Slobodan Cuk classD

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The first paper describes the use of a cuk converter for a class D amp. Looking at the title, it was a final year project by a Swiss polytechnic student.
They state that with this topology it is possible to have high dynamics, low noise combined with virtually no current ripple at the output of the amp. The lack of ripple renders additional filters needless and reduces EMI as well.
They write the hard part was to build a controller for this cuk converter since it shows a non-linear 5th order transfer function. The solution they found was a "one-cycle controller" with a simple 1st order PI transfer function.

My german is better than my english so I hope this translation was still usefull.
Btw, it's my first post after reading the forums for more than a year now :)
Cuk's converters for SMPS were first proposed in (I think) the 1980's as a way to get a variable ratio step-up/step down DC-DC converter with continuous input and output currents, thus avoiding large ripple currents at either input or output, using a single switch at the input side and a single diode at the output side.

The input and output inductors can be realised as a single coupled inductor (depending on isolation voltage requirements), the circuit can provide DC isolation using a transformer, and in theory the coupled inductors and the transformer can be combined into one integrated magnetic structure.

Difficulties with this design include the fact that the non-isolated version is inverting, there's an inband resonance between the input/output inductor(s) and the coupling capacitor(s), the stress on the switching device is large, and the magnetic structure doesn't lend itself to conventional manufacturing techniques.

For one or all of these reasons the design never caught on. I looked at this architecture many years ago for class-D amplifiers and found the following:

Using two Cuk converters driven in antiphase (one with widening pulses, the other with narrowing ones) can produce a self-oscillating class-D amplifier with constant frequency (because each output voltage is (D/(1-D)) where D is the duty cycle -- for example, if each output is +100V at 50% duty cycle, at 33% one output has fallen to 50V and the other has risen to 200V. at 20% duty cycle the outputs are 25V and 400V and so on. This does mean that the combined voltage and current stress on the switching devices becomes *very* severe...

This works fine at DC and doesn't seem to need any output filter, but it all goes horribly wrong when the output voltage changes and the LC resonance inside the converter shows up, typically inside the audio bandwidth.

Maybe more advanced control techniques can fix this issue, but I doubt it, it seems to be an inherent problem with the architecture. I think this (together with the other issues mentioned above) is why this was never used for class-D.

lumanauw said:
Hi, Iand,

Maybe that's why we doesn't see much of this topology, although it offers small ripple in and out.
Meanwhile, this paper is about Cuk's amp described in the patent.

He published lots of other papers too, especially about isolated versions of the converter which don't have the problem of needed both N and P switching devices and provide both regulation (step-up/step-down) and mains isolation.

This can be extended to the bridged class-D amp by putting a transformer in the middle and making the DC input rectified mains voltage (or PFC output). Then the mains side has 2 switching devices and so does the output side, with all bulk capacitor charge storage on the mains side -- however this then needs 2 coupled inductor/transformers (or "integrated magnetics" structures) per amplifier.

When I looked at it the biggest problem (assuming the LC resonance issue is solved) was the increased voltage/current stress in the switching devices, this is much less of a problem nowadays with the advances in power MOSFETs, IGBTs and high-speed diodes -- maybe it's time for another look at the circuit...

Isolated Cuk, do you mean something like this?

-- maybe it's time for another look at the circuit...

National made 1.4Mhz Cuk driver, LM2611


  • cuk.jpg
    63.5 KB · Views: 807

you can flunk the first linked paper right from the start. It says (my translation to English, last sentence in first paragraph):

The high phase shift in the output filter leads to a low loop gain, which results in higher non-linear distortion.

Low loop gain? No way. Look at UcD. Or any other topology. You can have loop gain as much as you want or as much you can handle.

Just another case of "he didn't looked far enough".

right, mr. baseballbat, the high phase shift in the output filter leads to a low loop gain, which results in higher non-linear distortion. if i wanted to have low loop gain and high distortion, i'd constructed an amplifier with valves during my diploma work ... . :hot:

the thought was that the cuk converter actually provides better dynamics due to the missing LC-filter of a conventional class-d amp ... . well: he does. and there's no current ripple. :rolleyes:

this was the plan: telling my professor's that conventional class-d amplifiers had: poor dynamics, poor linearity but low switching frequency, or: good dynamics, good linearity but high switching frequency an all the emi-problems of fast and hard switching converters, then: selling the 4-quadrant cuk converter as the solution to all those problems by high dynamics at low switching frequency due to the missing LC-filter, the missing current ripple and the low conducted emission, and finally: getting the job and getting a good evaluation by just doing what i used to to in my leisure time: constructing an audio amplifier. :angel:

now the truth about cuk converters: if they would work in practice as well as they do in theory, mr. cuk had sold more than only two of his patents ... :xeye:

the conclusion of my diploma work called "high fidelity cuk converter audio power amplifier" was:

- dynamics: good
- linearity: good
- current ripple: zero
- controllability: in theory excellent, in practice not sucessfully ...
- unsuitable for audio applications :whazzat:

you find the whole work here:

Only for those who understand it:

Ahhh - no 'ne angere Bärner wo'ne Schautverstärker aus Diplomarbeit 'bout het !!!

For David

A one cycle controller is quite a refined control circuit. You could look at it as the PWM modulator incarnation of a dual-slope converter.

It was invented by the ones who were the driving forces behind nu-force. Strangely they don't semm to use it for their amps.

look here:


To answer your question, lumanauw:

- yes, it produced sound. but in a very poor manner (due to the missing feedback and several problems with the switching devices ...).

- one cycle control is has been described by smedley in "control art of switching converters". in brief: the one occ-technique makes a non-linear switch linear.

I did some quick behavioural calculations for the isolated bridged Cuk amplifier, targeted at high-power amps (2500W/4ohms) similar to the QSC380:

-- two bidirectional amps with bridged output
-- one switch per amp on audio side, load bridged across two amps
-- one switch per amp on mains side, push/pull amps in series
(2 series 200V capacitors for DC bus, either from PFC or rectifier)

This way the switches are identical on both sides, and with 50% PWM duty cycle each has a peak voltage equal to the input DC bus voltage -- 325V nominal for rectified 230V mains, maybe 380V for output of a PFC.

For 2500W/4ohms the duty cycle varies from about 0.4 to 0.6 so gate drives can use transformers; the problem is the stress on the switches. The amplifier output is 100V/25A rms (140V/35A peak), the peak voltage across the switches is 400-450V (depending on DC bus voltage) which is fine for 600V devices (maybe even for 500V) -- but the peak switch current is 85-90A, which is about 2.5x the output current (even average current over a sinewave is about 30A).

The switches either have to be CoolMOS MOSFETs with series Schottky and antiparallel ultrafast diodes (like Ixys IXKF40N60SDC1) or ultrafast IGBTs with copack FRD (like Ixys IXGR48N60C3D1).

The IGBT can just about cope with the current (forward drop about 2.5V, which still means 300W loss), but the switching losses are too high -- more than 2mJ/cycle for 4 switches, which at 250kHz means >500W switching dissipation at 2500W output :-(

The MOSFET has lower switching losses (difficult to tell just how much lower) but will have large conduction losses, with Ron=60milliohms the forward voltage drop at peak current will be more than 6V.

In either case the combined losses due to on-voltage drop and switching means efficiency will be about 75%, which is pretty terrible for a class-D amplifier (though maybe not so bad if you consider that this also includes the SMPS function).

The reason for this is simple, the Cuk converter needs switches with *much* higher voltage and current ratings than the amplifier output signal-- in this case peak output is 140V/35A (5kW), but the amp needs 4 switches each with peak rating of 400-450V/85-90A (36-38kW), which is >7x higher than a normal full-bridge class-D amplifier...

Hi, Iand,

That's interesting :D What mechanism that makes Cuk needs higher voltage/current transistor?

Cuk converter has interesting equation, Vout=Vin*Ton/Toff.
It can magnify voltage or divide voltage, according to the ratio of Ton/Toff.
Is it possible to make big voltage swing with small voltage rail?

In Cuk converter (SMPS), after the capacitor there is only 1 diode. Why in Cuk classD, after the capacitor there is PNP transistor? What is this for, how does it work?

amuelhauser said:
right, mr. baseballbat, the high phase shift in the output filter leads to a low loop gain, which results in higher non-linear distortion. if i wanted to have low loop gain and high distortion, i'd constructed an amplifier with valves during my diploma work ... . :hot: [/b]

I build a ClassD-Amp in my diploma, too, a modified UcD (with bandwidth up to 100kHz and a switching frequency around 330kHz). And it worked, with high loop gain. 30dB is enough, especially when provided over the whole frequency. The little phase shift near the cutoff frequency of the output filter can be ignored.

You can have much higher loop gain if you build a standard phase controlled oscillator. Or an hysteresis controller. No problem.

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.