Parallel-ing Switching Power Supplies

[JohnT_diy]

Hi. I am a new member needing some info in paralleling of SMPS. I have searched these forums and Google but can't find the answers I need... Basic power supply design is 12V input, SG3525 directly driving IRFZ44's in push-pull with NO regulation, currently using ferrite core salvaged from a PC power supply and rewound. Output currently AC and is about 300Vp-p into a 240V/200W lamp, these are a cheap/convenient load. Adding a second 200W lamp to the output causes output to start to droop slightly and output waveform a bit rounded on the leading edge. I am testing throughput of the power supply at the moment... Eventually will set the required transformer turns and rectify the output etc. I want to make a possibly modular DC supply of high power (maybe 1000-2000W) but hoping I can parallel some smaller, 400W, already designed units. I don't want to use diodes to "OR" the DC outputs to a common bus... I could use the same single SG3525 to drive FETS on multiple identical power/rectifier stages and common up the rectified DC,,, or can I use common bank of fets and parallel the transformer primaries? Thaks, John.

[star882]

It's better to wind the transformer secondaries for only a fraction of the desired output voltage (but full output current) and connect them in series to add the voltages. Adding current requires load balance circuits and close tolerances.
Drive the transformer primaries with separate stages driven by the same controller.

[JohnT_diy]

Thanks Star. I'm not planning to use such an extreme output voltage as I have at present. I used the 300V output to test power throughput, because it's easy to use lamps as a load. At my required lower output voltage of around 100V, I keep burning up load resistors... Core warms up a bit with a 400W load too. The PC SMPS transformer is pretty small but doing an OK job.

I will look into paralleling the gate drive to multiple power stages soon as a bunch of ETD39 cores arrive.

My next question is about the secondary side:
Is is OK to parallel the rectified DC? or probably not the AC? I presume paralleling the DC is kinda like diode OR-ing the supplies to a common bus which would isolate the supplies from eachother. The individual power stages won't need to precisely share the load.

[BWRX]

Originally posted by JohnT_diy
Is is OK to parallel the rectified DC? or probably not the AC?

Think of it as you would a regular transformer. You can parallel the secondaries IF they are well matched and wired in phase.

[luka]

Hi

Your input voltage is 12v and output 100v?
Why don't you just make one that will be able provide that power, or like two would be enough. How much do you plan to use them?

[star882]

Originally posted by JohnT_diy
Thanks Star. I'm not planning to use such an extreme output voltage as I have at present. I used the 300V output to test power throughput, because it's easy to use lamps as a load. At my required lower output voltage of around 100V, I keep burning up load resistors... Core warms up a bit with a 400W load too. The PC SMPS transformer is pretty small but doing an OK job.

I will look into paralleling the gate drive to multiple power stages soon as a bunch of ETD39 cores arrive.

My next question is about the secondary side:
Is is OK to parallel the rectified DC? or probably not the AC? I presume paralleling the DC is kinda like diode OR-ing the supplies to a common bus which would isolate the supplies from eachother. The individual power stages won't need to precisely share the load.

Not a good idea. If it's even a little off (and it will be unless you wind all transformers *exactly* the same and use *exactly* matched components), one stage will (try to) carry all the load. Correctly paralleling the circuits requires the drive signals to be staggered so all stages see the load.

In a voltage regulator circuit for a Britney CPU, some (very high speed, refrigerant-cooled) Britney CPUs can use 150A or more at 1.45v. That's too high for even a 4 phase voltage regulator module, so two modules were used, each supplying up to 100A (rated for up to 125A). The two regulators were interlocked together so they supplied the same current and tried to attain the same voltage. (If the outputs are even 10mV off, the regulators will fight and circulate destructive amounts of current.) They ran off the same PLL, with one staggered 1/8 of a clock cycle behind the other.

At voltages above about 24v or so, it's easy to wind the transformer secondaries for part of the output voltage and then connect them in series.

[JohnT_diy]

Hi Brian. Yes, matching home wound transformers would never be exactly the same, and I guess slightly different switching times/waveforms would also rule out paralleling the AC outputs.

Hi Luka. I didn't want to massage an existing design into higher power, it may be simple though... I expect big bad things will happen and it will be more complicated to wind a single 2kW transformer, and the poor PCB traces.

Hi Star
I think I understand now...
1. Share the gate drive signals to each power stage
2. Wind low voltage (around 35V) high current (around 10A) secondaries and connect in series to get the required voltage. All power stages will supply and therefore share more or less the same current/power. Perfect.
3. The unfortunate thing would be a single rectifier module at a high voltage - sounds expensive - whether it be diodes or synchronous/FET

[star882]

Originally posted by JohnT_diy
Hi Brian. Yes, matching home wound transformers would never be exactly the same, and I guess slightly different switching times/waveforms would also rule out paralleling the AC outputs.

Hi Luka. I didn't want to massage an existing design into higher power, it may be simple though... I expect big bad things will happen and it will be more complicated to wind a single 2kW transformer, and the poor PCB traces.

Hi Star
I think I understand now...
1. Share the gate drive signals to each power stage
2. Wind low voltage (around 35V) high current (around 10A) secondaries and connect in series to get the required voltage. All power stages will supply and therefore share more or less the same current/power. Perfect.
3. The unfortunate thing would be a single rectifier module at a high voltage - sounds expensive - whether it be diodes or synchronous/FET

The MOSFETs each have their own gate resistor. They connect to a MOSFET driver chip that can handle the load.
When connecting the secondaries in series, test at low power to make sure the phasing is correct. Make sure the voltages add.
Fast recovery rectifiers are actually pretty cheap. You'll need 4 for a bridge or 2 for a voltage doubler.

[theChris]

effective parallel schemes work with current-feedback. each supply adjusts its output based upon some measurement of current. the easiest way is to apply current feedback to the supply itself, simulating a high output impedance. this allows multiple supplies to be paralleled. The simplist method would use actual resistors, but a control-based approach would reduce losses. if the supplies reference is reduced by an amount that scales with output current, the highest output voltage will force the most current, and will be forced back to the lower voltage (lowering current a bit).

Another option is to have each supply send a signal corresponding to its current output. If the output is a current type, multiple outputs can be wired in parallel. each supply then compares its output current to the average output current, with the setpoint adjusted as needed to bring it closer to equal current output. This allows the current sharing AND allows for regulation.

Of course, with multiple systems, you can get some fairly nice oscillations.

edit -- this is for use of multiple power supplies, not means of scaling up existing single supplies.

[darw82]

Paralleling the transformer.
I think it is better to connect the transformer in series.
You won't need such current mode control, to ensure current sharing. SG352X will be fine.

Those voltage drop you mention, is caused be leakage inductance in the transformer, series/paralleing transformer will help.