How do you build an offline isolated multiphase power supply

[FlapJack]

I have been trying to figure out how to build an offline isolated multiphase power supply. Many controller chips let you slave multiple chips to one master chip but this just synchronizes them to operate at the same time.

In a multiphase system each chip operates at 360 deg / n where n = number of phases. The tricky part is how each controller knows how to share the load.

I have spent days searching the internet and have found many non isolated low voltage solutions for multiphase buck converters for motherboards, but nothing for isolated offline or any information on how to do it.

 

[schmitt trigger]

If I remember correctly, there are two phase PFC controllers, which is the front end of an offline power supply.
Is that what you need?

 

[FvM]

Before you discuss switcher chips and their synchronization features, you would decide for a topology.
- which AC/DC stage, simple rectifier? PFC? 1 or 3 phase?
- multiple or a multiphase transformer?
- which primary switch configuration, multiple single switches? H6 bridge?

 

[mtwieg]

Load sharing is enforced by operating all phases with current mode control. The current control loops must be fed the same current setpoint, which is the output of one voltage-mode error amplifier. So if you combine multiple controllers, one will act as a master, controlling the clocks and voltage error signals of the other phases.

 

[FlapJack]

Before you discuss switcher chips and their synchronization features, you would decide for a topology.
- which AC/DC stage, simple rectifier? PFC? 1 or 3 phase?
- multiple or a multiphase transformer?
- which primary switch configuration, multiple single switches? H6 bridge?

I was speaking in general terms, But mostly for making higher power supply's like 5 to 15KW, with or without PFC, maybe using 2 switch forward converters, and any number of phases up to 9 or so with one transformer per phase.

I am trying to teach myself how to build switching power supplies so i just make up different scenarios to see if i can design it. Just for fun i was trying to design a 10KW power supply (just design not build) at 30 or 40 volts. Well at the hundreds of amps it would take i realized that multiple transformers was the way to go. At first i thought synchronized operation was the way to go and them i found some ap notes on motherboard multiphase power supplies and realized that was the best way to do it.

Another possible multiphase supply i would like to figure out would be for a 12V to 120vac inverter. I saw a picture of an commercial inverter and it had 6 small transformers in it. I do not know if they were synchronous or multiphase but once again multiphase would be better.

- - - Updated - - -

Load sharing is enforced by operating all phases with current mode control. The current control loops must be fed the same current setpoint, which is the output of one voltage-mode error amplifier. So if you combine multiple controllers, one will act as a master, controlling the clocks and voltage error signals of the other phases.

Well, i can see this working for synchronous supplies, all supplies will operate at the same time and give a little to satisfy the demand. But for multiphase supplies, each controller next in line would use 100% of it’s power to satisfy the control loop. The last of the controllers would rarely be used. This would negate the usefulness of multiphase controllers to reduce ripple current.

 

[mtwieg]

Well, i can see this working for synchronous supplies, all supplies will operate at the same time and give a little to satisfy the demand. But for multiphase supplies, each controller next in line would use 100% of it’s power to satisfy the control loop. The last of the controllers would rarely be used. This would negate the usefulness of multiphase controllers to reduce ripple current.

I'm not sure what you mean by synchronous supplies. Usually that means they use transistors instead of passive catch diodes, maybe you meant to say synchronized?

In any case, that approach does ensure sharing between interleaved phases, and is the standard method. Could you explain your reasoning a little clearer?

 

[FlapJack]

I'm not sure what you mean by synchronous supplies. Usually that means they use transistors instead of passive catch diodes, maybe you meant to say synchronized?

In any case, that approach does ensure sharing between interleaved phases, and is the standard method. Could you explain your reasoning a little clearer?

Yes i did mean synchronized.

Let’s say 6 interleaved supplies. 360 deg / 6 = 60 degrees, each supply supplies current 60 degrees after one another. If each supply is fed the same current set point then the first supply that the clock triggers will be the only supply supplying power at that instant. It will try to satisfy the full demand of the sagging power supply. Each subsequent power supply will supply less and less power.

 

[mtwieg]

Yes i did mean synchronized.

Let’s say 6 interleaved supplies. 360 deg / 6 = 60 degrees, each supply supplies current 60 degrees after one another. If each supply is fed the same current set point then the first supply that the clock triggers will be the only supply supplying power at that instant. It will try to satisfy the full demand of the sagging power supply. Each subsequent power supply will supply less and less power.

No, you misunderstand. Each phase has its own independent current control loop, and all of them are fed with the same setpoint. This means that all phases will always try to carry equal current. The ordering of the clock phases doesn't make any difference. There is no "first" phase, since the cycle is periodic.

 

[FlapJack]

mtweig:

Through nothing less than superhuman effort of 3 days of Googgling, I found a total of two sentences from a Linear ap note on “how to” of interleaved power supplies.

“The current-sharing can be easily achieved by implementing peak current mode control. In a current mode control regulator, the load current is proportional to the error voltage in the voltage feedback loop. If the paralleled regulators see the same error voltage, they will source equal currents.”

Which sounds like the same thing you had said.

I am still searching the internet for an actual design example using separate controllers. So far i have not found even one design.

 

[schmitt trigger]

FlapJack;
your lack of success in your search may have a silver lining.

It may mean that there is no prior art to it, and if you can actually solve the problem, you may have a patentable circuit.

 

[FvM]

What's the actual problem? The concept of current sharing is simple, if you understand the operation of switcher ICs, you'll know how to use it in multiphase topology.

 

[mtwieg]

mtweig:

Through nothing less than superhuman effort of 3 days of Googgling, I found a total of two sentences from a Linear ap note on “how to” of interleaved power supplies.

“The current-sharing can be easily achieved by implementing peak current mode control. In a current mode control regulator, the load current is proportional to the error voltage in the voltage feedback loop. If the paralleled regulators see the same error voltage, they will source equal currents.”

Which sounds like the same thing you had said.

Peak current mode control is normal, but average or valley current mode control can also be used.

I am still searching the internet for an actual design example using separate controllers. So far i have not found even one design.

To implement it, you just need synchronizable controllers whose current setpoint can be overridden with an external signal (easy to do when they use transconductance error amplifiers). And you also need a way to generate the interleaved clocks. Beyond that there's no real trick behind this. Just look up "polyphase" on linear's website, they make dozens of suitable parts.

 

[FlapJack]

FlapJack;
your lack of success in your search may have a silver lining.

It may mean that there is no prior art to it, and if you can actually solve the problem, you may have a patentable circuit.

Believe me, the world has not waited for me to show up and invent the multiphase power supply.

I suspect that the offline high power versions i want are only made by the professionals for a specific job and the schematics are not published.

- - - Updated - - -

What's the actual problem? The concept of current sharing is simple, if you understand the operation of switcher ICs, you'll know how to use it in multiphase topology.

As i said in my original posts i am trying to learn how to design switching power supplies. I did not say i know everything.

 

[mtwieg]

As i said in my original posts i am trying to learn how to design switching power supplies. I did not say i know everything.

Right, we're just pointing out that implementing multiple phases isn't a great leap from implementing a single phase supply. The controller(s) just need to have the capabilities I described above.

For a high power offline isolated supply, you would want to do PFC and isolation in separate stages, and either or both of them could make use of interleaving. I've used the UCC28070 interleaved PFC controller with success to design a non-isolated 20kW PFC stage. For the isolation stage you could in principle use any topology, but the best choice will depend on how much power you need per phase. If it's just a few hundred watts per phase then a single-ended topology like two switch forward is fine, but for anything more you should look at full bridge.

 

[FlapJack]

Peak current mode control is normal, but average or valley current mode control can also be used.

To implement it, you just need synchronizable controllers whose current setpoint can be overridden with an external signal (easy to do when they use transconductance error amplifiers). And you also need a way to generate the interleaved clocks. Beyond that there's no real trick behind this. Just look up "polyphase" on linear's website, they make dozens of suitable parts.

I was thinking of using the UC2825 controller. It looks like a full featured capable controller. I looked the block diagram over for any sign of the double overlapping circle that usually indicates a transconductance amplifier but did not see anything. Here is the link for the data sheet. If this controller does not have a transconductance amplifier can you recommend one that does.

https://www.ti.com/product/uc1825

The clocking does not seem to difficult. The only question i would have is the duty cycle of the driving waveform to the controller.

I looked at the polyphase controllers on Liners site. They are all single chip solutions for low voltage low power aps and only one output per phase.

 

[FvM]

I looked the block diagram over for any sign of the double overlapping circle that usually indicates a transconductance amplifier but did not see anything. Here is the link for the data sheet. If this controller does not have a transconductance amplifier can you recommend one that does.

There are other ways to achieve the same. The UC3825 error amplifier can be connected as a voltage buffer that copies the output voltage from one "master" error amplifier to the other controllers.

 

[FlapJack]

mtwieg:

Are you saying you only used one UCC28070 PFC controller for 20kw.

 

[mtwieg]

I looked at the polyphase controllers on Liners site. They are all single chip solutions for low voltage low power aps and only one output per phase.

Most of them are meant to handle one phase per each chip, but they are capable of being combined and interleaved easily.

mtwieg:
Are you saying you only used one UCC28070 PFC controller for 20kw.

Yes, but I implemented external isolated gate drivers and amplifiers for the current and voltage sense, so it was significantly modified from the typical low power application it was designed for. This is typical for high power applications, the control IC does not directly interface to any of the actual high power circuitry.

 

[FlapJack]

mtwieg:

I can understand you would need higher power drivers for the power fet's. But why would you need an amplifier for the current and voltage sense.

 

[mtwieg]

Because the controller was completely isolated from the power electronics. Voltage sensing was done through precision diff amps, and current sensing was with hall effect current transducers.