Electronics > Projects, Designs, and Technical Stuff
Problems with DIY SEPIC Converter
(1/2) > >>
Eldi4:
Hi,
I've been building a SEPIC Converter to power 8 Radio Access Point, and monitor each AP current consumption with some mcu, the power source is 2 x Car Battery in series, as i need to make the voltage stable on 24V i created SEPIC converter, previously i choose Inverted Buck-Boost converter, but because the output is inverted it makes the current monitoring each AP is hard, SEPIC is the only choice.

The targeted spec is 20~30V Input Voltage, 24V Output voltage, 8A.
after digging some information i start building it, The PWM Controller is UC3843 with roughly 100KHz frequency (Rt 3600R and Ct 4.7n), the Current Sense resistor is 0.05Ohm 10W, I use coupled inductor with core T90-26 and 22 turns 20AWG wire (Mathematically 34uH), The coupling capacitor is 4700uF, The MOSFET is 2xIRF1010E in parallel (previously was IRF640 but the leg was break), the output diode is MBR20200CT, the output capacitor is 3300uF electrolytic and paralleled with 4x22uF ceramic capacitor.

And here is the problem, the converter efficiency is horribly around 60% and the inductor is getting hot at 1.5A output current and it creates audible noise  |O , haven't tested at full load yet, as i does'nt have powerful enough power supply to supply the power.

Probably the efficiency can be improved by makin the mosfet closer to the UC3843? ( the range between UC3843 and mosfet is pretty far around 5cm ), i've been test to make the mosfet closer but the converter become unable to work, and using 2 mosfet in parallel actually increases the efficiency in my case, dunno why tho.

And probably the inductor is getting saturated?, but at 1.5A output current, the inductor RMS current is only  around 2.2A, and dunno why the inductor was causing a noise, the frequency is 100KHz but it can create audible noise, hmm....

Probably i will get it working better if i have an oscilloscope, but i don't have it :palm:, i don't even have a LCR Meter, only 2 multimeter there.
BFX:
Without some better measurement instrumentation I don't see big chance to get it working.
What kind of DMM you have? Are you able measure even frequency?
voltsandjolts:
The access points probably have DC-DC step downs anyway - try just running one direct from the car batteries. They won't need exactly 24V.
Eldi4:

--- Quote from: BFX on December 30, 2018, 01:02:27 pm ---Without some better measurement instrumentation I don't see big chance to get it working.
What kind of DMM you have? Are you able measure even frequency?

--- End quote ---

Sad but true  :'(
I can only measure frequency if it's zero crossing, my multimeter is cheapo chinese Zoyi VC17B+


--- Quote from: voltsandjolts on December 30, 2018, 01:07:24 pm ---The access points probably have DC-DC step downs anyway - try just running one direct from the car batteries. They won't need exactly 24V.

--- End quote ---

But it's for experimental and learning purpose too, hehe
T3sl4co1l:

--- Quote from: Eldi4 on December 30, 2018, 12:42:05 pm ---The targeted spec is 20~30V Input Voltage, 24V Output voltage, 8A.
--- End quote ---

Ok, so for a boost/flyback/SEPIC configuration, 1:1 winding ratio and ~1:1 voltage ratio, that'll be a peak switch current of about quadruple the output current, or 32A (assuming BCM)?

'3843 current sense voltage is 1V max, shouldn't that use (1V) / (32A) = 0.031 ohms, not 0.05?

Probably even smaller than that, considering minimum input voltage?  It can be a little more in CCM though.  Details, in any case; you're still at a fraction of the full output.

DCM, BCM and CCM (discontinuous, boundary or continuous conduction modes): are determined by the frequency relative to the inductance, supply voltage and load.  If Ipk > V*dt/L (where dt = 2/Fsw), it's in CCM, else DCM.  (BCM is the threshold between the two, where switch turn-on current is just barely ~zero.)

So, it looks like...



--- Quote ---after digging some information i start building it, The PWM Controller is UC3843 with roughly 100KHz frequency (Rt 3600R and Ct 4.7n), the Current Sense resistor is 0.05Ohm 10W, I use coupled inductor with core T90-26 and 22 turns 20AWG wire (Mathematically 34uH), The coupling capacitor is 4700uF, The MOSFET is 2xIRF1010E in parallel (previously was IRF640 but the leg was break), the output diode is MBR20200CT, the output capacitor is 3300uF electrolytic and paralleled with 4x22uF ceramic capacitor.
--- End quote ---

Ahhh yeah, that'll happen... ;D

- Hmm, 20AWG wouldn't really be enough for 8A, especially including ripple; doubled up (since this is SEPIC), it should be fine though.
- How was it wound?  A ball of wire on either side of the core?  Twisted pair?
- The T90 core sounds way too small to begin with, and mix #26 makes almost a better resistor than inductor at this frequency.  Try mix #8, or any of the fancy-named low-loss materials (Kool-Mu, Sendust, MPP, etc.; choose mu_r ~ 20 to 60 for best results), or gapped ferrite.
- What kind of 4700uF capacitor?  Just one?  What about supply (input and output) filters?  What kind of ceramic?
- Just one of those transistors is WAY MORE than enough to do the job.  Two in parallel is hurting you more than it is helping!  You may be better off with several IRFZ46N's (or STP50N06 or whatever cheap, old fashioned types are similar) in parallel, since the paralleling reduces stray inductance.  Stray inductance is critical at this peak switching current.

In applications like this, I like to use a current transformer to save on power lost in the current sense resistor.  That looks like this:



Note the CT primary winding in series with the MOSFET drain.  The secondary feeds a diode and burden resistor, which the 3842 senses.  The 1:150 turns ratio means the burden resistor sees 1/150th of the switch current, so effectively the 2.2 ohms acts as 2.2/150 = 0.015 ohms, for a peak switch current of 68A*; meanwhile, it dissipates 150 times less power than a single, direct resistor would!

*Hmm, sounds pretty high for a mere IRFZ46N.  What was the designer thinking? ;D

Note the inductor is spec'd as mix #26; I tried that initially, and it got way too hot.  I replaced it with severe overkill (a T156-2 with 10AWG equivalent Litz cable), which runs stone cold. :P  See side view here: https://www.seventransistorlabs.com/Images/Magamp_PSU2.jpg

The snubber (SB540 and 1uF cap) probably aren't necessary, but this type of clamping -- placed as close as possible to the transistor(s) -- can extend the voltage rating a bit, allowing you to accommodate more stray inductance between the transistor, coupling capacitor, output diode and output filter capacitor.

Keeping peak voltages in check is priority with a mere 60V transistor that you're running at 48V peak nominal -- that certainly leaves no room for automotive voltage variations, if this is an automotive application.

Anyway, ~30uH nominal, at 100kHz and 24V (nominal) input, suggests dI = 4A, much less than the 30A peak allowable here, so it's going to be CCM at more than light loads.  Which means...



--- Quote ---And here is the problem, the converter efficiency is horribly around 60% and the inductor is getting hot at 1.5A output current and it creates audible noise  |O , haven't tested at full load yet, as i does'nt have powerful enough power supply to supply the power.
--- End quote ---

...Chaos!  The peak current mode converter is a real, practical* implementation of the Logistic map, a beautiful bit of mathematics, which you don't really have to understand, just know that, when a control parameter (namely, inductor current) goes into a certain range... nutty things happen, and the result is a complex (not quite random; chaotic!) pattern that doesn't help us much, indeed it increases current ripple and makes the control loop somewhat unstable so we'd prefer to avoid it.  (It also makes a screeching or hissing noise, which is annoying!)

*Not that it helps us, just that, it's a good circuit... except for this behavior. ::)

So, what to do?
1. Choose a higher ripple current fraction.  4A ripple out of 8A load implies 50% ripple.  The instability goes away when inductor current is able to reach zero, i.e., in DCM, i.e., at >100% ripple.  Downside: more inductor losses, higher peak switch current.
2. Add slope compensation.  This is in the UC3842 datasheet or appnote -- check it out. :)  Downside: this worsens the current limiting behavior, which may cause the peak switch current to be much higher under some conditions (low input voltage and high load current), which you need to consider, to keep safe operation.
3. Don't use peak current mode at these power levels.  It's a good control but it's just not well suited to higher power levels -- there is an economy of scale working against you here.  More common are push-pull forward converters with average-current-mode control.  This can be implemented with the old standby TL494 (or '598 with integrated gate drivers, or because they're pretty weak, just add an external driver chip to either one).  You'll need an extra error amp to regulate output voltage -- the TL494 can't be wired that way by itself, unfortunately.

Depending on what capacitors you're using, I would guess the inductor, and maybe the switch and diode, in that order, are the major power losses in your circuit, and yes, efficiency that low is not unexpected given the above analysis.  (Well, I didn't really analyze much here, just talking about things I've analyzed or demonstrated in the past.  But yeah, I've been here, almost exactly.)



--- Quote ---Probably the efficiency can be improved by makin the mosfet closer to the UC3843? ( the range between UC3843 and mosfet is pretty far around 5cm ), i've been test to make the mosfet closer but the converter become unable to work, and using 2 mosfet in parallel actually increases the efficiency in my case, dunno why tho.

--- End quote ---

'3843 isn't very powerful, about 1A peak gate drive current.  That's equivalent to about 10 ohms resistance.  5cm of wiring will have a stray inductance of around maybe 40nH, and the gate equivalent capacitance is around 10nF (this is Cg(eq) = Qg(tot) / Vgs(on) ), so the gate circuit impedance is around Zc ~ sqrt(L/C) = 2 ohms.  To push this circuit into ringing, you need a driver below 2 ohms, and '3843 just isn't fast enough or strong enough to do it.

More importantly, with ~10nF hanging off it, that's a 10 ohm * 10nF = 100ns time constant, so we expect a gate rise/fall time on the order of 200ns, and a drain switching speed on the order of 50-100ns (the drain switching is faster, because most of the drain current change happens over a narrow span of gate voltages -- which is to say, the transistor has gain, and acts to sharpen the gate waveform).

At 100kHz, this isn't really terrible.  It's a maximum switching loss of around 100ns * 32A * 48V * 100kHz = 15W.  Still, it dominates over conduction loss, which is why I said the transistor is too big.

Out of a total ~200W capacity, 60% efficiency is 80W lost, about 60W of which is now unaccounted for.  Again, the inductor, and probably the diode, capacitors, I don't know what else -- will probably be the dominant losses here.



--- Quote ---Probably i will get it working better if i have an oscilloscope, but i don't have it :palm:, i don't even have a LCR Meter, only 2 multimeter there.

--- End quote ---

Unfortunately, designing and building SMPS is nearly futile without a scope. |O It can be done by an expert... but even then, there are so many small variables left uncontrolled that a 'scope is really necessary.

Good luck, hopefully you can find something useful!

Tim
Navigation
Message Index
Next page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod