MUFF WIGGLER Forum Index
 FAQ & Terms of UseFAQ & Terms Of Use   Wiggler RadioMW Radio   Muff Wiggler TwitterTwitter   Support the site @ PatreonPatreon 
 SearchSearch   RegisterSign up   Log inLog in 
WIGGLING 'LITE' IN GUEST MODE

Bad Citizens on the Power Bus
MUFF WIGGLER Forum Index -> Music Tech DIY  
Author Bad Citizens on the Power Bus
jakobprogsch
I'm currently working on/experimenting with my own linear power solution. This is what the current prototype looks like:


As a result I'm doing a lot of measuring. So far It mostly went as expected. I scoped ripple measurements during design etc. with power resistors and adjustable constant current sinks and now with the prototype rack also when it is fully populated with modules. Generally ripple stays around 2mV p-p it seems. This is what it looks like with >200 HP of various (analog) modules:



This was measured at one of the "far" points of the power distribution star. But it looks fairly uniform across the entire distribution system. Before putting in the modules I connected a 1A load at on of the farthest connectors and measured 2-3mV drop on the 0V rail (when only one rail is loaded) and 12-14mV on the +-12V rails. This is not including the ribbon cable which by itself drops about 4x that.

Anyway. there are some modules however that manage to introduce significant ripple over the whole distribution system just by themselves. For example a self oscillating filter by a popular brand:



So far I identified three modules with similar behavior. Two of them have clock generators that seem to cause this and one of them is the filter shown above. Here is a closer look at a shorter time interval:



Since this is visible over the whole distribution system including the copper bars at the center of it, which the regulators (LT1185 based) connect to via remote sensing, it would appear this is "caused" by the regulators. Or to be more precise by the apparently very sharp load transients (as low as 100ns rise time apparently) caused by the offending modules.

I also had noticed before in a different case with weaker power that the above filter has a tendency to bleed through high gain modules like wavefolders. I haven't heard this in this case but it shows up on the scopes.

I don't actually think I can really blame the power supply or distribution here. I haven't figured out a clever way to measure the magnitude of the load transient itself (suggestions welcome). But in the very least it doesn't seem too surprising that a linear regulator will not completely "keep up" with a <1uS load transient. The regulator data sheet shows a 100mA and <100ns load transient causing a 100mV voltage spike.

I conducted an experiment where I put a small power filter circuit consisting of ferrite beads, 100nF ceramic and 10uF electrolytic caps before the module. This smoothed over the transients slightly but unsurprisingly didn't "fix" them since this is presumably the same filtering/bypassing already found on the modules themselves. It did however get rid of most of the ringing seen above.

So overall Im tempted to blame the module design here. The question is, how do I handle this? put all the "nasty" modules on their own distribution+PSU away from all the finicky ones? Put them on their own busboard with very heavy handed filtering? Rely on properly designed modules to have sufficient PSRR? Or just banish the bad ones alltogether? Or is this still indicative that there is something wrong with my PSU/regulator design?
Kent
I moved this into the DIY section where it is more likely to be seen by suitable persons.
Mungo
jakobprogsch wrote:
I conducted an experiment where I put a small power filter circuit consisting of ferrite beads, 100nF ceramic and 10uF electrolytic caps before the module. This smoothed over the transients slightly but unsurprisingly didn't "fix" them since this is presumably the same filtering/bypassing already found on the modules themselves. It did however get rid of most of the ringing seen above.
Which side of the filters did you measure the changes at?
Most ferrite beads are only going to make a difference above 1-10MHz (the 100ns you have measured), changes below that frequency need large capacitances and/or inductances. Ferrite beads are far from ideal and have very loose specifications:
https://www.analog.com/media/en/technical-documentation/application-no tes/AN-1368.pdf

jakobprogsch wrote:
So overall Im tempted to blame the module design here. The question is, how do I handle this? put all the "nasty" modules on their own distribution+PSU away from all the finicky ones? Put them on their own busboard with very heavy handed filtering? Rely on properly designed modules to have sufficient PSRR? Or just banish the bad ones alltogether? Or is this still indicative that there is something wrong with my PSU/regulator design?
Isolating with filtering is one way, but the resistance in the wiring and along the bus boards form part of the filter too. Measure how much noise there is on one of your bus boards compared to the other, and how much is at the output of the power supply. Several mV or total noise is already quite low for such a large systems and getting lower is difficult.
jakobprogsch
Mungo wrote:
Which side of the filters did you measure the changes at?
Most ferrite beads are only going to make a difference above 1-10MHz (the 100ns you have measured), changes below that frequency need large capacitances and/or inductances. Ferrite beads are far from ideal and have very loose specifications:
https://www.analog.com/media/en/technical-documentation/application-no tes/AN-1368.pdf

I measured on the busboard side. These effects are visible on both busboards. I mostly conducted the filter experiment to figure out if there is maybe a defective bypass/filter cap or so on the module itself. In that case I would have expected more of an improvement. I guess the ferrite bead thing is more due to ease of adding them?

Mungo wrote:

Isolating with filtering is one way, but the resistance in the wiring and along the bus boards form part of the filter too. Measure how much noise there is on one of your bus boards compared to the other, and how much is at the output of the power supply. Several mV or total noise is already quite low for such a large systems and getting lower is difficult.

The scope shots from both busboards and the copper bars I used at the center of the "star" all look pretty much the same. I think the distribution is reasonably low impedance and all the modules I used are fairly balanced loads anyway.

To get a second point of reference I attached the offending module to a small MW DKE15A-12 (+-12V 625mA DC-DC converter) based board. That shows 10mV (~200kHz?) of ripple just by itself and the module can add an additional 40mV ripple on top of that. So I guess my own solution does acceptably by comparison. Especially considering the SMPS was only loaded with a single module and the rack was loaded with two full rows of modules.

On the other hand, now that I measured it and even though I can't hear it, it is of course annoying to me that a single badly behaved module can introduce as much noise on my carefully designed linear power solution as a cheap SMPS. But I don't see an obvious solution or flaw on the PSU side. So I guess I'll either have to wait until I'm over the knowledge and don't care anymore or start quarantining the modules.

Thanks.
jakobprogsch
So I rigged up a primitive current sensing device out of a TL074 and some 1Ohm resistors as shunts. I figured since I'm measuring modules with ~100ma draw the ~0.1V drop is acceptable for a measurement device.

This is what it looks like (10mV means 1mA of current):



The flanks are not as steep as I expected (Edit: I guess connecting a 1Ohm shunt resistor will also "improve" the filtering effect in conjunction with the modules bypass caps causing the Heisentransients to smooth over). But the magnitude surprises me. That is about 50mA of load variation... and worst of all asymmetrically. So this will dump the ripple current into the 0V rail... Considering this is a filter in self resonance I'm guessing this is the effect of a "generously" dimensioned Zener clamp in the feedback path? The spikes also significantly exceed the specified current draw of the module which is otherwise consistent with the "non spiked" current.
Mungo
jakobprogsch wrote:
So I rigged up a primitive current sensing device out of a TL074 and some 1Ohm resistors as shunts. I figured since I'm measuring modules with ~100ma draw the ~0.1V drop is acceptable for a measurement device.

This is what it looks like (10mV means 1mA of current):
Good to do the extra work and measure that, now you can estimate the impedance of the power system at a little under 0.1 ohm. Very typical for that sort of design and close to the specification of the regulator you are using. Improving from there would be replacing the supplies and busboards and distribution (everything!), and/or adding lots of capacitance around the system.
jakobprogsch
Mungo wrote:

Good to do the extra work and measure that, now you can estimate the impedance of the power system at a little under 0.1 ohm. Very typical for that sort of design and close to the specification of the regulator you are using. Improving from there would be replacing the supplies and busboards and distribution (everything!), and/or adding lots of capacitance around the system.


But... but... these ARE the replacements sad banana . The supply part anyway. I consider the Deopfer busboards a temporary solution until I get to engineering something for those myself. I was thinking connector boards directly screwed to a set of horizontal running copper bars. Anyway given this isn't local to a specific point in the distribution system I'm pretty sure I'm mostly seeing the transient response behavior of the regulator.

Here is a more interesting scope shot:



I tried to isolate the effects correlated to these transients by using the average mode on the oscilloscope.

I guess this does mean that the regulators have about the mentioned 0.1 Ohm output impedance at somewhere in the 50-100kHz range? Which is reasonably consistent with the output impedance chart in the data sheet (page 4 bottom right): https://www.analog.com/media/en/technical-documentation/data-sheets/11 85ff.pdf
I should probably do some "science" on the choice of output capacitor ESR as I'm currently at the top of the suggested range.

Either way I'm struggling to assess if this is good or bad compared to alternatives. Not many regulator datasheets contain output impedance specs.

The practical performance I measured on for example the MN skiff is worse though. So I'm not yet ready to accept that what I have is inherently bad and needs to be thrown out.
mskala
What amount of ripple is your goal, and why?
jakobprogsch
mskala wrote:
What amount of ripple is your goal, and why?

As low as I can manage and because I want to. Trying to engineer this myself is my way of dealing with feeling unable to assess who to believe and what to do with regards to modular power. So I'm "doing all my own science" to either arrive at a solution I trust because I understand it or in the very least have enough understanding to make decisions about what to buy.

Currently it seems like ~2mV ripple are achievable for this kind of configuration with 208HP of well behaved modules. Now I'm trying to figure out where I should try to solve the issue of a less well behaved module as seen above. Incidentally this also provides me with good data about what to look out for to make my own module designs well behaved.

So uh... the journey is the destination I guess?
mskala
jakobprogsch wrote:
mskala wrote:
What amount of ripple is your goal, and why?

As low as I can manage and because I want to.


Well, then you'll never be finished, because there will always be more you can do.

jakobprogsch wrote:
Now I'm trying to figure out where I should try to solve the issue of a less well behaved module as seen above.


It seems to me that whether this is an "issue" at all, has not been established. There's no indication that the ripple you are measuring has any consequences other than being visible on the scope, and I suspect that it doesn't. But if your goal really is to strive forever without a target, then I guess you're on the right track after all.
jakobprogsch
mskala wrote:
jakobprogsch wrote:
mskala wrote:
What amount of ripple is your goal, and why?

As low as I can manage and because I want to.


Well, then you'll never be finished, because there will always be more you can do.


So? That is also being overly literal though. Were it not for this (or these three) modules I would have probably stopped there for now. But at the moment the situation was:

    * PSU with 2x1.5A ohmic loads: <2mV ripple
    * PSU with 200hp of modules drawing around 2x1A: ~2mV ripple
    * PSU with the same modules + this one filter: up to 10mV ripple

So this seemed worth investigating. And I learned a bunch of useful things here already. Anyway, I hope this doesn't turn into a discussion where my motivations are being put on trial.
mskala
Seems to me that the next step, and probably necessary to get any significant decrease in ripple beyond where you're currently at, would be to give every module its own individual filter and regulator, like the old S-100 bus computers had. That'd make the bus board more complicated but wouldn't be completely unreasonable. Having a very low impedance for the 0V would also be worthwhile for keeping the actual audio clean, but that is a separate question from ripple on the power lines, and as others here will no doubt point out, Eurorack limits how well we can ever do on that score by putting both power and signal reference on the same 0V bus.
search64
I see you summoned Graham Hinton to your thread. It's a miracle he isn't already in here.
jorg
Because you seem to have a bunch of "polite" modules and one "rude" one, I'd suggest a filter for the rude module. Make something you could slip inline, in series with the ribbon cable for that module. A nice RC filter would probably work; make it a couple of stages, and remember that you are protecting the buss, not the module.
Pelsea
How does that filter behave when it is not self oscillating? What about when it just begins to oscillate?
Mungo
jakobprogsch wrote:
Mungo wrote:

Good to do the extra work and measure that, now you can estimate the impedance of the power system at a little under 0.1 ohm. Very typical for that sort of design and close to the specification of the regulator you are using. Improving from there would be replacing the supplies and busboards and distribution (everything!), and/or adding lots of capacitance around the system.
I guess this does mean that the regulators have about the mentioned 0.1 Ohm output impedance at somewhere in the 50-100kHz range? Which is reasonably consistent with the output impedance chart in the data sheet (page 4 bottom right): https://www.analog.com/media/en/technical-documentation/data-sheets/11 85ff.pdf
I should probably do some "science" on the choice of output capacitor ESR as I'm currently at the top of the suggested range.

Either way I'm struggling to assess if this is good or bad compared to alternatives. Not many regulator datasheets contain output impedance specs.
Without details of the power supply you have its not possible to provide much more information. Low ESR capacitors on the output may help, or they could make it worse if you don't design it to match the regulator:
https://www.tnt-audio.com/clinica/regulators2_impedance1_e.html
Noise on the power rails is a very relative "problem" and you could spend lots of time for little or no improvement. Measure the actual useful result, signal to noise ratio of the modules in use.
jakobprogsch
Mungo wrote:
Without details of the power supply you have its not possible to provide much more information. Low ESR capacitors on the output may help, or they could make it worse if you don't design it to match the regulator:
https://www.tnt-audio.com/clinica/regulators2_impedance1_e.html
Noise on the power rails is a very relative "problem" and you could spend lots of time for little or no improvement. Measure the actual useful result, signal to noise ratio of the modules in use.

Thanks for that link.

The PSU is of my own design using two identical regulator boards using a LT1185. The short version is: 12V 80VA toroid -> shottky rectifiers -> 2x10mF smoothing capacitors -> LT1185 as per data sheet with remote sensing. By that capacitor I meant the output capacitor the regulator needs for stability. My current choice (10uF tantal with 2Ohm max ESR spec) was towards the upper end of the recommended ESR range so I figured I should explore the range more. A quick experiment with a 22uF 0.06ohm polymer capacitor showed no change in terms of magnitude of the transient response except that it added some ringing. The results are also fairly consistent with a SPICE simulation. Playing some more within SPICE suggested that using 100uF/0.2Ohm would reduce the output impedance above ~10kHz but in practice that didn't make much of a difference. Sticking some capacitors on the busboard connectors did though. Not massively but it about halved the transient induced ripple from the filter.

I'm now fairly convinced that the PSU part is ok. I'll probably dig out my other prototypes from when I tested various other regulators (1085, LT1764) to see if those behave differently. I didn't focus on transient response/output impedance before so that seems worth revisiting.

Other than that I'll move on to experimenting with the distribution aspects for now.

Thanks for everyone's input.
MUFF WIGGLER Forum Index -> Music Tech DIY  
Page 1 of 1
Powered by phpBB © phpBB Group