Switching PSU filtering

[Neperdun]

Oh, OK, thanks again!

Did You solve this problem? It seems like I meet the same...

 

[Ricardus]

Did You solve this problem? It seems like I meet the same...

Whoever I was conversing with gave me the gerbers to order some PCBs to build a filter. I built one for another project that had similar voltage rails.

 

[Matt Syson]

It is important to remember that all switcher modules are a 'component' and NOT a finished power supply for audio gear. Output noise at around 100 millivolts might sound good on paper but compared to a 150 microvolts of even a half decent linear supply using a 'chip' or other circuitry it is consideably worse even if the 60 -200KHZ switching frequency is not immediately obvious there is a decent risk it will appear somewhere.

 

[amplexus]

I’ve moved to switchers for all new builds. I just built a replacement PSU for a Ramsa desk a couple weeks ago- 6 Rails, 1100W.

But even for preamp racking etc…

I’ve got two boards I did up that are simple CLC filters for multi rail projects- one os for basic 3 rail stuff (2 audio + phantom) up to 4 amps, and a larger one is for 4 rails, configurable however one needs, up to 18A per rail.

I haven’t used linear supplies in a build or replacement in… 3 years? Never had an issue. That Ramsa desk actually measures quieter than it did before (tho to be fair the OEM psu unit was hosed).

Havent had a single issue with noise or function. I don’t even bother with multi output modules usually. I just buy single outputs and roll my own grounding as needed.

 

[jasonallenh]

I don’t even bother with multi output modules usually. I just buy single outputs and roll my own grounding as needed.

For a console PSU, I imagine that works great. Might still go multi-output switcher or linear for something like a GSSL, though the thought of cramming all these switchers into a 3U GSSL is pretty funny

 

[living sounds]

For a console PSU, I imagine that works great. Might still go multi-output switcher or linear for something like a GSSL, though the thought of cramming all these switchers into a 3U GSSL is pretty funny

It's easy to get a GSSL hum-free with a linear PSU mounted off-board with a CRC filter in front of it. And following Rane grounding rules for the balanced connectors. I wouldn't bother with a SMPS in this case.

 

[jasonallenh]

It's easy to get a GSSL hum-free with a linear PSU mounted off-board with a CRC filter in front of it. And following Rane grounding rules for the balanced connectors. I wouldn't bother with a SMPS in this case.

Exactly what I'm saying. An SMPS per rail is somewhat excessive unless it's a bigger use case

 

[thor.zmt]

I built an 8 channel API-312 clone with a linear PSU inside the same chassis. It worked great but I’m starting to notice EMI in the channels closest to the PSU. I would like to swap out the toroidal for one of these:

3-Rail Switch Mode Power Supply +/-16v and +48v

The filter recommended on the page that sells this PSU has ~160Hz cutoff with a 30dB peak at 160Hz.

Adding some resistance to the choke, say 0.5 Ohm (which could be the DCR of the Coke) will critically damp this resonance.

An LC lowpass shows 40dB/decade roll-off, so at 16kHz the filter will knock the noise down to 1/10,000, which should be fine.

However, you might want to make sure the PSU has sufficient loading, low load causes most SMPS to enter cycle skipping or so-called green mode, which causes significant audio band noise, which the recommended filter may not attenuate enough.

Increasing inductor and/or capacitor values has limits, an active filter circuit may be preferred either stand alone or with an LC filter, as it can offer much much greater noise suppression at audio frequencies.

Thor

 

[ruffrecords]

The filter recommended on the page that sells this PSU has ~160Hz cutoff with a 30dB peak at 160Hz.

Adding some resistance to the choke, say 0.5 Ohm (which could be the DCR of the Coke) will critically damp this resonance.

An LC lowpass shows 40dB/decade roll-off, so at 16kHz the filter will knock the noise down to 1/10,000, which should be fine.


Thor

I am curious as to how the actual load affects the filter response and component value choices. In the example the load can be as much as 2 amps at 16V which is equivalent to an 8 ohm load. For a tube HT supply, the values might be 300V and 300mA which is equivalent to a 1K load.

Cheers

Ian

 

[Tubetec]

Amplexus ,
I was just going to ask about cooling , but I spotted the fan at the side ,
do you have a perforated lid on the box too ?

 

[sahib]

I am curious as to how the actual load affects the filter response and component value choices. In the example the load can be as much as 2 amps at 16V which is equivalent to an 8 ohm load. For a tube HT supply, the values might be 300V and 300mA which is equivalent to a 1K load.

Cheers

Ian

Ian, your reasoning does not apply as the output voltage is fixed. You would not be connecting a load that demands 200V operation to a power supply that provides 16V.

However, altering the load resistance/impedance will alter the output current and this will change the filter inductor/capacitor size.

 

[ruffrecords]

Ian, your reasoning does not apply as the output voltage is fixed. You would not be connecting a load that demands 200V operation to a power supply that provides 16V.

However, altering the load resistance/impedance will alter the output current and this will change the filter inductor/capacitor size.

I think we are talking at cross purposes. What I meant was that the components of the example filter were chosen for a specific output voltage and current. What I wanted to know was how these values would change for a different SMPS which could output 300V at 300mA.

Cheers

Ian

 

[sahib]

Apologies for misunderstanding.

Although the mathematics of it is treated in varying ways depending on the author, but if I can give an example from the book Switching Power Supply Design by Abraham I. Pressman, for the forward and push-pull topology the output inductor size is given by L= (0.5 x Vout x T ) / Iout . Here L is in Henry

So, the equation seems to answer to your question that the higher the output voltage the higher the L value.

I am in a hurry to go to the office. I'll look into it further but no doubt Thor will respond further.

 

[thor.zmt]

I am curious as to how the actual load affects the filter response and component value choices.

Well, THAT DEPENDS.

A lot of audio amplification circuits act as constant current or modulated current sinks.

So the AC load impedance onto the rail is very high. Thus my comment on damping LC filters/

In the example the load can be as much as 2 amps at 16V which is equivalent to an 8 ohm load.

NOPE, it is NOT equivalent to an 8 Ohm resistor. Unless you are feeding a resistor (e.g. tube heater), of course.

For a tube HT supply, the values might be 300V and 300mA which is equivalent to a 1K load.

Same comment, it is not. The AC Impedance of your rail is complex, series decoupling resistors, decoupling capacitors (if present) and then the audio circuits impedance (Anode load + Anode Impedance per stage.

Milkmaid calculations are of limited utility here.

Thor

 

[thor.zmt]

I think we are talking at cross purposes. What I meant was that the components of the example filter were chosen for a specific output voltage and current.

Yes and no, I suspect they were chosen on sufficient suppression of switching residuals, voltage and current are more related to practicality of available components.

What I wanted to know was how these values would change for a different SMPS which could output 300V at 300mA.

The same values will work the same. Same resonance frequency and roll-off, DC current and voltage do not change that.

If you increase L and reduce C to cover the issues of 2,200uF/350V Capacitors you alter the Q of the resonance. If we go for 4.7mH & 220uF without any DCR in the resonant circuit we get an over 35dB peak in the response at 156 Hz.

Ergo, the actual LC filter should be calculated based on the known lowest switching frequency of the SMPS to ensure sufficient attenuation of the switching residue and LC resonances must be considered carefully.

Capacitor ESR and Choke DCR form the worst case damping of the LC resonant tank. A resistive load will apply damping, but real power supply loads are rarely resistors.

With something like 30dB peaking at the PSU voltage, induced LC spikes during mains transients etc can easily blow out semiconductors. Poorly considered LC filtering is often the reason for inexplicable poor reliability of circuits, that work well most of the time and then bomb without apparent cause.

Thor

 

[ruffrecords]

Yes and no, I suspect they were chosen on sufficient suppression of switching residuals, voltage and current are more related to practicality of available components.



The same values will work the same. Same resonance frequency and roll-off, DC current and voltage do not change that.

OK, help me out here. This is basically an RLC circuit. Various elements make up the R including capacitor ESR and inductor DCR. But another element of it must be the resistive component of load itself. How is that accounted for in the calculation of Q?

Cheers

Ian

 

[thor.zmt]

OK, help me out here. This is basically an RLC circuit. Various elements make up the R including capacitor ESR and inductor DCR. But another element of it must be the resistive component of load itself. How is that accounted for in the calculation of Q?

How do you know the resistive component of the power supply rail of the supplied circuit?

Obviously it is not Vout / Iout as the load is an active circuit, not a resistor in most cases. And if it is a resistor, you don't really care about the resonance, do you?

The worst case PSU Rail Impedance is near infinity in parallel with multiple parallel decoupling capacitor. So unless you you know the AC impedance of the PSU rail, it is best to assume the worst case.

Thus load does not commonly factor in the LC filter Q, cut-off frequency depends on requirements for the SMPS frequency, capacitor and inductor values on what is practical.

A possible alternative is to make a LCRC filter where the LC uses strictly SMD components tuned for the best rejection of switching noise, then add a RC filter with a largeish electrolytic capacitor and a resistor so that the RC forms a Zobel that critically damps the LC filter (Q ~ 0.5) and additionally adds a further pole to make a 3rd order lowpass.

This circuit then is constant Q independent of the load and is completely unperturbed by the load. Downside is that the DC output is now not load invariant.

Of course, ideally we design the filter as part of the actual PSU, so that we use the DC part of the feedback from the final output of the filter and use AC feedback from nodes ahead of the filter, to get a stable control system. This way it is possible to make SMPS with noise levels comparable to linear PSU using TL431 as reference/error amplifier.

That is of course well past the topic of a post SMPS filter to kill the switching noise residuals in turn caused by commodity designs omitting a simple LC output filter and implementing a multi-lane feedback circuit, which would really not cost a lot extra.

Thor

 

[Matt Syson]

Using switcher mofules with inbuilt power factor correction adds another level of (different) noises to the DC output as modules supplying different currents will cycle out of sync and modulate the basic 60-100KHz main switcher leading to one heck of a messy output voltage. I use LC filtering then a LOW dropout Linear regulator ( half a volt at however many amps are needed) to clean up the resultant DC AND offer remote sensing in the 'linear' domain (at the expense of a little bit more wasted heat).
A replacement supply MUST give overall performance as quiet as an original properly functioning supply so in general terms less than say 5millivolts measured DC to 500KHz

 

[thor.zmt]

Using switcher mofules with inbuilt power factor correction adds another level of (different) noises to the DC output as modules

Yes. A modern agency compliant SMPS built to "commodity grade" standards makes quite an interesting wideband noise generator.

I use LC filtering then a LOW dropout Linear regulator ( half a volt at however many amps are needed) to clean up the resultant DC AND offer remote sensing in the 'linear' domain (at the expense of a little bit more wasted heat).

My personal take, commercially, is to use a high current Stepdown (Buck) converter to make a new "lowest common denominator" rail.

Say I want to handle 12V to 24V DC input, use a 100% duty cycle Stepdown with 36V withstanding voltage Step-down to 9V. Common IC's I employ are 5-8A output, over 800kHz switching frequency and can be synchronised externally, so two or four can be used in quadrature for higher current.

In an extreme case we might make a local PCB power bus using multiple pcb layers that operates effectively at 2.8MHz switching (4 pcs interleaved chips) with 20A @ 9V with a pretty much universal input DC for vehicular or other applications.

Now that's 180VA of DC. And what follows can be > 90% efficient, input power maybe 220VA from DC.

Then use externally synchronised boost or buck converters (> 1MHz switching) to make the actual voltages needed.

Modern switchers use bandgap references, so audio band noise is comparable to any linear regulators using unfiltered bandgap references.

Switching frequency, once you run at ~1MHz all components needed are tiny SMD. So while complex, such a system contains all noise on a "noisy island" and physical size can be very small, compared to power levels.



Example of such a design. The Mezzanine PCB to the left filters incoming DC (two inductors on top of PCB) and prevents conducted EMI.

Next a tiny drunk (legless) Buck IC makes 6V/6A (to heat tubes AND power the rest) and using a dual winding inductor makes an extra galvanically isolated ~6V/2A supply (logic, usb etc.).

A total of eight different supplies (including dual +/-18V supplies that are tracking negative and CPU adjustable positive via PWM) are then generated. Two of them use a greinacher cascade to make tube HT.

The big advantage of this approach is that after the buck converter we control everything, we can syncronise our switchers to any suitable clock, phase locked. We ignore the incoming DC totally and what noise we create locally we deal with locally.

The "Brick" vendor worries about agency compliance and power, we don't give a pair of fetid dingo kidneys how noisy his product is.

A replacement supply MUST give overall performance as quiet as an original properly functioning supply so in general terms less than say 5millivolts measured DC to 500KHz

Ahhhm, microvolt, not millivolt!

Otherwise 100% agreed.

Thor

 

[living sounds]

The commonly used Meanwell supplies already contain EMI filters at their inputs, according to the datasheets. Is there any point in installing additional filter caps (or even inductors) at the IEC inlets? I've got more than one device utilizing several of these little caged Meanwell switchers...