multiphase buck with single output inductor? 1

[Renovator1]

I'm having a tough time visualizing how this might work so I thought I'd pose the idea to the resident geniuses of eng-tips

Each phase in the typical multiphase buck converter has its own inductor which then feeds a common output capacitor while a single control loop adjusts the duty cycle to regulate the output voltage.

But what if the load is already highly inductive (ie - an electromagnet) and that it is the output current, not voltage, that needs to be regulated?

It seems to me that you could then eliminate the output capacitor and, more importantly, the individual phase inductors... I suspect, however, that each phase would then need its own current control loop to make sure the total load current is shared equally among the n phases.

What's tripping me up is that I know that if the "phases" are all switched at the same time - ie, all of the switches are in parallel - this configuration works just fine. It's when I try to visualize what might happen if the switches are operated in sequence (though with overlapping operation possible, of course) that my head starts hurting.

Oh, and the reason I'm giving myself a headache here is that using n phases rather than the same number of switches in parallel reduces the ripple current in the dc link.

Thanks!

 

[sreid]

I think you need a free wheeling diode for each inductor.

 

[Renovator1]

That you do, sreid. Not quite what I was asking about, though...

 

[Comcokid]

I'm not familiar with multiphase buck converters, but I'll try to think through what I believe your pondering.

"But what if the load is already highly inductive (ie - an electromagnet) and that it is the output current, not voltage, that needs to be regulated?"

Well, in current mode control, the inner control loop functions to make the switched inductor function like a current source, the voltage feedback loop takes the voltage at the output and feeds it back to control the current loop to provide a constant voltage at the output, making the it a voltage-controlled current source. If you break the voltage loop, and just put a controllable voltage, like a pot setup to provide a manually selected voltage, then the converter will function like a current source where you set the current with the pot and the actual output voltage will vary with load or input voltage since its only maintiaining a constant current.

You used the term "electromagnet", singular. If you have a single output inductor, youre no longer talking multiphase, but a regular buck converter. Unless the "electromagnet" your thinking of is typically three-phase AC power driven. If you drive a single electromagnet (inductor). with multiple switches, switching on different phase signals, this would become a rather odd arrangement, and will it actually reduce the ripple current? - you will probably need to take this back to the basic math and see. The inductor will already have induced flux as different phases switched.

"It seems to me that you could then eliminate the output capacitor..."

In current mode control, the output capacitor is part of the load which is why current mode control is desirable - the pole from the output capacitor dissapears from the transfer function resulting in a simpler control loop. Now when one normally thinks of the output inductor it is in series with the load, and an electromagnet is normally the load in itself, so you should be able to use the electromagnet as both the inductor and load, provided you don't drive it into saturation because that's where as an inductor it will have a very rapid decrease in inductance.

OK, I've mused over your post some to the limits of my thinking without the caffine fully kicking in.

 

[Renovator1]

Thanks for taking a shot at this, Comcokid.

The multiphase buck seems to be very popular for supplying 0.8V at 100A or more to the latest generation of CPUs. The idea is to divide the current among two or more smaller buck converters so that smaller and less expensive components can be used (though more of them, of course).

Here's a good article on how the multiphase buck is usually implemented:

http://www.eetimes.com/design/power...efits-of-multiphasing-buck-converters--Part-1

Referring to Fig. 1 in that article, I was thinking of replacing each mosfet synchronous rectifier with a regular diode and replacing each individual inductor with a current control loop (eg - shunt or Hall effect transducer) which then feed the electromagnet directly. I had no problem convincing myself this would work if I operated all of the switches simultaneously - that's just directly paralleling them - but when I tried to visualize switching each one in sequence I got stuck.

More specifically, I am trying to figure out whether the total current through the single inductor (and load) will split evenly among multiple switches or if each one will have to sustain the total inductor current, just for a narrower slice of time? Although the average current through the inductor will be the same either way, I suspect the switches will care which way this actually plays out quite a bit!

At any rate, I know this is sort of a weird question, but I thought it might be a more elegant solution to driving a big electromagnet by reducing the number and size of the electrolytic capacitors in the DC link (you know, the ones that always dry out then take the rest of the supply with them?). Then I remember that there ain't no such thing as a free lunch, and this looks suspiciously like a free lunch...

 

[vangjer]

If the MOSFETs are all controlled with a single PWM, why not build a current control loop to control the PWM.

 

[Renovator1]

Hmmm... I guess I'm not being clear but I don't know how else to phrase the question.

Making the example as simple as possible, let's assume the electromagnet needs 170V for 100A and that the dc link is 340V (rectified/filtered 3ph. 240V) and that a buck converter running at an average duty cycle of around 50% is the circuit chosen.

I know that this buck converter can be built with, say, two switches in parallel that are switched in synchrony at 50% duty cycle. Each switch, if well-balanced, will deliver around 50A of the 100A total load current.

My question is how the behavior changes if the two switches are still in parallel, but operated in anti-phase to each other (e.g. - 180 degrees apart)? This then becomes a multiphase buck, which is always drawn with individual inductors in series with each switch, presumably to ensure each one acts as a constant current source.

But if each switch had its own current feedback loop, would this not achieve the same result even if the switches are operated sequentially rather than synchronously?

C'mon guys and help an brother out!

 

[ScottyUK]

If the switches are in parallel and operate in antiphase then isn't the net effect that the switch is on all the time and freewheel diode never conducts?


----------------------------------

If we learn from our mistakes I'm getting a great education!

 

[Renovator1]

Well, ScottyUK, that's almost the question I am asking...

If the switches are paralleled and synchronous then every switch for the previous example is operated at 50% duty cycle. This is easy enough to understand, even for someone that's dead from the neck up, like me.

If the switches are in anti-phase, though, then it seems to me that each one would then be turned on for 25% of the period, so that the sum total of their on time was 50%. But if they are each to contribute half of the load current in half of the allotted time then their peak current must be twice as high, no?

 

[ScottyUK]

They can't truly be in antiphase then can they? In an antiphase switching pattern one switch is in the opposite state to the other at any given time. This sounds almost like you're trying to double the effective switching frequency by putting a double pulse into the inductor each clock cycle. I think you'd need discrete inductors feeding into a common bus for this to work - it would likely allow you to use a smaller output bus filter capacitor. I have half an idea about each switch feeding into a common inductor through a series diode, with a common freewheel diode, but I'm dog tired and my brain isn't firing on all four right now. I'll maybe sketch it out tomorrow and see if it makes any sense.


----------------------------------

If we learn from our mistakes I'm getting a great education!

 

[Renovator1]

I don't know why the two switches can't be in anti-phase, unless your definition of antiphase is 180 degrees of phase shift AND 50% duty cycle for each switch.

Attached is a picture of what I am trying to imagine doing. There's no output capacitor because even at a pokey 3kHz switching frequency the current through the inductor (~1mH) is almost pure DC.

I know this type of circuit works just fine if T1 and T2 are switched at the same time; it's when T1 and T2 are phase-shifted (as I tried to draw) that my imagination fails me.

 

[ScottyUK]

Not sure what the 'official' definition of antiphase says, but my own opinion is that in an antiphase switching pattern one switch is doing the opposite of what the other is doing.

In your scheme you're effectively increasing the pulse frequency.



----------------------------------

If we learn from our mistakes I'm getting a great education!

 

[sreid]

If you are looking to control current through a load inductor, the usual way is to PWM the voltage directly to the load inductor. The current is controled using a current sensor and a PI Control Loop.

There are single IGBT Modules that easily haddle 300 V and 100 A.

 

[Renovator1]

ScottyUK - Yes! The goal is to reduce the ripple in the input capacitor. Now, the question is, will this configuration work?

sreid - See above.

 

[Comcokid]

I don't quite see how this is going to work like you want unless you go to a true multiphase with seperate inductors.

If you have more than one switch switching on a single inductor, then as one switch opens, you have a brief dead-time, and you get conduction through the diode (as the inductor trys to keep the current flowing) and then you close the other switch and repeat the process - how is this not just equivalent to a standard buck converter running at twice the frequency (for two transistors) or at N times the frequency for N switches? Now, you really haven't changed the ripple current so much as you have upped the frequency of the ripple current.

Now, if you have a lot of switches where each one is closing on a different phase, and each phase slightly overlaps with another phase such that the inductor always has a least one switch conducting then you effectively have a dead short feeding your inductor, then the inductor is saturated, and Vin=Vout which is the result of a buck converter running at 100% duty cycle (which is the duty cycle where at least one switch is closed at all times). The fundamental thing of converters is not duty cycle on your switch (which is what one usually is focused on due to heating) but duty cycle on the inductor as it is the inductor that is the energy storage element.

Look at your diagram where you've taken a multiphase, and replaced the inductors with a single inductor. Now, you have two diodes in parallel, so if you eliminate the redundant diode then you have two switches in parallel, so you simplify by eliminating one of those and you have. . . . a simple buck converter!

Now, if your switches are really NPN, then they're not going to current share very well in parallel, so you go to MOSFETS (which have a Rds-on that increases with temperature that results in them current shareing rather nicely) then you are to no more than a high-power buck. {Side Note - in push/pull converters which are called 'buck-derived' as the circuits and equations simplify to a buck, it's not unusual to have as many as 16 or more MOSFETS in parallel to up the power.}

 

[Renovator1]

Comcokid - thanks, I finally got it as a result of your latest point, so you get a star! What cleared this up for me was you pointing out that it is the duty cycle of the inductor that is important, not the switch.

The example circuit was just that: an example. Most of the power semis I see in industrial applications are hockey puck thyristors and IGBT modules. 100A is almost a small-signal current in my world!

So now that it has been established that you must have an inductor for each phase, the obvious next question is, how big do these inductors need to be if the final load is also an inductor? Do you size them to result in a particular amount of ripple current regardless of the inductance of the load (even though it will further reduce the amount of ripple itself)?

At any rate, now at least I know why doing this isn't popular - big inductors are even more expensive than big capacitors!