My PSU design ripple and noise with picture measurements

[Vovk_Z]

I ordered Owon VDS-1022I USB scope, will arrive next week or so.

It is good that it has galvanically isolated power lines. So you wan't have an excessive induced common mode noise.

[VEGETA]

first time I know about microvoltmeter but it doesn't look like it draws like an oscilloscope but rather a value like multimeter.

I think there are reasonably cheap or affordable oscilloscopes which can do 500uV/Div but are they good enough?

VDS-1022I is isolated yes, which means no common-mode from source.

[VEGETA]

I've got OWON VDS1022I and the attached images are taken by it for 3.3v and GND of both my PSU and other PSU. Notice that my psu is the old version which I posted its measurements before... the new improved version with proper grounding + CMC + better filtering + bigger sized ceramic caps is going to start being manufactured in 2-3 weeks due to chip shortage. will need time to arrive though.

summery of pictures:

My PSU:

ground noise floor: about 14 mv P-P with some noise on top.
3.3v regulated rail: about 17-20 mV P-P ripple with nearly no noise on top, just ripple.

Other PSU:

ground noise floor: about 5 mV P-P ripple with more noise on top.
3.3v regulated rail: about 11~12 mV P-P ripple with a lot of noise (40~50 mV P-P) on top of it.

Please check the images and tell me what do you think?

[T3sl4co1l]

What does the power supply and probing setup look like?

Tim

[VEGETA]

I have it in attachments... just regular scope probing directly at rail itself.

[T3sl4co1l]

What about the power supply?

The loop area of the probe, and it laying over the input cable besides, may be picking up much more than you wanted, even with the CMC at the input there.

Another way to tell is if the spikes drop in level when you put multiple turns of ferrite bead on the scope probe.

Tim

[BrianHG]

     You need to use 2 channels on your scope in differential mode to get the authentic noise output of your PSP. 

     Here's how:  Channel 1 on your test VCC (right on your power supply's GND output) and Channel 2 on GND (right on your power supply's GND output).  Both scope probes should be calibrated and both channels should be set to the same voltage and same bandwidth options.  Then you need to select the scope's math function and select channel 1 'minus' channel 2.  The math result display, sometimes it's a new third trace on the display, will be the true voltage noise coming out of your power supply.  Just move the VCC probe to test both rails.

     To verify that you are setup properly, if you move either probe, tying VCC to GND, or vice-versa, (IE: Both probes on VCC or GND) the math output should read 0mv noise.  (Or, at least as close to 0mv as your scope's input and sampler can get.)

     Move your 2 probes connections to the Dreamcast's power input and you will get the true noise as the Dreamcast's input.

     Your 2 probe's GND clips should be connected together to a common GND point on your PCB or Dreamcast.

[VEGETA]

     You need to use 2 channels on your scope in differential mode to get the authentic noise output of your PSP. 

     Here's how:  Channel 1 on your test VCC (right on your power supply's GND output) and Channel 2 on GND (right on your power supply's GND output).  Both scope probes should be calibrated and both channels should be set to the same voltage and same bandwidth options.  Then you need to select the scope's math function and select channel 1 'minus' channel 2.  The math result display, sometimes it's a new third trace on the display, will be the true voltage noise coming out of your power supply.  Just move the VCC probe to test both rails.

     To verify that you are setup properly, if you move either probe, tying VCC to GND, or vice-versa, (IE: Both probes on VCC or GND) the math output should read 0mv noise.  (Or, at least as close to 0mv as your scope's input and sampler can get.)

     Move your 2 probes connections to the Dreamcast's power input and you will get the true noise as the Dreamcast's input.

     Your 2 probe's GND clips should be connected together to a common GND point on your PCB or Dreamcast.

I will do that today, I think this owon scope supports math functions for sure.

What about the power supply?

The loop area of the probe, and it laying over the input cable besides, may be picking up much more than you wanted, even with the CMC at the input there.

Another way to tell is if the spikes drop in level when you put multiple turns of ferrite bead on the scope probe.

Tim

power supply is external laptop-style power brick. it is connected directly via the connector at the white plastic part, then 2 wires (red and black) goes to dreamcast. i guess i could move the scope probes a bit off it but even at current setup i think they are away enough not to pick up anything major but will try tonight.

notice that this design doesn't have a CMC and also with the bad grounding we talked about above. new design will arrive in a month or so.


but besides all this, what do you think of waveforms of both PSUs? which one is better?

[VEGETA]

please check setup and results of differential probing in attachments.

[BrianHG]

The connection and math look correct.
2 tests to make sure everything is ok.

1. tie the 2 GND scope probes together and to a different GND point than the power connector.

2. Tie the both probe's tip to the same VCC.  In your scope capture, channel 1 and channel 2 should both show the same signal exactly, but, the math channel should be flat with no signal at all.  If the math still shows something then double check your scope settings, or re-orient the probes so that they approach your PCB from the other side.  EMI may be entering through the probe casing and probe coax.

If in case #2, your you see the math channel flat with no signal, then once you move one of the probes from VCC to GND, the output you see on the 'math' channel will be authentic power supply noise at the connector.

[BrianHG]

CORRECTION: Channel #2 looks delayed compared to channel #1.  This should not be.  The spikes in channel #2 should line up with the spikes in channel #1.  The 2 should cancel each other out in the math channel.

Is there some sort of setting in the scope which delays the sampling of channel #1 and #2?
Is the sampling set to be simultaneous between both channels?

Do you have a bandwidth limiter, IE 20Mhz bandwidth set on one channel but not set on the other?


Like what I said in my test #2 above, tying both channel inputs together should show you the exact same signal on both traces and the math output should reveal a flat 0 signal.


Is it possible that your scope may sample the entirety of channel 1, stop, then sample the entirety of channel 2 at a different time instead of simultaneously sampling both channels?  This would be bad and make the differential measurement setup fail.

[T3sl4co1l]

I think it's just aliasing from the super zoomed out view.  Try more like 100ns/div, trigger on a spike.

And you can take a closer look at that, see what (if anything) in the circuit correlates with the spike rate, and ringing frequency.

Tim

[BrianHG]

Another snapshot showing the exact same 2.25us delay on Vegeta's PCB.
That is such a huge delay for a scope, maybe it is a bug on the Owon software/firmware.
I cant imagine it being a circuit design delay in the scope's electronics as 2.25us would require a huge amount of wiring.  But in software, we are talking about having a bad relative sample offset, or, one channel is going through a software filter and the other isn't.

The samples speed is set to 50msps.  Maybe is Vegeta turned up the sample speed to the high MHz, maybe he can bypass the bug, but you would still need to check.

Though, connecting probe #1&#2 to the same point with these settings doesn't produce 2 identical readings, we know something is wrong with the scope, most likely a software or firmware issue.

There should not be a 2.25us skew between channels.  Maybe Owon didn't test this specific setup.  I wonder is turning off the 'math' may fix the problem and align the spikes, or choosing a different timebase.  If so, it is clearly a programming bug in Owon's software.  Unless you are experienced, you have no clue how many bugs I've been running into in recent years compared to the days of the old-fashioned CROs with regards to component datasheets and modern test equipment which haven't been properly tested with all use case scenarios.

[BrianHG]

I think it's just aliasing from the super zoomed out view.  Try more like 100ns/div, trigger on a spike.

And you can take a closer look at that, see what (if anything) in the circuit correlates with the spike rate, and ringing frequency.

Tim

You could be right, but the timed pairs coming out which I pointed to.

He is running the scope at 50msps for 2 channels but the scope is capable of 1ghz with 1 channel, 500msps with 2 channels.  Vegeta needs to increase record length to get the full 500msps as the expense of the USB display refresh might slow down, but with much greater precision display.  Though, if doing this shifts the timing of the 2 samples from 2.25us to 0.225us, it shows there still is a fundamental bug in the scope's software and it needs to be fixed to get the true Channel 1 minus Channel 2 view.

[BrianHG]

Found the 'problem', see image: (It's in you scope settings...)

Error: You should not be triggering on both channels 1 and 2.  Only trigger on channel 1 exclusively, otherwise, channel 2 will randomly or find the next trigger event and re-align to a non-parallel position based on its waveform completely screwing up the whole purpose of the differential measurement.  (IE, the scope is waiting for a trigger on channel 1, then it waits again for the next trigger on channel 2.  You do not want that.  Channel 1 and 2 need to be aligned in parallel, so, you should only be triggering on 1 channel.)

This is assuming using 2 separate triggers on 2 channels simultaneously does the same thing as it would do on my Tektronix scope.  With this setup, my scope would generate the exact same error and doing this when attempting the same A-B 'math' differential common-mode measurement, the 2 channels are no longer aligned giving me a flawed result.  The math is being done on the sample data in the window with the relative alignment position you see, not like an old cathode ray tube analog scope where the input A-B function is done at the analog stage in the probe input amplifiers where such an alignment error could not be made.

Once again, if set correctly, my test#2 above, with Channel 1&2 on the same VCC should generate an identical signal on each trace while the 'math' channel should show dead flat 0.

The 50msps should also be able to be increased by using a longer record length as the Owon scope in use can do 1ghz, with a 10 million sample record.  Increasing this setting would generate a far greater refined scope shot and more accurate measurements even if you keep the horizontal zoom level at the current 5us/div.  (Assuming the Owon has enough ram, otherwise, the only way is to zoom in the time-base.)

[VEGETA]

So to sum up, I need to disable channel 2 triggering while keeping channel 1 triggering? so channel 2 can be active but take triggering from channel 1? I am new to using the scope so I would need some assistance. I will try that tonight when I return.

Also, you recommend changing sampling to just 50 instead of 500?

[BrianHG]

So to sum up, I need to disable channel 2 triggering while keeping channel 1 triggering? so channel 2 can be active but take triggering from channel 1? I am new to using the scope so I would need some assistance. I will try that tonight when I return.

Also, you recommend changing sampling to just 50 instead of 500?

Point #1, you are correct.

Point #2, since I do not know about your scope, this speed might be a hardware limitation.  If so, zooming in on the horizontal time-base will improve the speed.  Or, you may be able to increase the sampling setting at the cost of the on-screen display refresh running slower.  This you will have to find out on your own.  500msps means the sampler is running at 500MHz instead of 50MHz, but, there will be 10x the data to send to your PC to display the same amount of width on the scope display.

[VEGETA]

i did this, please see.

i couldn't modify sampling, it is auto

[BrianHG]

In this snapshot: both_channels_on_3.3v.PNG, if both channels were on 3.3v, how is it possible that channel 1 shows the ripple and channel 2 is flat?  Could you have disconnected your probe?  If both probes are on the same source signal with the same settings, they should show you the same image.

Also, because of the sampling frequency, you should set channel 1 & 2 bandwidth limiter to 20MHz.  The view will be cleaner and for this level of zoom, probably more accurate.

[VEGETA]

In this snapshot: both_channels_on_3.3v.PNG, if both channels were on 3.3v, how is it possible that channel 1 shows the ripple and channel 2 is flat?  Could you have disconnected your probe?  If both probes are on the same source signal with the same settings, they should show you the same image.

Also, because of the sampling frequency, you should set channel 1 & 2 bandwidth limiter to 20MHz.  The view will be cleaner and for this level of zoom, probably more accurate.

I have re-taken the images right after I named them, so this picture is ch1 on 3.3 and ch2 is on gnd.

I have searched for limiting stuff but couldn't find, maybe I should search more, please tell me if you know the option. but what is your opinion on the waveforms regardless?

I have now modified the picture to have correct name.

[BrianHG]

I don't see the both channels on VCC 3.3 screenshot.

You wont find the limiting stuff, your scope is already only 25MHz analog bandwidth anyways.

[VEGETA]

I don't see the both channels on VCC 3.3 screenshot.

You wont find the limiting stuff, your scope is already only 25MHz analog bandwidth anyways.

I can take this picture for you if you need to see it, but I remember it was like the grounds one.

I feel like it didn't really make the subtraction, or am I wrong? the resultant waveform seem nearly identical to 3.3v one, I don't feel there is a reduction based on ground signal. what do you think?

based on waveforms for both PSUs, what do you think of each?

[BrianHG]

It shouldn't be like the GND snapshot.

Channel 1 should have the pulsing noise.
Channel 2 should have the same pulsing noise.
Channel M should be flat, assuming your setup is correct.

Your GND snapshot has 3 channels,  #1 & #2 & #M all flat.

We are verifying to see if your measurement scope setup and probes are all setup identically and properly.  1 mistake, and the result will not look as I described above.  1 mistake and what you see in the 'M' channel is useless for this type of measurement.

[VEGETA]

It shouldn't be like the GND snapshot.

Channel 1 should have the pulsing noise.
Channel 2 should have the same pulsing noise.
Channel M should be flat, assuming your setup is correct.

Your GND snapshot has 3 channels,  #1 & #2 & #M all flat.

We are verifying to see if your measurement scope setup and probes are all setup identically and properly.  1 mistake, and the result will not look as I described above.  1 mistake and what you see in the 'M' channel is useless for this type of measurement.

yes this is what i meant by line grounds... ch1 and 2 are similar and the result is flat.

I will see tonight, will replicate the test again. but for now please assume it is correct. so, what is your opinion on this?

will proper grounding and separate paths for 12v + 1206 caps + ground stitching done on the newer version (still not arrived).. will this make it better by noticeable amount? at least better than other_PSU

[BrianHG]

will proper grounding and separate paths for 12v + 1206 caps + ground stitching done on the newer version (still not arrived).. will this make it better by noticeable amount? at least better than other_PSU

I believe the other PSU is just operating at a much lower frequency, hence using larger inductors and larger caps makes the signal softer.

Yes, at higher frequencies, the choice of inductors and cap quality will play a part in the output ripple.

Your first measurements from 2 days ago show a leakage on both PSUs, a set of dual pulses at 50Hz.  This is usually a bleeding ground loop from the AC mains from your source supply and it Y1 cap and primary rectifier diodes.  It typically causes a faint buzz in the audio and hum-bar scrolling through the picture.  If this is part of the problem you are getting, no small switching supply can fix this as different measures are needed to get rid of this.  This would be an annoying AC mains looping frame ground problem I've seen with some 12v supply, perhaps like the one feeding your switcher when you have the output of the Dreamcast attached to another device like the scope GND clip, or your TV's video input.  If this was the actual problem, you would not see this with an analog power supply feeding the 12v input or using an isolator transformer on your AC mains feeding the 12v supply.  A proper differential measurement of the output on your switcher will determine if it is reacting to such a problem, but, we need to know the measurement you are making is true.

As for the 2MHz ring, yes, with a scope, we can sniff some of your PCB and find a solution.