SMPS design for power amps (Class D in particular)

[Golgoth]

Re: 'zero ripple' isolated PFCs

Ripple cancellation depends on large capacitors to work. It is a misnomer and would be better known as ripple reduction. By adding an extra winding or two in the right places almost any convertor topology can be made zero ripple. Cuk (rhymes with kook) converters are not unique in this regard and do not do better than any other four filter element topology. Please reread Jaka's post about power converters and the laws of physics.

Regards -- analog(spiceman)

Hi Analog,

(the conventional quotes in 'zero ripple' indicate it's not quite zero, but reduced by one order of magnitude or so, which is still quite a bonus regarding EMI filter size reduction)

Indeed the 'zero ripple' technique (discussed e.g. in http://www.hamill.co.uk/pdfs/azrtatad.pdf ) can be applied to other converters by adding extra windings, and the laws of physics are what they are, but can you think of one converter other than isolated Sepic (simpler than isolated Cuk), that can do 'zero ripple' isolated PFC with just 3 windings on a single magnetic component, or that can do the function with less component count, or less footprint? (cf Fig 1a in the paper I gave the link to)

And is one 0.78µF capacitor (like in the 170W Sepic converter of the paper) what you call "large capacitors"?

Cheers

 

[Jaka Racman]

Hi,

just 2 quick remarks. In proposed Cuk converter EC70 core was used for 170W. In normal converters any half decent SMPS designer would aim for 500-1000W with such a core. I can not comment for Sepic design since I could not find any core data on Magnetics website.

Second there are possible problems with leaky magnetics designs. I once designed boost PFC front end followed by series resonant capacitor charging converter. In resonant converter I used UI93 core for both transformer and series inductor. Inductor was simply 22uH of primary leakage inductance since I wound primary and secondary on different legs of the core. The end result was that it was almost impossible to attenuate third and fifth harmonic of the series converter. While mains filter attenuated PFC harmonics by 80db, harmonics of the resonant converter were attenuated by mere 20db. This was due to the induction of harmonics in magnetic components of the filter itself, although there were two 1mm thick steel walls betwen converter and filter.

Best regards,

Jaka Racman

 

[analogspiceman]

The isolated, double ripple "canceled" `Cuk (remember, it rhymes with kook ) converter has four independent energy storage elements: two capacitors and two magnetically independent air gaps sharing one core. Compare its performance to a standard isolated buck convertor with input filter that has equivalent total energy storage (I^2*L/2 and V^2*C/2) and equal total Vp*Ip products for its semiconductor switches and you will find that they both have similar terminal ripple components. They won't compare directly because the buck converter is inductive (high impedance) on its input and capacitive (low impedance) on its output whereas the double ripple "canceled" `Cuk actually is capacitive (low impedance) on both its input and output.

Unlike with the `Cuk converter, the input and output LC sections of the buck converter are less constrained and can be selected such that they do not interact via the switching action. Thus, the buck converter has the advantage of not requiring large RC damping networks across the converter capacitors in order to achieve control loop bandwidth beyond the filter corner frequencies.

There is no free lunch. -- analog(spiceman)

 

[Golgoth]

...There is no free lunch. -- analog(spiceman)

You may be right about zero ripple isolated Cuk (pronounces Chook actually), but what about zero ripple isolated SEPIC then? Simpler magnetic component, no RC damping network across the converter capacitor needed with the right control scheme, cf the Sepic.pfc paper I gave the link to, and this other one too (active voltage clamp isolated SEPIC PFC):
http://ece-www.colorado.edu/~pwrelect/Paper_archives/PCIM01.pdf

I have also come across a remarkably detailed and well written series of Unitrode Seminars papers by Lloyd Dixon describing the design of a 200W "zero ripple" SEPIC PFC, not isolated unfortunately but I suspect an isolated version could be extrapolated relatively easily:
http://focus.ti.com/lit/ml/slup103/slup103.pdf
http://focus.ti.com/lit/ml/slup104/slup104.pdf
http://focus.ti.com/lit/ml/slup105/slup105.pdf

About Jaka's remark regarding leaking magnetics, Dixon seems to suggest that a pot core with center post ground (like Bruno used in UcD) could be used for the Sepic magnetic component, this would yield minimal flux leakage.

Last but not least, to get a specialist's advice I asked Dragan Maksimovic head of CoPEC by email what he would recommend for a 1kW single stage isolated PFC for our application, here is his answer:
------------
I think active-clamp SEPIC would then be a good choice. Efficiency wise, I
would still think the two-stage would be comparable to SEPIC - the SEPIC
transformer is usually a problem. More recently I had a student (Praneet
Athalye) who did some work on high-performance active-clamp SEPIC - you can
take a look at the papers he published (I am a co-author).

Dragan

----- Original Message -----
...
> Hi Dragan,
>
> Thanks for your kind answer, sorry I wasn't specific enough. Here is why it
> seemed to me a zero ripple single stage PFC would make sense:
> - Output current ripple at 100/120Hz isn't a problem as the particular class
> D (switchmode) amplifier has very high supply rejection, and we need large
> rail capacitors to swallow back current from highly reactive speakers so
> output voltage ripple will be reasonable anyway.
> - Also I would like to keep cost and occupied space, including those of EMI
> filter, as small as possible.
> - I thought efficiency would be better than two-stage (but you tell me it
> will be comparable).
>
> Would an isolated SEPIC (necessarily with active clamping I guess) be a good
> choice do you think, or would another single stage topology be more
> appropriate for this kind of power?
------------
(the Athalye work DM refers to is probably the PCIM01.pdf I link to above)

Cheers

 

[Jaka Racman]

Hi,

I would say that on paper Sepic looks attractive. I did some experimenting with Sepic years ago. It was meant for capacitor charging applications from 24V supply bus. As soon as you go for a non leaky core (I used PM cores and stacked EE cores), you must add a small external inductor in series with the input inductor and you need a relatively large coupling capacitor because of good coupling between input inductor and primary of the transformer. Another drawback of close coupling is a lot of HF garbage on input inductor, so external filtering is still mandatory.

Best regards,

Jaka Racman

 

[Golgoth]

...As soon as you go for a non leaky core (I used PM cores and stacked EE cores), you must add a small external inductor in series with the input inductor...

I may be wrong but I disagree.

In my understanding you need a small external inductor when the windings are too closely coupled (to make up for the lack of leakage inductance), but I see no incompatibility between loosely coupled windings and a non-leaky core.

Imagine a pot core with center post ground, an inner winding (or two in closely coupled arrangement for an isolated version) + several mm insulation + an outer winding (cf Dixon's third paper for calculations). It should leak no flux to the outside although there will be a large leakage flux inside the pot and therefore a large leakage inductance. Or am I missing something?

 

[Jaka Racman]

Hi Golgoth,

I can only speak from experience. In my experiments I used interleaved primary and secondary, and then inductor wound on top. I might have used a much larger core and fill in some insulation between inductor and primary to increase leakage, but what is the advantage? External inductor was very small anyway, less than 10% of the main core. And you still have problems with HF garbage and inrush current because of large coupling capacitor. For your 1000W goal I would not consider any topology that requires use of more than 600V switches or rectifiers.

Regarding vast number of different topologies present in various articles one must ask himself: Is my goal pursuing an academic degree of working industrial design?

Best regards,

Jaka Racman

 

[Golgoth]

Jaka,

I gladly admit I haven't much experience in PFC, many thanks for sharing yours. I am only trying to find out what would be the best single stage isolated PFC topology for the practical UcD amplifier smps application we are discussing here, and from what I have seen up to now zero ripple SEPIC seems a good candidate.

I mentioned 1kW to Dragan in anticipation of future higher power UcDs, and according to him this topology is still a good choice at this kind of power, which is a good thing.

I thought your argument about perturbations generated by leaky cores was pertinent, so it seemed to me a center-gapped pot core with a spacer between windings might be a practical solution to have both high leakage inductance and virtually zero flux leakage out of the core, even if you never tried that experimentally you (or somebody else) might have an opinion on that.

BTW here is an interesting paper concerning ripple cancellation, which mentions the use of pot cores:
http://www.analogzone.com/pwrt0712.pdf

 

[analogspiceman]

You may be right about zero ripple isolated Cuk (pronounces Chook actually) [...]

Of course. It's a pronunciation that very appropriately rhymes with kook, because `Cuk actually believed and was blinded by his own misleading, contradictory hype about the magical qualities of his topology. To this day, I don't think he truly understands how the double ripple cancelled version of his converter works (I've quizzed him about it personally). However, in his heyday he did bring a lot of exciting controversy to otherwise arcane technical conferences. It's just too bad that the legacy of his hype still lingers and waylays the unsuspecting.

Shown below is a non isolated switch mode power converter that is the simplest possible embodiment of the double ripple "cancellation" principle. I call it the "X" converter for obvious reasons. It has two capacitors and two inductors, each of which are split into two equal windings. It is capable of producing a non inverted output voltage from zero to many times the input voltage (just the thing when your application needs a regulated 12 volt output from an unregulated 12 volt battery).

When you ponder the schematic below, keep the following in mind:
o The output is non isolated and the same polarity as the input.
o When the duty cycle for Q1 is zero the output is zero.
o At 50 percent duty cycle the output is the same as the input.
o L1A and L1B are two equal windings on the core of L1.
o L2A and L2B are two equal windings on the core of L2.
o The winding polarities for L1 and L2 are indicated by the asterisks and are all positive when Q1 is on.

    +Vi o-+                  +-o +Vo
          |                  |
          @)*                @) 
      L1A @)             L2A @) 
          @)                 @)* 
          |           C1 ||  | 
          +-----.     ,--||--+
          |      \   /   ||  |
       |+-+       \ /       _|_ 
     o-||   Q1     X     D1 /A\ 
       |+-+       / \        |  
          |      /   \   ||  |  
          +-----'     `--||--+
          |           C2 ||  |
          @)*                @)
      L1B @)             L2B @)
          @)                 @)*
         _|_                _|_
        ////               ////

The "X" converter is an almost unknown power converter topology based on split inductors. Its name comes from the x-shaped cross connection of components in the schematic of the most basic form of the topology. The split inductors add another degree of freedom that has kept the "X" converter virtually undiscovered in spite of efforts to classify all power converters with four or less energy storage elements.

There are several forms of this topology with positive or negative output, with ground referenced switches, and with natural isolation. This power converter has the same static conversion characteristics as a buck-boost or a `Cuk converter, and it has reduced ripple current on both input and output ports (sometimes referred to as "zero" ripple).

Although these ripple currents are not actually zero, it can be shown that the "X" converter has the very interesting and useful behavior of confining the bulk of its inductors' ramping magnetizing currents to internal loops. Through this mechanism, sometimes called "current steering", the "X" converter achieves ripple current characteristics similar to an equivalent buck converter with an input filter.

Since the topology has four energy storage elements, its undamped dynamic behavior is somewhat unwieldy. The control-to-output transfer function has two left half s-plane complex pole pairs and a right half s-plane complex zero pair due to the lattice topology and the switching action. This combination yields 540 degrees of phase shift! With proper damping, the zero pair can be pulled into the left half s-plane such that it nearly cancels one of the pole pairs and the overall response appears almost second order.

Getting ground referenced drive to Q1 may appear problematic due to the flying nature of the switch, but this is easily done simply by adding a small bifilar gate drive winding to L1B. Since first coming up with it in the mid 80's I have built and experimented with several versions of the "X" converter circuit.

It is fun and easy to simulate in LTspice (both in the frequency and time domains). -- analog(spiceman)

 

[subwo1]

Interesting, I once found that in a push-pull converter, once of the MOSFET drains could be capacitively coupled to the transformer.

 

[Golgoth]

Hi Analog,

I am perfectly aware that Dr Cuk's converter is nothing magical. It was only thought to be magical at the time because it was the first found to exhibit 'zero' ripple as I recall.

Regarding your X converter, thanks for sharing this little jewel with us, but I don't suppose you suggest it might be a candidate for the application we are discussing? (two cores, two capacitors, about twice too many!)

You haven't commented on my current soap box, the zero input ripple isolated SEPIC (one core, one capacitor), is it just because there is no Dr Sepic you could nag at?

Cheers

 

[Jaka Racman]

Hi,

Analog, do I have a reason to believe that "Analysis of the Near Zero Input Ripple Condition in a Symmetrical Push-Pull Power Converter" by Edward Herbert, HFPC 1989 is nothing new to you?

Michel,

I must admit that currently I have no idea what would be the best topology for audio. I searched ICEpower website. Their 500W module with integrated power supply is able to sustain full output power for 60s. It is capable of 90W of output power at 25 degrees C and 35W at 50 degrees C ambient temperature.

So basically we are looking for a design with peak to average power ratio somewhere in between 3:1 and 8:1. It is not a problem to design transformers that can be grossly overloaded, but inductors are a little bit more problematic. We are looking for material with high saturation flux density, which might not be a low loss one. (HighFlux, amorphous materials or even iron comes to mind). So for the time beeing I am not sure if integrated magnetics is the way to go. But you are right about the possibility of Sepic with a single core. Because of pulsed power, less copper than usual is needed and that gives space for required leakage.

If using Sepic, I would probably run it in the critical conduction mode with passive lossless turn off snubber and would use IGBT instead of FET. I think this would be a more robust solution than active clamping.

Best regards,

Jaka Racman

 

[analogspiceman]

Originally posted by Jaka Racman
Analog, do I have a reason to believe that "Analysis of the Near Zero Input Ripple Condition in a Symmetrical Push-Pull Power Converter" by Edward Herbert, HFPC 1989 is nothing new to you?

Actually, I've never seen that paper as far as I know. The university library in my area doesn't have that series of proceedings (probably because they are not an IEEE publication). I think I own just one year myself (97?) and I have access to a few others through friends. Is that paper something I should see?

Originally posted by Jaka Racman
I must admit that currently I have no idea what would be the best topology for audio. I searched ICEpower website. Their 500W module with integrated power supply is able to sustain full output power for 60s. It is capable of 90W of output power at 25 degrees C and 35W at 50 degrees C ambient temperature.

So basically we are looking for a design with peak to average power ratio somewhere in between 3:1 and 8:1. It is not a problem to design transformers that can be grossly overloaded, but inductors are a little bit more problematic. We are looking for material with high saturation flux density, which might not be a low loss one. (HighFlux, amorphous materials or even iron comes to mind).

Keep in mind that such an inductor will have a seemingly high number of turns and large air gap because of requirement for high peak currents at low duty cycle. The relatively high number of turns will lead to correspondingly low ac flux excursions so that high loss core materials should not be a problem.

Regards -- analog(spiceman)

 

[analogspiceman]

Originally posted by Golgoth
I am perfectly aware that Dr Cuk's converter is nothing magical. It was only thought to be magical at the time because it was the first found to exhibit 'zero' ripple as I recall.

Regarding your X converter, thanks for sharing this little jewel with us, but I don't suppose you suggest it might be a candidate for the application we are discussing? (two cores, two capacitors, about twice too many!)

You haven't commented on my current soap box, the zero input ripple isolated SEPIC (one core, one capacitor), is it just because there is no Dr Sepic you could nag at?

Hey, in spite of my "kooky" opinions, I'm no poison pen and am not particularly anti-sepic either. I just don't have any experience with that topology (might be just the ticket for the UcD180 power level). I find understanding integrated converters easier once I've devolved them into their equivalent non integrated magnetic versions. Haven't done that for a Sepic yet. Am guessing it is functionally equivalent to a boost plus flyback (or maybe forward).

By the way, I didn't intend to imply that my X converter was an appropriate topology for the application du jour (it's not). I offered it as a thinking tool for understanding the operation of any of the double ripple reduced family of converters (including the fully isolated super 'Cuk - rhymes with kook). The reason 'Cuk himself never came up with it is because he never really understood the equivalent magnetic circuit of his double ripple "canceled" converter (IMO).

Regards -- analog(spiceman)

 

[Golgoth]

...
But you are right about the possibility of Sepic with a single core. Because of pulsed power, less copper than usual is needed and that gives space for required leakage.

If using Sepic, I would probably run it in the critical conduction mode with passive lossless turn off snubber and would use IGBT instead of FET. I think this would be a more robust solution than active clamping.
...

Note that required leakage could also be achieved without wasting space, using a side by side arrangement instead of a concentric one (cf fig 4 of the pwrt0712.pdf paper I linked to), moreover the leakage inductance value would be more repeatable (spacer thickness being difficult to control).

Passive lossless snubber? Sounds good, what is it and how does it work? A schematic (or better an ltspice simulation) would be welcome.

 

[Jaka Racman]

Is that paper something I should see?

Hi, only if you are interested in the following:

-take your x converter and substitute diode for another fet
-wind all the inductors on the same non gapped core and treat resulting structure as primary
-add secondary windnding with classical center tapped push pull rectifier and LC filter

What you get is a single non gapped magnetic component buck type converter with zero input ripple.

If you have good nerves you might also read some of the latest patents by the good old professor Cuk. They are quite a lectures, each having over 100 pages. You will find if his knowledge of integrated magnetics has expanded. Hint: his latest preoccupation seems to be flux cancelation in a single core, where you end up with smaller core because of cancelation of fluxes produced by the inductor DC currents. Here is one example

Best regards,

Jaka Racman

 

[Jaka Racman]

Hi Michel,

here is a quick link.

Best regards,

Jaka Racman

 

[Golgoth]

Hi Jaka,

Thanks for the lossless snubber link, very interesting. Besides, cf p.10 of the slup103.pdf paper, Sepic has intrinsically virtually no turn-off loss, so maybe one could use an IGBT even without a turn-off snubber? Still, with or without a turn-off snubber I imagine the use of an IGBT would severely limit the switching frequency.

Concerning the transformer, I was surprised to see on page 2 of the Sepic.pdf isolated Sepic PFC paper that their primary and secondary were wound bifilar, is that allowed by the norms? (in the other isolated Sepic PFC paper -PCIM01.pdf- the transformer has the usual insulator layer between the windings).

Ah and a last thing, is isolated Sepic suitable for multiple outputs? (we need at least two!) I don't see why not but I saw no reference to such a thing in Google. Admittedly there are very few references at all to isolated Sepic, which apparently is a very recent development, as it hadn't been experimented yet at the time of the slup103.pdf Dixon paper (1993), cf page 11.

 

[Jaka Racman]

Hi Michel,

lossless turn off of mosfet in the cited app note is more wishful thinkig than reality and is not function of the Sepic converter. The same could be said about any other hard switched converter with mosfet. It would require very fast turn off of the mosfet, with self capacitance acting as a snubber. With IGBT it would be impossible. What I had in mind was running Sepic on the border between continious and discontinious conduction mode. In this case you get zero voltage zero current turn on for free, so only turn off snubber is required. Because of higher peak current and high voltage (switch with 800V capability would be required) I thought that IGBT would be a better choice.

If you choose so called "triple insulated wire" then bifilar winding is indeed possible.

As Sepic is basically boost stage followed by flyback converter, it has a very good cross regulation, so it is very suitable for multiple outputs.

Best regards,

Jaka Racman

 

[Golgoth]

...What I had in mind was running Sepic on the border between continious and discontinious conduction mode.
...

Hi Jaka,

I found your idea of BCM (boundary conduction mode) for the Sepic PFC attractive, so I did a Google search and found that it has already been experimented with success (non-isolated though), by a CoPEC team again:

http://ece-www.colorado.edu/~pwrelect/Paper_archives/PESC01SEPIC.pdf

Nice things with this scheme:

- soft turn-on as you said (no diode recovery losses since transistor is turned on at end of diode conduction), which means low radiated EMI.

- simplified control (no current loop necessary)

- cherry on top: BCM is _perfectly compatible_ with input ripple steering as demonstrated in the cited paper, which gives design rules for both reduced and zero ripple. Actually in their experimental setup they only went for reduced ripple, but I don't see why zero ripple couldn't work. They say the zero ripple condition would lead to excessive switching frequency, but since switching freq is highest at line voltage zero crossings in BCM I suspect one could clamp switching freq to a suitable value without overly deteriorating power factor. So it should generate very low conducted EMI.

And since bifilar (well, at least trifilar since we have two outputs) is possible as you say, isolating the BCM Sepic PFC converter should be feasible without excessive primary to secondary leakage inductance.

Well, it looks like we're talking very high performance for a minimal component count! Admittedly it requires high voltage and current transistor and diode, but apart from that what's the price to pay for this free lunch Analog?

Cheers