My PSU design ripple and noise with picture measurements

[VEGETA]

yes the other power supply uses lm2596 which has 150 khz switching frequency, with big inductor has "330" label on it, but output caps are just 100uF elec. caps per rail. it has solid ground plain though.

Mine runs about 2 MHz switching, therefore smaller inductors. I assume solution is already done on the newer version, perhaps check the previous replies to find it. we put CMC + one ground plain with lots of stitching + short loops for switchers + bigger ceramic caps + more input filteration + making 12v output using a separate path with separate filters than the 12v fed into the switchers. I think we all agreed back then that it would be good to solve all issues but we are waiting the PCBs which will take 4 weeks.

So lets discuss the benefit of choosing slow frequency vs fast frequency switching here... what could be the best to use? faster switchers seem more desired these days but what in terms of noise\ripple?

[T3sl4co1l]

Best for what?  Ripple?  Non sequitur -- ripple can be made arbitrarily low, given adequate filtering and shielding.  It's the other constraints that limit performance: size and cost primarily, with efficiency, component selection, V/I range, etc. also being relevant.

Tim

[VEGETA]

Best for what?  Ripple?  Non sequitur -- ripple can be made arbitrarily low, given adequate filtering and shielding.  It's the other constraints that limit performance: size and cost primarily, with efficiency, component selection, V/I range, etc. also being relevant.

Tim

i mean for my application, 50x50mm board, low noise\ripple requirements as much as possible.

my switchers promised < 1 mV of total ripple and noise but the case in real life is not like this. we addressed this to be a grounding problem with what I explained in the previous post. so the new version of the circuit is on route.

down through that we landed on the idea that it is also common-mode noise, so CMC is added. I guess CMC can also help in regular noise suppression since it is a big L after all right?

however, low frequency stuff could have less worse ripple\noise if filtered through large L and C right?

I didn't address the PSRR of the switchers, which is a good value really. but the < 1mV is supposed to be assuming no ripple\noise on the input to begin with. but in my case and most cases, there are a ton of ripple\noise. So I needed to use more filters, bigger caps, better grounding to address this issue.

I am hyped about the new design, hoping to be the last one.

[VEGETA]

this is both channels on 3.3v for my psu

[BrianHG]

Ok, you have proven that your scope can properly do a differential measurement.
Meaning if you place channel #1 on the VCC and #2 on the GND with this setup, the MATH channel will show your true noise at the output.

Also, looking at the current background noise in the MATH channel, your scope's approximate background noise level is ~15mv p-p.  It might be possible to turn on the scope's measurement stats for the MATH channel as well.

Now, your scope's timebase is 10ms/div and the waveform is approximately 2 divisions long.
IE: 1s/0.020 = 50Hz.  Is this correct?

If so, this interference is not your switcher oscillating, it's a leakage from your AC mains bleeding into/through your circuit.  Changing a tiny output cap or inductor on your PCB will not solve this problem.  (Those changes are to correct problems at the 2MHz switching speed which we cannot currently see.)  Now that your scope is properly setup, please do the differential measurement of the rail, with both probe GND clips tied together, then clipped to the dreamcast's GND through a series 100ohm-up to-1kohm resistor.  The purpose of this measurement is to verify that the route of this 50Hz interference isn't coming from a looping GND throughout your multiple AC main power supplies throughout your measuring scope's GND.  Verify that with the same time-base setting that your competitor's board doesn't bleed through the same 50Hz.

(You might as well test the 12v rail too.  This interference may be coming through there.)

If your video output is 50Hz, this may be CPU processing load once every V-sync.  To verify this, load a different game and this waveform's duty cycle should change or disappear.  If this is the case, then we need to look at the regulation feedback filter on your PCB and again, the same test is worth looking at on your competitor's PCB.  Also, this 50hz signal may not be on the 5v rail if it is due to CPU processing modulating current draw when rendering each frame.

[VEGETA]

Ok, you have proven that your scope can properly do a differential measurement.
Meaning if you place channel #1 on the VCC and #2 on the GND with this setup, the MATH channel will show your true noise at the output.

this was the case in previous images. last image was just because I forgot to do it with the rest.

Also, looking at the current background noise in the MATH channel, your scope's approximate background noise level is ~15mv p-p.  It might be possible to turn on the scope's measurement stats for the MATH channel as well.

I didn't find it, but since all channels are the same vertical resolution (20mV/Div)... I just use ch1 measurement to measure all waveforms.

Now, your scope's timebase is 10ms/div and the waveform is approximately 2 divisions long.
IE: 1s/0.020 = 50Hz.  Is this correct?

I wasn't paying attention... perhaps the pictures above contain this info.

Now that your scope is properly setup, please do the differential measurement of the rail, with both probe GND clips tied together, then clipped to the dreamcast's GND through a series 100ohm-up to-1kohm resistor.

Will do it soon.

The purpose of this measurement is to verify that the route of this 50Hz interference isn't coming from a looping GND throughout your multiple AC main power supplies throughout your measuring scope's GND.  Verify that with the same time-base setting that your competitor's board doesn't bleed through the same 50Hz.

So i should do this measurement setup for both my psu and the other one?

ok but what to expect in both cases? I mean why the resistor?

If your video output is 50Hz, this may be CPU processing load once every V-sync.  To verify this, load a different game and this waveform's duty cycle should change or disappear.  If this is the case, then we need to look at the regulation feedback filter on your PCB and again, the same test is worth looking at on your competitor's PCB.  Also, this 50hz signal may not be on the 5v rail if it is due to CPU processing modulating current draw when rendering each frame.

this is interesting.

I cannot do it now but will find an SD Card to load games on it then use it since I have gdemu not a gd-drive.

[BrianHG]

Now, your scope's timebase is 10ms/div and the waveform is approximately 2 divisions long.
IE: 1s/0.020 = 50Hz.  Is this correct?

I wasn't paying attention... perhaps the pictures above contain this info.

Yes they do.

Now that your scope is properly setup, please do the differential measurement of the rail, with both probe GND clips tied together, then clipped to the dreamcast's GND through a series 100ohm-up to-1kohm resistor.

Will do it soon.

Ok.

The purpose of this measurement is to verify that the route of this 50Hz interference isn't coming from a looping GND throughout your multiple AC main power supplies throughout your measuring scope's GND.  Verify that with the same time-base setting that your competitor's board doesn't bleed through the same 50Hz.

So i should do this measurement setup for both my psu and the other one?

ok but what to expect in both cases? I mean why the resistor?

The resistor separates the scope's GND from your PSU GND just in case the 50Hz ripple is running throughout the GND.  It may also be useful to check if with the Dreamcast disconnected from any TV if it is currently wired there.

If your video output is 50Hz, this may be CPU processing load once every V-sync.  To verify this, load a different game and this waveform's duty cycle should change or disappear.  If this is the case, then we need to look at the regulation feedback filter on your PCB and again, the same test is worth looking at on your competitor's PCB.  Also, this 50hz signal may not be on the 5v rail if it is due to CPU processing modulating current draw when rendering each frame.

this is interesting.

I cannot do it now but will find an SD Card to load games on it then use it since I have gdemu not a gd-drive.

Yes, if the Dreamcast is currently rendering a really simple scene, IE drawing 3 amps for part of a frame, then going down to 0.5 amps for the rest, this 50hz ripple we see would be within the regulator's transient load response and would probably only show on the 3.3v rail.  The other PSU you have may be doing the same thing, you just need to measure that supply with the same scope settings.

Typically, this type of problem would be solved with either a low esr bulk capacitor on the 3.3v output right at the PSU power connector, or, modification of the feedback filter for the regulator's 3.3v sense circuit, or use a heavy-duty over-current regulator which wont react to such a transient.  However, now with a scope capable of a differential measurement, you can now measure ripple on the feedback resistor network VS the output to trace the problem.

[VEGETA]

So to sum up, which test should I do for now without the game? which ones with the game? so I can do it.

[BrianHG]

You may skip the game for now if you like.  Just measure the 2 PSUs with the same settings on the scope.

If you are going to test a game:
Does the Dreamcast have a demo or power-up test mode?
Look for a game with just a still intro screen which stays on.
We just want to see if it changes the waveform you see now because a 50hz signal can be coming from the AC mains, or, just the right CPU intensity processing in a PAL vidoe system.

[BrianHG]

Also, if it's power sag or interference from your power source, have you considered it's your input power choke?
Did you try shorting it out?

Also, what about placing a 2200uf cap on the power output connector?  (When powered off of course, the powering on.)
What does your scope output look like with the probes on the cap?

In your scopeshots, I didn't see anything saying that the probe was set to 10x or 1x.  Do you have a 10x/1x setting on your probes?

[VEGETA]

Also, if it's power sag or interference from your power source, have you considered it's your input power choke?
Did you try shorting it out?

Also, what about placing a 2200uf cap on the power output connector?  (When powered off of course, the powering on.)
What does your scope output look like with the probes on the cap?

In your scopeshots, I didn't see anything saying that the probe was set to 10x or 1x.  Do you have a 10x/1x setting on your probes?

I got the probes to x1 to be able to measure beyond 50mV/div.

I don't have a CMC in this circuit, please remember that this is the old circuit... CMC and grounding solution is on new circuit which is on route.

you mean putting very big elec. cap on 12v input from the ac-dc power brick? since I cannot put that much capacitance on 3.3v or 5v output since switchers specify 200uF maximum capacitance as a must... I am about 20uF away from that on this board and about 200uF on new board.

so the tests i should do for now includes this resistor from DC ground to my scope ground leads for both PSUs right?

[BrianHG]

Also, if it's power sag or interference from your power source, have you considered it's your input power choke?
Did you try shorting it out?

Also, what about placing a 2200uf cap on the power output connector?  (When powered off of course, the powering on.)
What does your scope output look like with the probes on the cap?

In your scopeshots, I didn't see anything saying that the probe was set to 10x or 1x.  Do you have a 10x/1x setting on your probes?

I got the probes to x1 to be able to measure beyond 50mV/div.

I don't have a CMC in this circuit, please remember that this is the old circuit... CMC and grounding solution is on new circuit which is on route.

you mean putting very big elec. cap on 12v input from the ac-dc power brick? since I cannot put that much capacitance on 3.3v or 5v output since switchers specify 200uF maximum capacitance as a must... I am about 20uF away from that on this board and about 200uF on new board.

so the tests i should do for now includes this resistor from DC ground to my scope ground leads for both PSUs right?

Yes on the resistor PSU tests.
The elect cap should be on the 3.3v.
Yes, I know you exceed the recommended switcher's rating.  If you are afraid of oscillation due to a lemon value of cap, going to 4700uf or 10000uf cap should be too high to allow the switcher to inject odd oscillations into the 3.3v rail.

[VEGETA]

Hello

done the resistor test for my psu, check it in pictures with names.

Here are some comments on pictures:

- odd behavior picture:

this happened after testing several caps to get results for other pics, such as 220uF, 1000uF, and 2200uF. then I noticed the waveform is flat with periodic bump as you see. I hooked up the video output and saw the picture completely messed up!! please see it in detail. it blinks each half a second or so while the colors are messed up a lot... slow functionality too like some delay until it registers a button move. I really regret doing the cap thing now since it ruined my dreamcast for reasons I don't know! messed up screen picture is in 2nd post.

[VEGETA]

here is the messed up screen.

any way to know the issue or fix it?

[VEGETA]

hello

after disassembling it and re-assembling it, it stopped turning on! it turns on for a brief moment then shuts off.. i measured about 86 ohms from 3.3v to gnd unlike 5v which is open load, so suspected something wrong.

checked many components which deals with 3.3v, left pins, etc.. but still 85 ohms... to make it worse I destroyed a sharp 07vz1h regulator! even after this regulator is out, still 85 ohms.

check pics: https://slow.pics/c/JOSrApga



EDIT: it now powers on completely with fan and so on, voltages are ok but there is no output picture. I measured voltages on caps and found the cap highlighted in picture to have 0v on its + side and about -0.5v on its negative side: https://slow.pics/c/PGsUaNuK

EDIT 2: that sharp regulator takes 3.3v and outputs 2v according to this schematic: https://archive.org/details/Dreamcast_Schematics_U
what is going on? could it be that that regulator supplies power to output video?

[T3sl4co1l]

Yeah, it's a good idea not to plug in capacitors randomly to powered circuitry.

Toasted regulator?  Ah, seems likely the large capacitor effectively shorted out the main supply, reversing the regulator, with more current than it can handle.  Lucky it wasn't something much worse?  Or maybe that's just the first thing of notice...

Tim

[VEGETA]

Yeah, it's a good idea not to plug in capacitors randomly to powered circuitry.

Toasted regulator?  Ah, seems likely the large capacitor effectively shorted out the main supply, reversing the regulator, with more current than it can handle.  Lucky it wasn't something much worse?  Or maybe that's just the first thing of notice...

Tim

most of the times i shut the system off, then plugged the cap and turned it on. but yes did put caps while it was working, just not many times.

I still don't understand what went wrong till now. the regulator made by sharp wasn't bad or at least I didn't verify it. I went along the board trying to disconnect all 3.3v related regulators and destroyed it by being a super saiyan in a wrong time and place.

but even if without destroying it, the picture was distorted. but now it works and turns on but no picture at all... which makes me thing the video circuitry got the problem. don't know if it is the analog video or the gpu itself.

that cap is showing reversed voltage which is very weird. it needs more investigation but now I ordered a new dreamcast locally which I learned to be more careful.

[BrianHG]

I so sorry to hear your dreamcast got damaged up.  Like 'T3sl4co1l' and what I wrote in my original post:

Also, what about placing a 2200uf cap on the power output connector?  (When powered off of course, the powering on.)

The cap had to be secure and not attached/detached while powered on as it may appear as a short, or, a voltage surge if the cap had a charge.

here is the messed up screen.

any way to know the issue or fix it?

It looks like a data bit in memory is stuck low.
This is odd because corrupt memory usually means the CPU wouldn't be able to do anything at all.
And yet your CPU is still working and executing code.

When you say it may be in the video output section, it is rare that this could be the video DAC, but not impossible.

Does anything else work?
Is there audio?

Were you able to replace the busted regulator?

Looking at the schematics, can you probe IC 401, with your probe in 10x mode for the high frequency, and check the data line inputs to see if any are stuck as well as the outputs?

Found cleaner scan of schematics here:  https://console5.com/wiki/Dreamcast

[VEGETA]

Hello,

new boards have arrived, and looks like we nailed it perfectly this time!
 
I probed using differential probing of 2 channels then math function as done previously, also using the same 12v regulator. Now on a new dreamcast which is working fine. no game used though, and has its gd drive in place while the previous one used a clone gdemu.

Here are the details:

- org: original PSU from SEGA.
- mine: 2 pics, one measuring ripple only and one including noise spikes.
- competitor: 2 pics, one for ripple only and one including noise spikes. this one similar to the one used before.


Results:

original: about 65-70mV p-p ripple with no noise spikes. I kinda expected better.

mine: about 9mV p-p of ripple and about 31mV p-p noise spikes which are not a lot.

competitor: around 20mV p-p ripple and about 45mV p-p noise spikes.

please check the waveforms and tell me your opinion.


I conclude that CMC + grounding enhanced the design significantly, it is now more than 2 times better than best competitor. certainly more than good enough as a high quality product. Stock PSU still excels more at grounding as seems from the yellow noise or maybe that it doesn't use any high frequency switching (fly-back) plus its own CMC made it better to suppress noise, or more clearly it doesn't make noise.

Looking forward to your feedback!

[T3sl4co1l]

Shrug, that's just output resistance.  Well within limits it seems.  What about the switching ripple, what's it look like around 1us/div?

Tim

[BrianHG]

Zoom into and trigger/capture the spikes.
They are too high frequency and being mucked up by the slow sampling rate and 10ms/div speed.

Your new supply seems immune to that 50hz buz in the signal.

For the high frequency spikes, if they are there in the differential when zoomed in on the time base, then they are authentic.
Because of their high frequency, strategically placed 1-10nf caps in the right place or strategically positioned ferrite bead should get rid of them.  Then you will be pretty much on par with an analog supply.

[VEGETA]

Shrug, that's just output resistance.  Well within limits it seems.  What about the switching ripple, what's it look like around 1us/div?

Tim

I don't understand what you mean by just output resistance.

I probed exactly like all previous times, literally the same. the psus are on the actual device and it is running. now gd-drive instead of gdemu. gonna try gdemu next time which will add a small load

Zoom into and trigger/capture the spikes.

aren't the spikes fully visible hear? or you want the trigger level to be on them so that normal signal is not triggering?

They are too high frequency and being mucked up by the slow sampling rate and 10ms/div speed.

I will try that next week (tonight is just this) but I think I changed to different time bases and saw nothing hidden or special. same method for other supplies as well.

Because of their high frequency, strategically placed 1-10nf caps in the right place or strategically positioned ferrite bead should get rid of them.

adding a small 10nf ceramic on each rail is doable but where? the space is limited, and I think I have some small through hole ceramic caps in this range available right now... so I could use them in this current circuit without building a new one. next pcb should be a small production batch not a test one.

Then you will be pretty much on par with an analog supply.

can you kindly explain?



so assuming this psu still the way it is here, is it good in your opinion? what specs should I write for it in its product page besides (high quality low noise PSU...)?

[BrianHG]

Zoom into and trigger/capture the spikes.

aren't the spikes fully visible hear? or you want the trigger level to be on them so that normal signal is not triggering?

They are 1 pixel wide.  What is their shape and frequency?
Spikes seem to exist on both the GND and VCC channels, and with 1 pixel wide spikes, maybe you are seeing a sample on one input channel and not the opposite one due to the sample being missed as it is 1 pixel wide.

As for analog power supply comparison, most switching supplies have a continuous ringing on their output when under load.  Since you are not zoomed in, I don't see this interference problem.  Though we still cannot also see perhaps an EMI coming out of your switcher.  Most likely, you will need to run your probe in 10x mode with the scope time-base set to the sub uS range to see anything there as your new supply operates in the 2MHz range and if switching noise is getting through the output filter, that signal will have edges up in the 500-200ns range.  An analog linear supply will not generate this type of noise.

[T3sl4co1l]

I don't understand what you mean by just output resistance.

Measure load current.  I think you will find it's pulsing up and down at the vertical refresh rate.  (This will be a PAL or SECAM model, I expect?).  Thus, the voltage is changing in relation to load current: output resistance.

This has nothing to do with the converter's switching cycle, at some 100s of kHz.

aren't the spikes fully visible hear? or you want the trigger level to be on them so that normal signal is not triggering?

Beats me.  I don't know your scope.  Is it aliasing the spikes?  Are they really spikes?  Is it switching ripple?  Ringing?  Lots of things it could be.  No way to tell from a 10,000m view.

Then you will be pretty much on par with an analog supply.

can you kindly explain?

A linear supply only rectifies mains voltage, filters, and regulates it if applicable; they have no switching noise besides the diodes, and that's at mains frequency and harmonics so is easily handled by a regulator.

There's nothing inherently impossible about a quiet SMPS, but they do take more effort to make quiet: you can't be lazy about layout, and may need shielding.

Tim

[VEGETA]

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

I mean, what do you really want to see or verify? 2MHz switching frequency (exactly the same as previous design) is big enough not to be shown and therefore it didn't need big inductors, yet we used 4.7uH main inductors. It also has ferrite beads integrated inside the feedback loop itself as explained later.

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 just don't want to do more pcb revisions for now.


I do report that the device works perfectly fine and image is very clear with no problems. 3.3v and 5v are spot on as value too.

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