My PSU design ripple and noise with picture measurements

[Mechatrommer]

Yes!

Also test with some idle load, so it's not going from 0-100% each time.  Maybe a 10 ohm always connected, then pulse a 1.2 ohm across it.  Or 2 and 2 ohm (50-100% step).

it has been proposed earlier with low side switcher rig, hopefully this time you got his attention. Looks like he's into finding the solution in simulator..

[VEGETA]

Yes!

Also test with some idle load, so it's not going from 0-100% each time.  Maybe a 10 ohm always connected, then pulse a 1.2 ohm across it.  Or 2 and 2 ohm (50-100% step).

it has been proposed earlier with low side switcher rig, hopefully this time you got his attention. Looks like he's into finding the solution in simulator..

Originally I didn't understand that he wanted me to do this test on a separate load... but rather thought full load is only the dreamcast itself. now I know exactly what to do.

I will do a very simple rig. just some little boards with resistors on them, hook them to the supply with the scope using alligator clips after preparing it to do either single shot or continuous as agreed.

as for simulation, it will go in parallel to all this. won't hurt to try and won't delay the progress.

[Mechatrommer]

It needs sharp eyes to see mV differences between 20% load and say 80% load if you do switching with hand. If you switch the low side nmosfet with function generator or mcu programmed to output 100hz signal to the nmosfet's gate, you should be able to see that mVpp 100hz on your psu output, then thats your psu load regulation problem. If psu is stable, then problem is something else. If you do the test, post pictures on psu output and what kind of load you are giving it. Then we have better picture what your problem is.

[VEGETA]

I have finally done all resistor mini boards and here are the tests: https://slow.pics/c/Di7BsJkH

each picture has its name for it, all done via 3.3v rail.

quick notes: some readings of multimeter does not match the current, meaning i say 0.5R load and it gives 2.2A @ 3.3v while it should be 1.5 ohms. my multimeter is Aneng8009 and maybe it is not good after many years despite not used a lot.

for example, it gives around 1.5 ohms for that 1 ohm 50w resistor.

still, i think we have achieved our goad having these figures. however, I don't see much wrong when stepping up or down even when zooming.

for your kind review.

[T3sl4co1l]

Sanity checks:
- What rail is being tested?
- What is the other rail doing?

I have no idea!

test08_0.5ohm_2.2A_stepdown, test08_0.5ohm_2.2A_stepup: why is this so chunky?  Sanity check: real waveforms aren't made of line segments.  This is clearly a sampling error.  (You can further confirm by changing display mode to points instead of vectors, or enabling sinc interpolation, if these options are available.)  Perhaps you took an acquisition at very low sample rate, and zoomed in?

Hm, also very visibly aliased.  The mostly-flat parts dither between discrete values.  Surely the scales used during acquisition are completely different from what's shown (500mV/div, 1 / 2 ms/div).

Also, you're saying this thing steps ~3V with a 0.5Ω or 2.2A load?  No, that must be power-on/off.  So add that to the list:

- What test is being performed?

I assumed these were all load-step tests, because that's what we were talking about.  You can never provide too many details when documenting things!

test01_20ohm_130mA: OK so this will be a light-load (burst mode) waveform?  Why is the amplitude so unstable?  Is that an aliasing problem (too zoomed out), is it interference from the other channel (what's it doing? -- hence why that is important to note!)?  Does it change if the other channel is disabled (remove EN, or bias V(FB) up with a resistor, setting its output to 0V)?  Does it change if you touch anything in the circuit (maybe it's an instability or sensitivity)?

Be suspicious of inconsistencies.  Investigate.  See what affects them, what correlates with their appearance (did the other channel fire a burst at that moment?), what affects their frequency (repeat rate, wave frequency, pulse length, etc.) and amplitude?

test02_1ohm_1.3A: That looks more like the switching ripple I'd expect to see, though it is square not triangular, implying induction from the switching inductor (it's the same shape as the square wave applied to the inductor), or maybe ground-loop voltage.  But what the hell are the bursty things?  Are they present on every edge, or just erratically?  Does anything affect their presence?

test03_1ohm_1.3A_stepup: I guess this is 1.3 to 0A step?  Why does it rise so far?  Or, not very far at all, it's DC coupled and only 0.8V tall?  Wait is this V(FB) startup or something?!

test04_1ohm_1.3A_stepdown: I guess this is the shutdown then?  Why is it slow (an RC exponential, as one would expect!) when test08 is ~instantaneous [as far as the scope can tell]?

test05_10ohm_240mA: This is the first one that looks like real actual output ripple voltage.  It's smooth, reasonably periodic, a little lopsided as one might expect from a filtered square wave at some odd duty cycle.

test06_2ohm_875mA, test07_0.5ohm_2.2A: Essentially identical, but like test02, what are the spikes/bursts?  Why is the switching ripple square?

Tim

[VEGETA]

Hello

Sanity checks:
- What rail is being tested?
- What is the other rail doing?

I have no idea!

the rail tested is just the 3.3v rail as I mentioned somewhere above. the 5v rail does nothing but has on-board load resistor of 10k (same as 3.3v), while 12v rail is taken from the input itself, no other load is on it.

test08_0.5ohm_2.2A_stepdown, test08_0.5ohm_2.2A_stepup: why is this so chunky?  Sanity check: real waveforms aren't made of line segments.  This is clearly a sampling error.  (You can further confirm by changing display mode to points instead of vectors, or enabling sinc interpolation, if these options are available.)  Perhaps you took an acquisition at very low sample rate, and zoomed in?

I have never missed around with sampling. it says 250KS/s on the scope software for this particular shot for 1ms/div. when I lower the time base, sampling increases. for example 2us/div has 100MS/s.

kindly inform me which to choose for such shots.

Also, you're saying this thing steps ~3V with a 0.5Ω or 2.2A load?  No, that must be power-on/off.  So add that to the list:

here is the procedure I went with:

I powered the PSU, hooked the scope, and prepared single shot mode. Then I hooked the resistor first lead on 3.3v, then the other lead via a multimeter in current mode. Now I hold this lead and quickly hook it to the GND pin of the PSU , which makes the scope get the reading.

Here is a simple diagram for simplicity:

3.3v ------ Resistor+ ------ Resistor -----
                                                             |
                                                             |
GND --- multimeter- --- multimeter + ---

So it goes from low load (only the 10k on-board resistor) to the specified load. this is the case for all stepup\down pictures.

test01_20ohm_130mA: OK so this will be a light-load (burst mode) waveform?  Why is the amplitude so unstable?  Is that an aliasing problem (too zoomed out), is it interference from the other channel (what's it doing? -- hence why that is important to note!)?  Does it change if the other channel is disabled (remove EN, or bias V(FB) up with a resistor, setting its output to 0V)?  Does it change if you touch anything in the circuit (maybe it's an instability or sensitivity)?

I assumed it is the case for such low loads too. the amplitude is not constant maybe because I have chosen low time base, but if increased time base will show the full pattern. and no, nothing changes by touching, actually even if kept for minutes it will not change. so stability is ok.

test02_1ohm_1.3A: That looks more like the switching ripple I'd expect to see, though it is square not triangular, implying induction from the switching inductor (it's the same shape as the square wave applied to the inductor), or maybe ground-loop voltage.  But what the hell are the bursty things?  Are they present on every edge, or just erratically?  Does anything affect their presence?

yes I noticed the weird squares and no idea why it is like this. just notice it is 5us. bursty edges are maybe due to low timebase, when bigger time base it may not look the same.

test03_1ohm_1.3A_stepup: I guess this is 1.3 to 0A step?  Why does it rise so far?  Or, not very far at all, it's DC coupled and only 0.8V tall?  Wait is this V(FB) startup or something?!

I assure you it went to 3.3v as usual but maybe this is one of the pictures which was not correct. I had ones which went to 3.3v

test04_1ohm_1.3A_stepdown: I guess this is the shutdown then?  Why is it slow (an RC exponential, as one would expect!) when test08 is ~instantaneous [as far as the scope can tell]?

I zoomed in that is why it looks slow.

test08_0.5ohm_2.2A_stepup is for 2.2amps step up, and another test8 is for 2.2amps step down.

the only continuous ones are the ones not labelled stepup\stepdown.

test05_10ohm_240mA: This is the first one that looks like real actual output ripple voltage.  It's smooth, reasonably periodic, a little lopsided as one might expect from a filtered square wave at some odd duty cycle.

if I zoom out, all of them appear constant amplitude.

test06_2ohm_875mA, test07_0.5ohm_2.2A: Essentially identical, but like test02, what are the spikes/bursts?  Why is the switching ripple square?

spikes are maybe more visible if we zoomed out, I don't know really.

___

so after all these measurements, what can we conclude? for me I saw no ringing on step up and down despite sampling wasn't perfect. but they exist on actual Dreamcast load.

I am ready for more tests if needed.

[VEGETA]

Hello,

I have checked simulation, and all those parasitic L were 5nH inductance but also 5n DCR. I changed this DCR to be 1m, and done the new simulation. Results here: https://slow.pics/c/SUiVWKDV

waveforms:

yellow -> input 12v, has 200mv ripple p-p.
green -> 3.3v @ 3 amps.
red -> 5v @ 2 amps

I have noticed the exact same square pulsing on the output! is it a coincidence?

[Mechatrommer]

what is this? (A and B) A is about 150Hz 3-5mVpp (also can be seen in test06_2ohm_875mA and test07_0.5ohm_2.2A), where is that coming from? you must figure out where is that coming from. the switching IC? your room? or some ground loop caused by another device in your lab? turn OFF other devices, lamps etc you measure in dark room if you have to. but since your using PC connected DSO, so its difficult to rule out your PC noise. B is what? related to IC switching? or what? you must find it yourself, we cant access your circuit, maybe better to turn on dso channel 2 and see simultaneously the IC switching or Vinput noise (like i demo'ed in my thread linked earlier) thats why many channels dso is important in diagnosing this kind of circuit. does your simulation shows exactly like in the picture? if not, your simulation model is not good enough ymmv.

[VEGETA]

what is this? (A and B) A is about 150Hz 3-5mVpp (also can be seen in test06_2ohm_875mA and test07_0.5ohm_2.2A), where is that coming from? you must figure out where is that coming from. the switching IC? your room? or some ground loop caused by another device in your lab? turn OFF other devices, lamps etc you measure in dark room if you have to. but since your using PC connected DSO, so its difficult to rule out your PC noise. B is what? related to IC switching? or what? you must find it yourself, we cant access your circuit, maybe better to turn on dso channel 2 and see simultaneously the IC switching or Vinput noise (like i demo'ed in my thread linked earlier) thats why many channels dso is important in diagnosing this kind of circuit. does your simulation shows exactly like in the picture? if not, your simulation model is not good enough ymmv.

as for A, please check the simulation pictures I linked in previous post. they look very similar, almost identical (check sim last image). we can guess they are related to switching ic.

From scope pictures, A is about 1.5uS period = 0.67 MHz, therefore I conclude that it is the switcher since the switcher has a fixed frequency of 700khz. very close so it has to be it, especially that it exists in simulation almost identically.

imagine that it is the only thing that exists... I will be overjoyed by 5mV p-p total figure!

however for B, which is about 20-30 mV and it happens not in a regular pattern but overall slower frequency. could be the input 12v supply itself since I suspect its switching frequency to be around 300 khz or less.

I am out of home for now, will return in days. I will conduct new tests where I will put high continuous current and measure both 3.3v output and the input on the switchers + switchers output directly at inductor (before or after?).

then I will remove 4.7uH main L and install 10uH in its place, then repeat the tests as well.

so we need to take care of 2 switching frequencies: 700 khz one from the switcher, and the low frequency one from unknown source till now. plus, the 50-60hz due to actual final load as discussed.

as for loads in my house, only laptop with PC monitor + mobile chargers (hooked to power strip but no mobile was connected) + neon light (no switching lights). these are the ones in lab room, rest of the house is the same but with an old fridge. no effect is expected.

if the 12v source itself was the problem, then why the other psu delt with it correctly? I can use a chinese lab power supply that i got to see if it has other effect.


what do you think of this till now? waiting my return to lab + teslacoil's response.

current discoveries and discussion seemed to result in making 10uH instead of 4.7u, with a second stage of 1uH instead of the ferrite bead followed by filter caps (including 1 electrolytic). adding to all that a good LDO to deal with 50-60hz stuff if possible. making 2 versions will not be easy but I can manage to put them next to each other and then mouse-bite them together... making 2 versions one with LDO and one without can be time saving. but as agreed, first we need to understand more.

thanks

[VEGETA]

I am home and I can resume testing soon.

I will perform these tests:

1- Remove 4.7uH inductor and put 10uH in its place, then see what is the final result is.
2- Use another power supply source, plus measuring ripple and noise of it (and the currently used one). this is to see what type of noise is related to the input source.


On another level, I will get this soon (1-2 months):

https://www.amazon.com/dp/B082F36D3T/

I think it is enough, 60 watts and 10 amps are very good for my needs. there are others available but higher wattage is not required for me at this stage. what do you think about it?

[VEGETA]

Hello,

I have made a new update on this as follows:

since I the demand (and my ultimate goal) was to have a true universal AC power supply drop in replacement to original, I decided to do just that. so I have changed the design to take universal AC with the following changes and design choices:

- using Recom power module as AC-DC converter which delivers 12v. its size is more compatible than meanwell module but it is not as cheap. it should be better quality.

- I added RC212-0.5-10M common-mode choke since I got advised by many that such AC-DC modules probably do not have enough CMC filtering.

- the AC-DC module is filtered by 2x 330uF elec caps + 1uH inductor (similar to 2nd stage one mentioned below) + 2x 1210 22u ceramics + 1u ceramic.

- main 12v filter before each switcher is filtered by one 22uF elec cap + 2x 1210 22u ceramic + 1uF 0402 (very close to pin)

- switcher IC used is AP62600SJ-7, very cheap and capable.

- main inductor used is TaiTech_TMPC1265 which has very low DCR of 1.7mOhm which should give excellent ripple performance, it is better than eval board choices. huge size but I managed to make it fit as seen in images. 1uH is supposed to be enough especially that I will operate the switcher at 1.2 MHz.

- main filter caps are all X7R ceramics 10uF 0805 and 2 1210 22uF ones as seen. layout made them very close to SW and GND to have good coupling.

- 2nd stage filtering also using 1uH inductor but not as good, it is MWSA0603S which has 7.4 mOhm DCR. I compensated for this when choosing switcher output voltage.

- all ceramic caps are X7R, elec. caps of 22uF were put in every stage to help dampen the stages according to advice I got. Final output cap of 2nd stage, besides the 2 1210 ceramics, is a 330uF solid polymer cap with just 9mOhm of ESR. should give better performance than regular elec. caps for sure while having huge capacitance.


_________


i have made the layout as recommendation, but i am worried about if some of the pours i mentioned in pictures won't solder properly due to large copper without thermal reliefs.

for your kind opinion.

[VEGETA]

Edit: I have modified all zones to have thermal reliefs suitable to each pad... only switcher IC pads have solid connection due to limited space.

[VEGETA]

I re-arranged some components in a minor way and added proper silkscreen stuff.

any comment before sending it to create a prototype?

[VEGETA]

Hello,

I have got the design and tested it. please check images here: https://imgur.com/a/3Lqw5NZ

I also attached the images with this post. + pictures of it installed.


measurement results (ac coupling, short scope spring ground):

3.3v no load:
mine: 29mV p-p
org: 70mV p-p

5v rail:
mine: 7mV p-p
org: 33mV p-p

3.3v full load - main menu screen:
mine: 47mV p-p
org: 70mV p-p

3.3v full load - crazy taxi game (should be max it can take in its operation):
mine (ripple only): 26 mV p-p
mine (ripple + noise spikes): 65mV p-p
org: 90mV p-p


the one and only thing that needs changing is ac connector mechanical holes... need to be 1 mm bigger. I filed them to fit this time.

honestly, what is your opinion overall?

[VEGETA]

I wish to remove the 50-60hz big noise spark which is due to the console operating at those frequencies. if these big spikes removed, I will consider the design to be perfect.

I will try increasing the output capacitance since original PSU had much more capacitance than this, will see if it works.

output currently have 2x22uF ceramics + 330uF solid poly cap as final output caps + 22uF elec. cap for damping.

[Vovk_Z]

I wish to remove the 50-60hz big noise spark which is due to the console operating at those frequencies. if these big spikes removed, I will consider the design to be perfect.
I will try increasing the output capacitance since original PSU had much more capacitance than this, will see if it works.

IDK why your oscilloscope says frequency is 50 Hz or 60 Hz if as I can see pulsations have 100 Hz frequency (10 ms period), which is typical for 50 Hz mains and not enough 100 Hz main filtering capacitance after rectifier (otherwise it's just a pure PSRR of your regulators due to not ideal PCB wiring).
For test you may add the same capacutance (or about 50%) in parallel to existing main filter caps and look at those output voltages under load again.
If pulsations drop proportionally to increased main filter capacitance - then you see that it must be increased. If it doesn't change pulsations then you have fine filtering but pure regulator PSRR (pure PCB design/wiring etc, IDK).

Upd: I see now, main filter capacitor I'm talking about is a high voltage one in that Racom RACM-30 power module. And it has glued case, hard to disassemble...
This module gives you a lot of 100 Hz pulsations and it seems to me your posr-regulators don't do much with it.

[T3sl4co1l]

I think you're misunderstanding (you're not alone, the same confusion was had much earlier in this thread).  It's not line frequency, it's update rate of the console -- he's done (almost?) all load testing with the real system.  So you're seeing the pulsing load as the CPU/GPU prepare a frame, every frame.

Scope appears to be 10ms/div and pulses repeat 20ms or 50Hz, must be a PAL 50Hz version?

Edit: a bit less, 6 cycles in 5 divs = 60Hz.

Tim

[Someone]

I think you're misunderstanding (you're not alone, the same confusion was had much earlier in this thread).

That's this thread 100%, no 200%. It's trivial to simulate what more capacitance will do for the 50/60/100/120Hz noise/ripple, and experienced people already know exactly where this is headed. But the OP is set on their methods and isn't planning to expand their skills/knowledge.

[VEGETA]

Hello,

I wanted to do this update yesterday but couldn't because forum didn't work on my browser. Now using portable browser.

I added 3x2200uF at the final filtering stage to get rid of that low frequency spike and it worked as intended. please check the attached image which is full load while running a game.

Now only the 50-60 hz ripple or power sips which are as previous reply mentioned, just the console update rate which is typical and is seen on the original power supply itself.

Yesterday tried many many capacitance variations and looks like the key is high bulk caps at the very end to deal with power sipping demands. today if I get time I will try and add more caps at different stages especially after the module itself to see what difference will it make overall and how the module will tolerate it. the result in this image looks like the perfect form already, I don't think we can get rid of these cycles right?

next and final revision is where I will add the new caps to 3.3v rail and this time will make them through hole caps to save space (and cost), will add another low esr solid cap as well. so it will be total of 3x2200uF elec + 2x 470uF solid + 2x 22uF 1210 X7R ceramic. as for 5v rail it is good the way it is now.

[VEGETA]

Here are some new measurements.

I probed the module output after filter stage (just before the switchers), and as it seems even adding many capacitance didn't remove the 50Hz component. I tried adding 4400u and 6600u caps but no use either on its own measurement or at final 3.3v.

I also added 8800u for 3.3v rail (full_load_game_3v3_8800u picture) and as it seems a maximum of 20-25mV p-p total figure with 0 noise of any kind. I also tried making it 9800uF but didn't improve much, 6600uF also was slightly worse but overall similar. main menu is 50Hz since the console is PAL but game is NTSC crazy taxi thus 60 Hz.

therefore, I have made the following changes to be done on upcoming test sample:

- use 3x 220uF caps for main module pi filter instead of the 2x 330u ones.
- use 1x 220uF cap for each switcher input instead of 22uF elec one while keeping everything else.
- use 3x 2200uF caps for 3.3v rail final output + 2x 470uF solid polymer caps (instead of 1x 330uF).

caps are Lelon through hole RXW series which are low ESR and long life:

220uF elec.: https://www.lcsc.com/product-detail/Aluminum-Electrolytic-Capacitors-Leaded_Lelon-RXW221M1CCC-0611P_C916211.html
2200uF elec.: https://www.lcsc.com/product-detail/Aluminum-Electrolytic-Capacitors-Leaded_Lelon-RXW222M1ABKF1020_C2920647.html
470uF poly: https://www.lcsc.com/product-detail/Solid-Capacitors_Lelon-ORZ471M0JBK-0608_C434469.html or https://www.lcsc.com/product-detail/Solid-Capacitors_Lelon-ORK471M0JBK-0608_C176697.html

these caps are certainly better for SMPS (labelled this way from manufacturer too) than previous SMD ones, and for sure better than the crappy ones I tested now (with very dirty soldering job).

I am confident that this will keep it (and maybe make better than) maximum of 25mv p-p absolute full load. I mean, realistically can't be better than this TBH. I consider this the best available design from my side, extremely low ripple due to console itself and 0 noise. even by its current no big bulk caps readings it is already 2-3x better than original PSU.


I remember once that we spoke about LDO capability to eliminate such 50-60hz but now I am satisfied with this. I may try making an LDO version as a future testing and learning project though.

for your kind opinion.

[VEGETA]

Hello,

more interesting tests.

I have done 2 tests tonight:

1- using DC load itself, not the final console.
2- using a 1 ohm power resistor.

as seen below in images (with description)... the design looks very good. the measurements showed 2mv p-p final noise and ripple figure at full load, even similar reading on previous stages like 1st filtering stage and 12v rail.

however, in dc load 3.3v and 5v @ 3 amps measurement pictures you seem some random (yes random, did it many times to verify this) spikes and stuff happening above the 2mv p-p figure. this is for sure due to long wiring from the device to the dc load connection + input connection (lots of wiring). notice that switcher is running at about 1.2 MHz

therefore, I wanted to do a pure resistor load with short wires (and no input to AC power outlet, no noisy psu supplying it...etc) and the results were very nice! 2mv p-p on 3.3v and about 3v p-p for 5v.

notice that 3.3v rail has the extra 3x2200uF caps while 5v does not. these caps are necessary for final connection of the device on the console to be able to supply the huge power sips mentioned above. they are only required for 3.3v not for 5v though which is nice!

I attached the pics of my setup for reference.

I am very satisfied with this result, especially when I see it on the final design (with new elec caps mentioned before).

what is your honest opinion after this marathon?

[Vovk_Z]

Power rails look nice and clean now. (Working with pure resistive loax with short wires). Using just pure resistor you may identify pulsation and noise of you power supply more esily. Addind your pulsating load you'll know output impedance of your power supply.
Those spikes can be not connected with your power supply. It can be anything  like WIFI or LED lights or IDK. But it is good to investigate and find a reason otherwise you can't call your design 'low nose one'.

[VEGETA]

Power rails look nice and clean now. (Working with pure resistive loax with short wires). Using just pure resistor you may identify pulsation and noise of you power supply more esily. Addind your pulsating load you'll know output impedance of your power supply.
Those spikes can be not connected with your power supply. It can be anything  like WIFI or LED lights or IDK. But it is good to investigate and find a reason otherwise you can't call your design 'low nose one'.

yes after putting the resistive load I saw the waveform as mentioned, just about 2mv p-p. the 50-60hz ripple is just the console sipping power at those rates which can't be negated, at least with conventional means.

By "pulsating load" you mean what exactly? the 50-60 hz stuff is the console as mentioned.

I don't have LED lights, and for wifi it is not so close to the setup too. the same result is using the original power supply, meaning showing the 50-60hz cycles but at higher amplitude than mine.

I put a big common-mode choke on the ac side before the power module (which has its own), this was the key to filter all noise from the source. in previous designs, i couldn't beat the original power supply in this particular area. the original power supply always looked cleaner due to big and good CMC but after adding it to my design, it got the same.

what do you suggest I test next?

[Vovk_Z]

By "pulsating load" you mean what exactly? the 50-60 hz stuff is the console as mentioned.

Yes, the pulsating load is your console.

[Vovk_Z]

what do you suggest I test next?

This topic became qute large so I don't remember if there is a suggestion for you to read some helpfull books about noise and related things.
1. Highspeed digital design. A handbook of black magic. Howard Jonson, Martin Graham.
2. Noise Reduction Techniques in Electronic Systems. Henry Ott.
Those books give you a full.picture.