What's wrong with coupled inductor SEPIC converters?

[treez]

Hello,

Why does Dr Ray Ridley, one of the world’s leading SMPS Engineers, state that the SEPIC converter with coupled inductor should never be done?

The last paragraph, on the right hand side of the third page (actually called page18) of the following states that the SEPIC with coupled inductor should never be used...........

**broken link removed**


So what’s wrong with the SEPIC converter with coupled inductor?

 

[mtwieg]

The article specifically warns against "almost unity coupling" between the two inductors, which is good advice. Using low coupling coefficients often works fine.

 

[Orson Cart]

The sepic is meant to have high value inductors, with no coupling, to function as intended, when you couple you change the topology of the circuit and introduce a resonant aspect that is not immediately obvious, this gives a partial ring in the current waveform and usually higher losses, less effect for lower coupling.

 

[treez]

Thanks Orson, you mean the ringing between the coupled inductor leakage and the sepic capacitor?

The non coupled sepic also suffers a ringing, and this is with the two inductors and the sepic cap. This is a low frequency ring which is dreadfully hard to snub out....(sometimes you see people using huge electrolytic cans as a snubber cap, in series with a huge axial power snubber resistor.)

So i'd say that ringing is far worse in the non coupled sepic. In the non coupled sepic, there is a relationship between the L-C-L resonant frequency and the crossover frequency, the crossover frequency should be at least three times higher than the L-C-L resonant frequency.......and with the SEPIC, this is not at all easy to do because it tends to go unstable. Some people get round it by using 500KHz plus switching frequencies, which makes it easier to get a high crossover frequency.

This is also why you are saying to do the non coupled sepic with big value inductors.....because it reduces the L-C-L resonant frequency , and makes it easier for you to get your crossover frequency three times higher than the resonant frequency.....though its still damned hard.

 

[Orson Cart]

Hello there, usually the load provides enough damping to give good enough performance, we have designed all manner of Cuk converters at 70kHz, which have much the same power stage dynamics as the SEPIC, and although we have a few control tricks to aid stability from no load to full load, essentially the high sw freq and the use of large value inductors, helps enormously in getting to a good starting point, if you can post your ckt so far no doubt some improvements can be suggested.
Regards, Orson,

 

[mtwieg]

When you guys say ringing, keep in mind that there's a big difference between ringing higher than the switching frequency (which would be problematic with SEPIC converters with very high K, or when layout parasitics are severe) and "ringing" much below the switching frequency. The latter type can be somewhat addressed with the feedback loop, the former type cannot.

 

[treez]

Regarding the ringing frequency and the crossover frequency................

I think the relationship is the same as in voltage mode full bridge converters....in these, the crossover frequency must be either three times greater than the output LC filter resonant frequency, or less than half of the output LC filter resonant frequency.

By "LC" resonant frequency , i am not referring to an output EMC filter, i am referring to the "L" which is the output power inductor, and "C" as the output capacitance.

 

[Orson Cart]

At 70kHz and moderate values for the 2 x L's and C, you should be able to get a control loop with a bandwidth of 500Hz, there are some feed forward techniques to apply to the volt loop which can push this out to 1kHz, it does require careful tuning of the feedback loop (esp at very light loads) - this is why many engineers shy away from the Cuk & Sepic, as they are quite a bit harder to get a stable system with good bandwidth than other power stages, e.g. resonant forward.

 

[mtwieg]

Having conjugate RHP zeros definitely will scare a lot of engineers, but such zeros can be eliminated (shifted to the LHP) with damping. Though there's still a good bit of math to it.

However in the case of coupled vs uncoupled inductors in the SEPIC, I'm quite sure it doesn't change the number of poles and zeros (though it does shift them around a bit).

 

[Orson Cart]

The coupling introduces extra terms into the transfer function of the power stage, usually restricting the useable bandwidth that can be obtained with a simple control loop.

 

[treez]

The coupling introduces extra terms into the transfer function of the power stage, usually restricting the useable bandwidth that can be obtained with a simple control loop.

coupled sepic (k>0.97) is pretty much the same as flyback, so the control is easier?

 

[Orson Cart]

Well you could run a sepic in DCM, where the control is greatly simplified with wider bandwidth, this is similar to a flyback in DCM, usually though you run it mostly in CCM, which is where the control gets trickier, slightly more difficult than a flyback in CCM, which also has control issues due to the inherent delay and drop in o/p power when you suddenly increase the ON time of the power switch.

 

[treez]

which also has control issues due to the inherent delay and drop in o/p power when you suddenly increase the ON time of the power switch.

i agree , though as you know, in flyback this is simple to handle...you just make the crossover at least three times less than the RHPZ frequency.

Good idea about DCM, do your points refer to coupled or non coupled sepic regarding DCM?

 

[Orson Cart]

I was talking with reference to CCM, DCM is not that efficient in terms of the amount of matter used for a given amount of power conversion, and the peak currents are typically 4 x the average processed current, although it is EMC quiet, and is used a lot for interleaved boost converters to get the switching losses and EMI right down, also eff higher as you can use "slower" boost diodes with lower Vf, and the cap ripple is n x the operating freq - but I digress.

 

[mtwieg]

The coupling introduces extra terms into the transfer function of the power stage, usually restricting the useable bandwidth that can be obtained with a simple control loop.

When adding coupling, it doesn't actually add any poles or zeroes, it just shifts their locations (as K is increased, the conjugate RHP zeroes will shift up in frequency). I've done the modelling myself to confirm it. If anything, a coupled sepic should have higher bandwidth than a non-coupled sepic (assuming that the inductances are the same).

coupled sepic (k>0.97) is pretty much the same as flyback, so the control is easier?

Well from a pure control standpoint it might be fine, but a sepic with very high K will still have those ringing currents which will really harm efficiency.

 

[treez]

mtwieg

but a sepic with very high K will still have those ringing currents which will really harm efficiency.

thanks, but your words send shivers down my spine, because i've just done a 5W coupled inductor sepic led driver with the MSD1583-473 coupled inductor, which is said to have K>=0.98.

MSD1583-473 coupled sepic inductors
**broken link removed**

The sepic capacitor is a 22u, 16V, X7R ceramic, which i hope is big enough to keep the ringing currents from getting too high in amplitude.
How can i get coilcraft to only send us the "k=0.98 and no higher" ones?



.....the ten prototypes that ive done so far are all showing efficiency of 70% with 5V load, and 79% with 40V load (vin = 6V)

...but with your words, i am now wondering if we will get a "bad" batch of coupled inductors with really high K.
-What do you think the chances of this are?

The switching frequency is 36KHz with a 40V, 5W load, through to ~126KHz with a 5V, 5W load. (its a constant off time coupled inductor sepic)
The leakage is quoted as 0.46uH.....which gives a resonant frequency of 50KHz with the 22uF sepic capacitor.

 

[mtwieg]

I wouldn't go above K=0.9, at that point you've mostly got the benefits from coupling, and any higher just makes ringing more problematic.

Coilcraft makes coupled inductors with lower K (0.8-0.85) specifically for SEPIC designs: https://coilcraft.com/msc1278.cfm

 

[Anna Conda]

A quick look at the current in the mosfet will tell you how the converter is performing, if the current is rectangular / trapeziodal , then likely all is well, if they have a noticeable, dip and peak, then the coupling is too high (not enough proper series inductance)...

 

[treez]

I see what you mean.

Strange that the MSD series from coilcraft (as in thread #16) also is stated as being for sepic, even though they have k>=0.98.

Have coilcraft got this slightly wrong?

 

[Orson Cart]

When adding coupling, it doesn't actually add any poles or zeroes, it just shifts their locations (as K is increased, the conjugate RHP zeroes will shift up in frequency). I've done the modelling myself to confirm it

Unfortunately - In the real world - when you build and run up a Sepic or a Cuk, you find that the coupled version is harder to stabilise than the un-coupled, if you write the complete transfer function for the power stage (now with mutual inductances, and magnetising inductances in the transformer/choke), and not a simplified version, you find there are more terms in the transfer function and the net effect is more phase delay making stable control harder to get - esp at light loads/no load where the RHP's are lowest in freq.