A couple of days ago I replaced my 12V/0,7A power supply to a 12V/2,1A power supply that I use for a couple of purposes. Everything looked okay until I found that my 1-wire network is completely down which I know happens as a result of noise. (The network is running from a completely different 5V power supply, sharing only the ground with the 12V supply.)

So I went on to check the new power supply and found what is seen on the attached image. I forgot to set my 10X probe correctly, which should mean 500mV spikes that are most likely causing my trouble.

The question is: how could I get rid off these? I know this is very broad topic, but I am hoping in some practical advice. Right now I believe a simple inductor should do the lowpass filtering job, but what size? Is it correct at all?

Thanks.

 

A couple of days ago I replaced my 12V/0,7A power supply to a 12V/2,1A power supply that I use for a couple of purposes. Everything looked okay until I found that my 1-wire network is completely down which I know happens as a result of noise. (The network is running from a completely different 5V power supply, sharing only the ground with the 12V supply.)

So I went on to check the new power supply and found what is seen on the attached image. I forgot to set my 10X probe correctly, which should mean 500mV spikes that are most likely causing my trouble.

The question is: how could I get rid off these? I know this is very broad topic, but I am hoping in some practical advice. Right now I believe a simple inductor should do the lowpass filtering job, but what size? Is it correct at all?

Thanks.

The dark portion of the trace looks to be about 5us which is about 200khz so could well be the switching frequency of the power suppy. You could try a simple LC or PI filter to give you about 55dB attentuation at the 200kHz and see if it helps. This should bring the 200mV and place the in the uV's. The fine 500mV pulses look to be much higher frequency (i can't see what frequency this is) so is likely to be a leakage spike in the switching converter and will likely be more problematic. I would suggest have a look at the low frequency first as the high frequency could well require a snubber or narrow notch filter to address. I attach an example cirucuit based on the 12V 2.1A capability of your power supply. I'd start there and see what you find.

 

The dark portion of the trace looks to be about 5us which is about 200khz so could well be the switching frequency of the power suppy. You could try a simple LC or PI filter to give you about 55dB attentuation at the 200kHz and see if it helps. This should bring the 200mV and place the in the uV's. The fine 500mV pulses look to be much higher frequency (i can't see what frequency this is) so is likely to be a leakage spike in the switching converter and will likely be more problematic. I would suggest have a look at the low frequency first as the high frequency could well require a snubber or narrow notch filter to address. I attach an example cirucuit based on the 12V 2.1A capability of your power supply. I'd start there and see what you find.

Thanks for your response. I haven't even noticed the smaller spikes at 5µs, I thought the switching frequency is given by the distances of the huge spikes (which is about 13µs, so 80kHz). Anyway, I will put together the filter you showed and see how it changes...

 

Thanks for your response. I haven't even noticed the smaller spikes at 5µs, I thought the switching frequency is given by the distances of the huge spikes (which is about 13µs, so 80kHz). Anyway, I will put together the filter you showed and see how it changes...

Perhaps i’m

Thanks for your response. I haven't even noticed the smaller spikes at 5µs, I thought the switching frequency is given by the distances of the huge spikes (which is about 13µs, so 80kHz). Anyway, I will put together the filter you showed and see how it changes...

The smallest timebase spike which is the biggest amplitude looks to be way smaller than 5us. To get a better picture you need to zoom in on one cycle to get the detail. My guess is you have a leakage spike near 9Mhz, the lower frequency could be some reverse recovery of a diode.

 

So today I took the supply out of the cabinet, put it on my desk, prepared a ~70Ω resistance as test load and checked it again. Surprisingly the spikes were there, but way smaller then they were before (image 1).

I suspected that some external noise is making fun of me, so moved back to the cabinet and checked again (image 2). The spikes are a little bigger, but still not that bad.

Now added little longer than half a meter wires to connect the resistance (the wires I originally used to connect the power supply output). This is when the spikes got a lot bigger (image 3). So now I completely shut down all the other units (other power supplies etc.) in the cabinet, in fact I switched off everything in the house. The spikes stayed, so they must come from the power supply.

Moved back again to the desk with all the attached cables. The spikes were still there (image 4). I prepared also the filter you showed (150µH, 2µF+2.2µF), but this time I could not see the lower frequency components. I attached anyway, but it had little effect to the big spikes. I also tried inspecting these spikes a little deeper, but I think my simple scope just not good enough for that, images attached. Anyway, the frequency of them must be way above 1Mhz for sure.

Question: how come a lowpass filter leaves those so high frequency components untouched? Any thoughts?

So now I pick my screwdriver and see what's inside that power supply....

 

So today I took the supply out of the cabinet, put it on my desk, prepared a ~70Ω resistance as test load and checked it again. Surprisingly the spikes were there, but way smaller then they were before (image 1).

I suspected that some external noise is making fun of me, so moved back to the cabinet and checked again (image 2). The spikes are a little bigger, but still not that bad.

Now added little longer than half a meter wires to connect the resistance (the wires I originally used to connect the power supply output). This is when the spikes got a lot bigger (image 3). So now I completely shut down all the other units (other power supplies etc.) in the cabinet, in fact I switched off everything in the house. The spikes stayed, so they must come from the power supply.

Moved back again to the desk with all the attached cables. The spikes were still there (image 4). I prepared also the filter you showed (150µH, 2µF+2.2µF), but this time I could not see the lower frequency components. I attached anyway, but it had little effect to the big spikes. I also tried inspecting these spikes a little deeper, but I think my simple scope just not good enough for that, images attached. Anyway, the frequency of them must be way above 1Mhz for sure.

Question: how come a lowpass filter leaves those so high frequency components untouched? Any thoughts?

So now I pick my screwdriver and see what's inside that power supply....

So today I took the supply out of the cabinet, put it on my desk, prepared a ~70Ω resistance as test load and checked it again. Surprisingly the spikes were there, but way smaller then they were before (image 1).

I suspected that some external noise is making fun of me, so moved back to the cabinet and checked again (image 2). The spikes are a little bigger, but still not that bad.

Now added little longer than half a meter wires to connect the resistance (the wires I originally used to connect the power supply output). This is when the spikes got a lot bigger (image 3). So now I completely shut down all the other units (other power supplies etc.) in the cabinet, in fact I switched off everything in the house. The spikes stayed, so they must come from the power supply.

Moved back again to the desk with all the attached cables. The spikes were still there (image 4). I prepared also the filter you showed (150µH, 2µF+2.2µF), but this time I could not see the lower frequency components. I attached anyway, but it had little effect to the big spikes. I also tried inspecting these spikes a little deeper, but I think my simple scope just not good enough for that, images attached. Anyway, the frequency of them must be way above 1Mhz for sure.

Question: how come a lowpass filter leaves those so high frequency components untouched? Any thoughts?

So now I pick my screwdriver and see what's inside that power supply....

Hello again,
Sorry to hear of your little success. As you seem to have proved the noise is on the cables so the longer the cables the more coupling paths you get. The only way to adress the issue is to remove the noise through filtering, damping it at its source, or remove the source of noise.

To answer your question theres a number of reasons why a filter will not work

1. The noise isn't differential but is instead Common Mode, therefore any differential filter will not be effective regardless of values. (Try putting a ferrite clamp around your cables instead as this is more effective for common mode issues) if that is effective this can be the solution or a common mode filter can be designed if the qty of wires is low.
2. The SRF of the inductor is below that of the ringing causing the problem. therefore the inductor is more capacitive than inductive at the problem frequency and the noise will just pass through it, this can be addressed by re-designing the filter to the problem frequency which will yeild a smaller inductance value with much higher SRF. For example change to change the filter to 55dB at 1.7Mhz change the filter values to 17uH (approx) and the C's to 550nF and it will knock 55db off if its differential. It will be less effective against low frequencies then though.

Unless i'm reading your plots wrong that ring looks to be approx 1.7Mhz so should be filterable using the previous filter values so if it isn't then its likely to be common mode.

 

Hello again,
Sorry to hear of your little success. As you seem to have proved the noise is on the cables so the longer the cables the more coupling paths you get. The only way to adress the issue is to remove the noise through filtering, damping it at its source, or remove the source of noise.

To answer your question theres a number of reasons why a filter will not work

1. The noise isn't differential but is instead Common Mode, therefore any differential filter will not be effective regardless of values. (Try putting a ferrite clamp around your cables instead as this is more effective for common mode issues) if that is effective this can be the solution or a common mode filter can be designed if the qty of wires is low.
2. The SRF of the inductor is below that of the ringing causing the problem. therefore the inductor is more capacitive than inductive at the problem frequency and the noise will just pass through it, this can be addressed by re-designing the filter to the problem frequency which will yeild a smaller inductance value with much higher SRF. For example change to change the filter to 55dB at 1.7Mhz change the filter values to 17uH (approx) and the C's to 550nF and it will knock 55db off if its differential. It will be less effective against low frequencies then though.

Unless i'm reading your plots wrong that ring looks to be approx 1.7Mhz so should be filterable using the previous filter values so if it isn't then its likely to be common mode.

Thank You for your suggestions, Marc. I am really interested in at least knowing the source of the noise, even if I cannot eliminate it directly. I sketched and attached the schematic of the supply, even I see that this is a really low-end stuff. Apart from this noise it would still do for me, so I will try to suppress this noise at least to a level that my 1-wire network will accept.

If you or anyone can see anything that could be easily improved to this schematic (or it is worth a try), I would appreciate the help.

 

Thank You for your suggestions, Marc. I am really interested in at least knowing the source of the noise, even if I cannot eliminate it directly. I sketched and attached the schematic of the supply, even I see that this is a really low-end stuff. Apart from this noise it would still do for me, so I will try to suppress this noise at least to a level that my 1-wire network will accept.

If you or anyone can see anything that could be easily improved to this schematic (or it is worth a try), I would appreciate the help.

The primary source of the noise will be the around the switching transformer nodes 2 & 3 and also around the core of the transformer itself. On a flyback converter the hightest currents at the switching frequency is normally on the secondary side which is dependent on the primary switching current and the transformer turns ratio. You could try putting a 470uH and 15uF LC Filter onto the output which should remove a lot of switching noise from the cables. Also try adding a 2.2mH and 10nf LC filter on the front end too, at the moment you have current paths for all the switching noise (rich in RF harmonics) going onto your cables which is the issue.

It looks like the designer has already attempted to damp the ringing through the use of snubbers. R1, C2 & D5 being the primary one and C5, D6 for the secondary. They've also used CY1 to control the amount of common mode emissions but there is a complete lack of fltering on the power lines for both the input or the output sides so both would really need to be addressed for any filtering to be effective. Its very difficult to do using voltage measurements really, you need to measure current ideally with a spectrum analyser and RF current probe.

 

The primary source of the noise will be the around the switching transformer nodes 2 & 3 and also around the core of the transformer itself. On a flyback converter the hightest currents at the switching frequency is normally on the secondary side which is dependent on the primary switching current and the transformer turns ratio. You could try putting a 470uH and 15uF LC Filter onto the output which should remove a lot of switching noise from the cables. Also try adding a 2.2mH and 10nf LC filter on the front end too, at the moment you have current paths for all the switching noise (rich in RF harmonics) going onto your cables which is the issue.

It looks like the designer has already attempted to damp the ringing through the use of snubbers. R1, C2 & D5 being the primary one and C5, D6 for the secondary. They've also used CY1 to control the amount of common mode emissions but there is a complete lack of fltering on the power lines for both the input or the output sides so both would really need to be addressed for any filtering to be effective. Its very difficult to do using voltage measurements really, you need to measure current ideally with a spectrum analyser and RF current probe.

You are definitely right, I am basically trying to solve this issue blindly, my basic instrument set is just not meant to study these problems. On the other hand I find this very interesting, so I started to dig myself into the theory of all this. In the meantime I decided to test all your suggestions in the hope that I can make a rough improvement for now.

So I put together a filter with 18µH and two 660nF capacitors. Well, the result is confusing, I would even say my scope is just a piece of crap, but I can reproduce it, so there must be something behind it. I put the filter right to the output of the supply just before the half meter cable. When I have the filter on, the scope shows much higher frequencies (at least like 5MHz) and interestingly the shape of the signal is very similar. When I disconnect the filter I get a much smoother shape with less transitions, like the ones I sent earlier.

I am sure that this is related to the limitations of my scope, but in what way?

Anyway I move forward to the common mode filters. I have two old computer PSU with full of coils, I think I need one with two coils on the same ferrite ring. Size-wise bigger is better, or I should choose a smaller one?

 

My guess is the change in the spikes depending where the PSU is , on the bench or in the cabinet, is down to how you have earthed the scope.

Are you using that 2 inch long Wire that comes with most scope probes ?
if so that is unlikely to give you a real measurement,
there is going to be just to much radiated "stuff" around.

You need to use the short leg on the scope probe, and measure across a load resistor.
Real PSU measurements are done using coax type connectors to avoid this


This noise you see, is that on the output of the PSU or the input ?
generically, capacitors are probably better at filtering than inductors in the output case.

have you looked at the one wire circuit ,
what does the signal look like on that ?
Again short scope earth pin.
Its typically a very resilient protocol to noise,
The chips have built in low pass circuits to disregard the noise on the data ,
and assuming your self powering, then your signal would have to dip a long way to make a difference.

show us a picture of the one wire please.

 

My guess is the change in the spikes depending where the PSU is , on the bench or in the cabinet, is down to how you have earthed the scope.

This is also I suspected. I literally just moved to PSU with attached resistor there. First I thought that the noise is coming from the mains line, so I checked each of the 3 mains lines, but I got the same results. Then I attached the wires between the resistor and the PSU, the spikes became much greater. So I thought they were like antennas, so I shut down everything except the mains power to the PSU. And nothing changed. Unless I have the noise on all 3 mains lines or have something "in the air", the spikes must be coming from the PSU.

(By the way, we have a nuclear power plant in 2-3km distance, but then why I do not see the noise on the old PSU with the same probe and setup?..)

Are you using that 2 inch long Wire that comes with most scope probes ?
if so that is unlikely to give you a real measurement,
there is going to be just to much radiated "stuff" around.

You need to use the short leg on the scope probe, and measure across a load resistor.
Real PSU measurements are done using coax type connectors to avoid this

Yes, I am using that (around 3 inches) with crocodile clip on it and always measure directy on the 70Ω load. I guess you mean that small pin that can be attached right above the center probe tip and results in another pin like 5mm off the center tip. I do not have one and even I had I could not touch the resistor (resistor network made of 3 resistors to handle the heat) properly like that. It makes sense, but I do not know how that is generally used.

Sure my measurement is not professional at all. All I would like to get rid of or at least make a quarter size of those spikes. The real shape of them or the absolute magnitude of them is not something I am really interested in. (Though getting more and more curious.)

This noise you see, is that on the output of the PSU or the input ?

You mean I could check the line input? I do not dare to measure high voltages, I do not think my probe and scope could handle that.

generically, capacitors are probably better at filtering than inductors in the output case.

Any specific idea? It is welcome because whatever I do, the spikes are healthier than ever.

have you looked at the one wire circuit ,
what does the signal look like on that ?
Again short scope earth pin.
Its typically a very resilient protocol to noise,
The chips have built in low pass circuits to disregard the noise on the data ,
and assuming your self powering, then your signal would have to dip a long way to make a difference.

show us a picture of the one wire please.

This is a good idea. I will do that tomorrow. From previous experience I know that at this total network length of 80-100m they are sensitive to noises. For this reason I used low capacitance ProfiBus cable for the purpose. They have been working well for years, but the new 12V PSU is immediately "detected", even it is not running from that.

Thank you for your response and pointing this out.

 

OK, I think I finished with experimenting blindly. For the sake of fun I sketched the current setup. It looks like I could suppress the spike amplitude to about 70% of the original size. Confusingly making a measurement right at the PSU output gives actually lower amplitudes. Probably I am missing very basic concepts here, I have no idea how it is possible.

Anyway, my conclusion is that for studying such situation one needs good equipment. Once the noise is correctly diagnosed (common-mode or differential, frequency etc.) that is when any design procedure can be started.

Tomorrow I will check the 1-wire network to see why it is failing with this PSU.

 

have look here
https://www.keysight.com/upload/cmc_upload/All/Probing-6-Hints-Scopes-Dec12-2007-webcast.pdf

Ok its trying to sell HP stuff, but look on page 9
you have two pictures side by side of the same signal , one with a short earth , one with the long clip you are using.
Notice how "crisp" the one with the short earth is and how many ups and downs the long earth has,

https://www.analog.com/en/technical-articles/short-ground-leads-make-better-scope-photos.html

This is what you are suffering,
your equipment is not at fault, a USD 100K scope, probed as you are will show just the same.

 

Your seeing resonances,

think I finished with experimenting blindly. For the sake of fun I sketched the current setup. It looks like I could suppress the spike amplitude to about 70% of the original size. Confusingly making a measurement right at the PSU output gives actually lower amplitudes. Probably I am missing very basic concepts here, I have no idea how it is possible.

OK, I think I finished with experimenting blindly. For the sake of fun I sketched the current setup. It looks like I could suppress the spike amplitude to about 70% of the original size. Confusingly making a measurement right at the PSU output gives actually lower amplitudes. Probably I am missing very basic concepts here, I have no idea how it is possible.

Anyway, my conclusion is that for studying such situation one needs good equipment. Once the noise is correctly diagnosed (common-mode or differential, frequency etc.) that is when any design procedure can be started.

Tomorrow I will check the 1-wire network to see why it is failing with this PSU.

OK, I think I finished with experimenting blindly. For the sake of fun I sketched the current setup. It looks like I could suppress the spike amplitude to about 70% of the original size. Confusingly making a measurement right at the PSU output gives actually lower amplitudes. Probably I am missing very basic concepts here, I have no idea how it is possible.

Anyway, my conclusion is that for studying such situation one needs good equipment. Once the noise is correctly diagnosed (common-mode or differential, frequency etc.) that is when any design procedure can be started.

Tomorrow I will check the 1-wire network to see why it is failing with this PSU.

Unless your scope is isolates that will be part of your issue. As soon as you connect the scope earth you are providing a path to Gnd for the noise to flow via your scope. As a side note you may want to try some Y capacitors to earth from the left side of you common mode choke lines which will reroute the RF currents to earth rather than your wires. A CM choke with no Y caps will probably only give about 10dB per decade.

 

have look here
https://www.keysight.com/upload/cmc_upload/All/Probing-6-Hints-Scopes-Dec12-2007-webcast.pdf

Ok its trying to sell HP stuff, but look on page 9
you have two pictures side by side of the same signal , one with a short earth , one with the long clip you are using.
Notice how "crisp" the one with the short earth is and how many ups and downs the long earth has,

https://www.analog.com/en/technical-articles/short-ground-leads-make-better-scope-photos.html

This is what you are suffering,
your equipment is not at fault, a USD 100K scope, probed as you are will show just the same.

Thanks Andrew the links, I read them. Following the topic I found a video at the Texas Instruments youtube channel where the guy talks about just this and is checking PSU noises (

). Interestingly he just shows the same kind of spike I have measuring it at above 140MHz!!!

I think the main thing here is that I am measuring the spikes at more than 1Vpp while it looks to me that the probing technique is getting important when you are checking noise in the 10mV range.

Anyway, I took a Cat5e cable, stripped the coating of one of the wires, coiled it on my screwdriver and it now perfectly fits on my probe which has a little larger diameter. Even I am surprised how stable it is. So I made the checks using that.

See attached the images of my 1-wire network with the old and new 12V supply connected. I checked it with the same settings I check the 12V PSU output (200mV/div), but I switched my probe also to x1 mode as well (20mV/div).

I see the ripple and also some spikes, this is probably coming from the 5V supply of the network, but the noise stays in 140-150mVpp. But then when I switch to the new 12V PSU, I can see those narrow spikes on the network. They are not that giant like at the output of the 12V PSU, but the noise on the network now gets to 400mVpp and it seems to be enough to make the communication fail.

I checked again the 12V PSU output, giant 1.2Vpp noises.

 

Well done,

they look more like reality. its amazing how many people loose the short earth clip on the scope..

The scale looks at 2.72 v / division, I hate these new scopes, what happened to good old easy numbers like 2V per division , much easier to read the display. And your sampling at 5.22 Ms/s !!!!

So I'd say the noise is coming through the 5V PSU to the one wire,
What circuit do you have droving the one wire ?

Have you looked at the 5v Supply, it strikes me that that might be letting a lot of noise through ,
may be we can look at that end of things.

 

Your seeing resonances,




Unless your scope is isolates that will be part of your issue. As soon as you connect the scope earth you are providing a path to Gnd for the noise to flow via your scope. As a side note you may want to try some Y capacitors to earth from the left side of you common mode choke lines which will reroute the RF currents to earth rather than your wires. A CM choke with no Y caps will probably only give about 10dB per decade.

Marc, if you didn't send this "side note", I would have given it up. I was so out of ideas. As I started looking at these Y capacitors (since I have never heard about them before) and you saying I need two of them on the choke, I remembered that I have seen a similar setup in a microwave oven I have taken apart long ago. I knew it was a mains filter, but of course I had no idea how it worked. Thanks to the recent days of fighting with those bastard spikes and your side note, the whole thing came together. I attached an image of it.

So I put it now turned around on the output of the PSU (so I connected the OUT connections to the outputs) and... IT WORKED! Attached an image, the spikes were still there, but with a maximum of 400mVpp (instead of 1.2Vpp like before). So it was really something. Quickly put the whole thing (with the filter hanging on the wires) into the cabinet, replaced the old PSU, and tadaam! No communication failures!

I checked with the scope (at the output of the filter) and found the attached image. Weird, I see what looks like a ripple, I have never seen that before. The spikes are also larger than on the bench. Then I switched on a 12V unit (so increased the load), and I got the same level of noise than on the bench. So it looks like the load matters.

Nevertheless I checked the 1-wire network and it really looks good, like with the old PSU. In fact not exactly. I still see smaller spikes "intermittently", but probably that causes no trouble.

So it is not a 100% solution, but I originally wanted an improvement only, and now I have done it. I believe the frequency of these spikes are way above 10MHz or so (like for the guy in the video I sent before). This is the reason why my poor little scope with 20MHz bandwidth just takes some samples of it and the spike is sometimes smaller or bigger or seemingly not there at all. For any further improvement I would need a better scope, look right in the eye of the bastard and then design a choke for exactly this purpose.

All in all, thank You for all your suggestions that led me to the right track!

(Last questions: as I understand the Y capacitor is Y capacitor because of its safety level. Would any generic 4.7nF low voltage ceramic capacitor work in place in my case? The big yellow capacitor 220nF/~250V and the parallel 1MΩ resistor on the mains filter makes any sense in my case?)

 

Well done,

they look more like reality. its amazing how many people loose the short earth clip on the scope..

The scale looks at 2.72 v / division, I hate these new scopes, what happened to good old easy numbers like 2V per division , much easier to read the display. And your sampling at 5.22 Ms/s !!!!

So I'd say the noise is coming through the 5V PSU to the one wire,
What circuit do you have droving the one wire ?

Have you looked at the 5v Supply, it strikes me that that might be letting a lot of noise through ,
may be we can look at that end of things.

Andrew, this is also something I was thinking about after taking a look at the 1-wire signal. But since 1-wire with this 5V supply has been working great for now almost 4 years and the fault is introduced clearly by attaching the new 12V supply I kept on trying to fix that.

By the way the 5V supply is a wall-plug supply, so probably the same poor quality like the new 12V PSU... No, at least not that bad.

The scope is showing a DC offset when I am using my probe in x10 mode. So that 2.72V is supposed to be the DC level, but is clearly wrong, because in case of the 1-wire it should show 0.5V in x10 mode. When I switch back to x1 mode with the probe, then the DC level is correct, on my images it is shown 4.94V, which is more or less the reality. I do not know the reason for this, but I found that the AC levels are good, so I did not care.

In fact this scope is not a scope, but a logic analyzer with a very basic scope functionality. It has a 20MHz bandwidth and the minimum voltage setting is 20mV/div. So it is very good to see "something", but is clearly not for making measurements.

As I mentioned in my previous comment I think I will stop now, it works. Not perfect, but works. It was a great learning curve and I am really thinking of buying a real scope somewhere in the future and I enjoyed making experiments.

Thank You also for your comments.

 

Has feeling of the ground between the two power supplies not being "hard" .
so the 5v and the 12V grounds Wander,

2.72 offset is well bad.... !!

IMHO, and what I try to teach, the art of engineering is seeing some thing that look strange, coming to a conclusion as to what it is caused by , then deciding is it ok or not.

management is , it works , ship it.,.,,,, I'm not a manager..... !!

The chips on the one wire do not see the AC levels, they see the difference between their ground and the one wire,
You need to sort out the scope probe,

But why are you not looking at the one wire bus with the logic analyser ?

Anyway

 

Marc, if you didn't send this "side note", I would have given it up. I was so out of ideas. As I started looking at these Y capacitors (since I have never heard about them before) and you saying I need two of them on the choke, I remembered that I have seen a similar setup in a microwave oven I have taken apart long ago. I knew it was a mains filter, but of course I had no idea how it worked. Thanks to the recent days of fighting with those bastard spikes and your side note, the whole thing came together. I attached an image of it.

So I put it now turned around on the output of the PSU (so I connected the OUT connections to the outputs) and... IT WORKED! Attached an image, the spikes were still there, but with a maximum of 400mVpp (instead of 1.2Vpp like before). So it was really something. Quickly put the whole thing (with the filter hanging on the wires) into the cabinet, replaced the old PSU, and tadaam! No communication failures!

I checked with the scope (at the output of the filter) and found the attached image. Weird, I see what looks like a ripple, I have never seen that before. The spikes are also larger than on the bench. Then I switched on a 12V unit (so increased the load), and I got the same level of noise than on the bench. So it looks like the load matters.

Nevertheless I checked the 1-wire network and it really looks good, like with the old PSU. In fact not exactly. I still see smaller spikes "intermittently", but probably that causes no trouble.

So it is not a 100% solution, but I originally wanted an improvement only, and now I have done it. I believe the frequency of these spikes are way above 10MHz or so (like for the guy in the video I sent before). This is the reason why my poor little scope with 20MHz bandwidth just takes some samples of it and the spike is sometimes smaller or bigger or seemingly not there at all. For any further improvement I would need a better scope, look right in the eye of the bastard and then design a choke for exactly this purpose.

All in all, thank You for all your suggestions that led me to the right track!

(Last questions: as I understand the Y capacitor is Y capacitor because of its safety level. Would any generic 4.7nF low voltage ceramic capacitor work in place in my case? The big yellow capacitor 220nF/~250V and the parallel 1MΩ resistor on the mains filter makes any sense in my case?)

Hi,
Glad some of the input helped. The parts your using are definately not ideal but i'm glad you reached a suitable solution.

The load dependant ripple your seeing is probably caused by low frequency resonances caused by the filtering you've added, its not optimised to your load nor does it have any damping which would minimise these oscillations. You could also try rotating the CM filter too with the caps the other side as depeding on the current flow the filtering can change substantially. The CM choke looks quite large so i think its a large CM inductance for relatively low frequencies.

Functionally a general purpose capacitor does the same work is the same as a safety capacitor so it would work, safety capacitors have self healing capability which make them suitable for bridging an isolation barrier so the answer is it depends. If failure doesn't represent a shock hazard you'll be ok to use a lower spec part on the ouput of a PSU but its a definate no for the primary side.

The yellow capacitor accross the CM choke is a differential filter capacitor. The CM Choke has some leakage inductance which means is has some differential inductance as well as its primary CM inductance. The yellow capacitor forms a LC filter with this leakage inductance to prodive some attenuation it will still provide some filtering so as your happy you can leave it there.. The Resistor is just there to discharge the capacitor as in the original application it would pose a shock hazard when charge to 300V when you pull the plug.