Difficulties with DC/DC converter 12V to +/-40V, 8 amps

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

I want to show you something. It's a push-pull converter from 12V DC to +/-40V, 8 amps (640W). I added here some pictures, schematics and efficiency graph. The converter looks quite nice, but there are some problems with stability :confused:. Could you please give me some advice.

Thanks

schematic:
An externally hosted image should be here but it was not working when we last tested it.


Vds1 and Vds2:
An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


Efficiency graph:
An externally hosted image should be here but it was not working when we last tested it.
 
I can't help you with your problem, but I thought I would say that you should compress photographs to JPG instead of PNG. This will result in smaller file size with no appreciable loss of quality (or tiny file size and very low quality, depending on how high you set the compression). PNG is great for graphics with solid blocks of colour, but bad when it comes to gradients and noise, like photos.
 
OK, the images are now on the different server.

efficiency:
16axt8k.jpg


Vds1 and Vds2:
2cgc46b.jpg


Vgs1 and Vds2:
1j9ks4.jpg


shcematic:
r7lseh.jpg


smps pic 1:
fdwnpt.jpg


smps pic 2:
np3m9i.jpg
 
Gate drive looks OK, but turn on is way too fast. The control IC can drive these MOSFET without buffers at the required speed, the 9A drivers are overkill.

Transformer leakage inductance looks high too, it can be improved with a properly wound toroid. E cores are a pain to wound for 12V push-pull.

Output inductors must be wound bifilar on the same core for proper coupling and control loop stability.

Frequency compensation is completely wrong (absent). If component values are right, you need a zero around 700Hz, and it must be implemented before the optocouplers. Forget about the zener and dual optocoupler approach.

Input EMI filter is missing.

Implementing a current transformer with an iron powder core is a very bad idea, it lacks precision, magnetizing inductance usually shunts most of the sensed current and is hard to predict and keep stable.
 
Eva, OMO, looking at the circuit diagram I notice there aren't any small decoupling caps (smd) on the supply rails esp near driver ic's and push pull centre tap. Ringing in the duty cycle deadband..
with no small decoup caps I question the scope probe ground....on tram designs this has fooled me many times due to charge induction effects. If the waveform changes when moving the ground lead around then what is one looking at ? I can't quite make out if vias are on the ground plane to underside.
To me each half primary isn't symmetrically switching, one side toward hard saturation looks like current mode ain't working properly.
The primary snubbers 10R &10nF ? seem high to me. The waveform suggests one side is getting hotter than the other.

I question the efficiency figures. Above 0.8 great care has to be taken with each drop and measurement. Can one expect more from the topology with only a 12V supply ?

I have wound an E core for 12V p-p. Not recommended unless one has a 100% precise understanding of leakage inductance and is good with copper foil. The terminations are quite critical.
The planar is an expensive alternative but also good if one likes a niche for such designs. Again a copper exercise.

richy
 

Attachments

  • cores.jpg
    cores.jpg
    42.8 KB · Views: 620
to Eva:

The current transformer is not made of iron powder core. I used ferrite toriod core (material 3C94 by Ferroxcube I think), so this should not cause any problems.

Could you give me some advice for the frequency compensation (e.g. some schematic or link on website). I would appreciate it, because the converter unstability really make me sad :(

Thanks
 
simingx said:
Why not just run the SMPS open loop (i.e. full duty cycle)? Then you can do away with regulation and it's associated headaches...

SMPS is a tough world; and one has to be responsible; the circuit, duty current limit must prove to work in both open and short circuit conditions. At the moment that doesn't look the case. It's a normal SMPS requirement to make sure this does work to avoid destruction.

Vena2a; the ground plane under, have you got a pic of it ?

richy:-
 
The requirement for regulation depends mostly on the load and the expected input voltage range... For example, for class D amplifiers I think that regulation is not required, unless you want maximum output power even at low input voltages.

For frequency compensation consider calculating the average of the absolute value of both rails, then substracting the target voltage from it, then applying a pole in the origin and a zero at 700Hz, then applying this signal to the optocoupler and then just to the duty cycle control. This may be achieved with TL431, op-amps or even transistors. Do you understand the terminology? The coupled output inductor is a must too. The 700Hz is roughly the frequency at which 12.5uH and 4000uF resonate [ Fr=1/(2*pi*sqr(L*C)) ]

By input EMI filtering I meant a PI filter on the 12V line providing good attenuation above 1Mhz (>40dB). The inductor should not saturate at full load and together with the ESR of the capacitors should result in a damped system. Filter corner frequency should obviously be placed well below clock frequency. In other words, HF stuff should not flow through the 12V line...

I thought that the core of the current transformer was iron powder because it's yellow. Remember that some dead time is required in order for the core to reset. You have to check with oscilloscope that it's really having enough time to reset at full load.
 
Question: Why don't you try to simulate with SwitcherCAD from Linear?
SwitcherCAD

What irritates me is the topology with the two optocouplers.
First you have only one symmetric DC-Link voltage (with a mid point) but it is not possible to regulate each side of your supply side individually! Therefore you only need to feedback the total output voltage with one OC. In your topology it is not possible to power positive and negative supply individually.

Secondly optocouplers are quite nonlinear components and make compensation worse - especially if two of them are connected in series!! Better: try using only one like in LTC Datasheet or add at least a parallel resistor to each OC.

In your schematic is no coupling between primary and secondary side and your transformer has no shielding inside. I recomend minimum a capacitor between the two GND's to have a short circuit for capacitive coupled commonmode currents.

If you like to increase your efficiency you could try using the synchronous recitification option in LTC datasheet. Yes it is more expensive but probably worth the money.

@Eva : The deadtime is only required for the Switches to desaturate and turn off. Otherwise a short circuit will occur. IMO the core needs no time to reset.
 
Aoxomox:

Both secondary voltages really can not be controlled individually with this connection. My idea for this connection of optocouplers was to contol the supply by the secondary voltage which is more loaded. This allows the supply to work with different loads on both outputs whereas the output voltages have at least required value.
But as luka said ... it's probably useless. It was just such idea.

About the synchronous recitification. I thing, that synchronous rectifier would not help a lot. Most of the power dissipation is in the switching MOSFETs ... there is weakness of this converter. It's strange for me because the MOSFETs (IRFP2907) should have Rds(on)=4,5mOhm and I have two in parallel.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.