 or more while the lower frequencies like below 2MHz will be a short of 0.01 ohms or less.

[VEGETA]

ok so adding 100nF cap can solve it? I will try that soon, I have some 0.1uf elec caps around

[BrianHG]

Take a look at your capacitor impedance charts.  You want to choose capacitors with the lowest impedance in the 20-50MHz range.  This may be a smaller value than 100nf.  You need the true specs from a cap company with true readings.

The same goes for the series ferrite bead or inductor before that cap.  You want as close to 0 ohm as possible, but, at 20-50Mhz, you want as high impedance as possible.

[KubaSO]

For supply evaluation, I always add an AC-coupled noise amplifier on the board – say x20 or x50, and an active rectifier and active peak detector. That way it’s easy to see ripple on any oscilloscope, and measuring peak and average ripple/noise can be done with a multimeter. But those three circuit functions need to be also well designed and fairly broadband (say 100MHz bandwidth) – and their layout is critical too. So it’s a chicken and egg problem: you have to already know quite a bit to design all that. For ripple itself, you’ll do fine with a lower bandwidth amplifier, since otherwise the scope isn’t sensitive enough, so even 10MHz is better than nothing. That still calls for an op-amp with 100Mhz GBW, and you’ll need two stages of gain. And it that opamp is quite hot when idle, you know you got oscillation – then just compare the op amp’s supply current with the quiescent value in the datasheet. You’ll want current measurement shunt resistors installed on the PC board to avoid messing with multimeter’s ammeter series connection. Heck, that amplifier could be discrete, to save on cost. But without a scope it’ll be hard to make sure it all works. All amplifiers can oscillate and their DC readings taken with a multimeter will look great, but it’ll be misleading. And some wideband amplifiers just run hot by design – especially the older video amps.

You’ve got many pointers from other replies. One thing I’d add: whatever “layout examples” the manufacturer provides don’t mean that you can just copy them and are guaranteed stuff will work. It still takes understanding of what the layout decisions mean and why they were made. Those are handy “sanity check” references for experienced engineers. For novices who can’t but copy without understanding, they are only mildly helpful.

Furthermore, in switching supply design, the non-ideal properties of components do matter, and things like ceramic capacitor dielectric type and ESR and SRF (self resonant frequency) of any capacitor type do play an important role.

And then, just looking at the size of that switcher chip: the inductor physically looks like it’s way too big. Those things usually scale together, so the physical size of the switcher chip and that of the surrounding passives go together at least roughly. Not always, but such big disparity is a red flag.

[VEGETA]

I didn't quite understand last point about capacitors ratings of impedance vs frequency. But know that I opted for caps from Lelon brand since they are good well known and available for JLCPCB, that is for elec. caps. for ceramics, I get what is available on LCSC and JLCPCB SMT service.

However, for ferrite bead, I get it from Digikey\Mouser from well-known brands... I posted few suggestions which has quite good impedance at 20-50mhz which could replace the originally recommended by chip manufacturer.

as for inductor size, it is 4.7uH and no way I can find one 2x2mm (size of the switcher). Recommended inductor is smaller than what I got for sure but physical small size isn't important to me, but price is. LCSC inductors are a lot cheaper than say Pulse branded ones and do the same. I got ones with greater ratings than required to eliminate any chance of margins.

I didn't blindly follow the example layout but rather did my own using best practices. You can find pictures of it in previous replies, people here helped doing it.

So to sum-up, here are some suggested solutions:

1- keep all the same, and go for a 2nd ferrite bead using one of the suggested above after the final 47uF elec. cap. a bead which has good impedance at 20-50mhz as the parts I listed above.

2- add another LC circuit comprised of 1uH inductor (small size) + 22uF caps as PI filter. meaning, split the output caps to be before and after the added L. ensure 47uF elec. cap to be the absolute final before the connector.

3- same as 2 but also add another ferrite bead.

4- same as 3 and 2, but add 10nF or 100nF caps next to big caps.

note: i am ok for getting one nicely branded cap but it should be say 0805 so I can hand solder it. one extra BOM line won't ruin the entire design economy. Please check the inductor I chose too.

which do you think should we go with? if not, please suggest a practical move i can use.

Notice that I am fine with <9mV of total ripple but just wanted to filter out the high frequency noise spikes. I don't think it will affect the final output of the system but still should be solved if possible. this design is already x2 times better than all other solutions. also, better in terms of ripple than original one, but worse in noise... original one doesn't have any noise spikes at all.

[VEGETA]

Take a look at your capacitor impedance charts.  You want to choose capacitors with the lowest impedance in the 20-50MHz range.  This may be a smaller value than 100nf.  You need the true specs from a cap company with true readings.

The same goes for the series ferrite bead or inductor before that cap.  You want as close to 0 ohm as possible, but, at 20-50Mhz, you want as high impedance as possible.

I wanted to make a final revision (hopefully) so I am interested in your suggestion, which is after the final 47uF elec. cap... adding a series ferrite bead and a 100nF or so to ground.

So I am trying to search for values similar to what you have explained, so can you help?

I found this cap and here is its data. click characteristic graph -> frequency graph -> |Z|

I found its lowest Z is at 25 MHz, but this is the resonance frequency right?

what do you think?

Also, here is another one which is 10nF: GCM155R71H103KA55D this one has 0.5-0.2ohms at 25-50mhz and its resonance frequency is about 90mhz.

[VEGETA]

I posted the question to TI support to see if they have a good recommendation, plus asked them about exceeding maximum output capacitance of 200uF.

Looking forward to your comments on my previous reply.

[Vovk_Z]

plus asked them about exceeding maximum output capacitance of 200uF.

I'm not sure simple increasing a capacitance at the output will help with noise or spikes. It doesn't work like that in real life. You just have to make some experiments with different types of caps of a low value (1-47 uF), consider playing with electrolytic caps of a general type with high ESR too.

[T3sl4co1l]

And like I said. It might be common mode noise still.  It is impossible to tell without a proper setup, free from conflating sources (from the power adapter, load, etc.), and preferably in a jig that gives standardized conditions (typically LISNs over ground plane).

Tim

[VEGETA]

And like I said. It might be common mode noise still.  It is impossible to tell without a proper setup, free from conflating sources (from the power adapter, load, etc.), and preferably in a jig that gives standardized conditions (typically LISNs over ground plane).

Tim

I do appreciate your pro tips and setup but I am not that advanced nor the product itself. I mean, we came a really long way to achieve this low noise level after much help... just one more issue remained. I will be very happy if we could attenuate it by any factor for now.

TI chip will be in stock in May but my PCBway can get them now for 10$ per IC... so I can make one more prototype and 1 assembled board.

[VEGETA]

Hello,

I have been idling for a long time, I thought of some minor changes could be done to enhance the thing with little to no cost. I thought of changing the CMC to a better one, here are the candidates I searched for:

https://www.digikey.com/en/products/detail/pulse-electronics-power/PA4339-102NLT/9659623 -> 1.02k ohms @ 100MHz, about 600 @ 20mhz and 1000 @ 50 mhz. this is a massive upgrade of attenuation from previous one  (DLW5BTM501SQ2).

https://www.digikey.com/en/products/detail/tdk-corporation/LCV70-701-2PL-TL00/11682555 -> 700 ohm @ 100 MHz. Upgrade from old one but the first candidate is better IMO. Unless Z @ freq. is not the only factor.

https://www.digikey.com/en/products/detail/w%C3%BCrth-elektronik/744237152/14011659 -> better than both but higher price.

^ all this assuming 12v @ 3amps of absolute maximum, reality is less.

original CMC was noticeably cheaper, these are easily 0.5~0.8$ more in price. So I thought of removing some inductors and caps here and there to save some cost, and put only one big main filter inductor instead of one per switcher. What do you think of that? so something like this:

12v -> new CMC -> elec. caps (47uF x 2) -> big inductor (cheap one but big) -> 1000uF elec cap. then input to each switcher will get 2 or 3 47uF caps + 2 22uF ceramics. while the 12v output only has like 1 47uF since it is not really used much.

or instead of one main filter cap I put one per regulator but leave the 12v without inductor?

I thought of this due to the harmonics seen are mostly common-mode as many here agreed, plus getting more diff. inductor filtering should help.

I am open to suggestions.


EDIT:

inductors selected for now:

100uH: https://lcsc.com/product-detail/Power-Inductors_TAI-TECH-TMPC1206HP-101MG-D_C357256.html
33uH:   https://lcsc.com/product-detail/Power-Inductors_PROD-Tech-PSPMAA1040H-330M-ANP_C436575.html
10uH:   https://lcsc.com/product-detail/Power-Inductors_PSA-Prosperity-Dielectrics-MCS0630-100MPY1_C2937094.html

so putting one 100uH as main filter, then just bulk caps before regulators seem good or putting 33uH per regulator without main L. what is best? the 33uH is slightly more expensive than 4.7uH one! still gonna need a 4.7uH for the switcher itself as indicated by datasheet.