smps voltage feedback / General Science and Electronics / Forums

Forums

4hv.org :: Forums :: General Science and Electronics

« Previous topic | Next topic »

smps voltage feedback

previous 1 2 3 next

Move Thread

LAN_403

GeordieBoy

Mon Feb 15 2010, 11:46AM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

Closed-loop control is a big subject. Perhaps a basic starting point for compensating buck derived converters is this article:

"AN012 A Handy Method to Obtain Satisfactory Response of Buck Converter"

You should be able to find a PDF link to this article. There are also a number of other excellent documents written by a bloke called "Lloyd Dixon" on the subject of compensating the feedback loops of various switched-mode power supplies. I would definately recommend browsing them at least to get a feel for what it's all about.

All you are really trying to do is make a feedback loop that has loads of gain at DC to achieve good regulation (the "loads of gain" acts to minimise the error between what the output voltage is and what you wanted it to be...) BUT the feedback loop must be designed so that it doesn't have too much gain or phase shift at higher frequencies around the point where the converter's output filter peaks and causes a rapid phase shift. So, in short you want as much gain as possible at low frequencies (integrator action) and then you want to roll off the gain in a controlled manner at some high frequency in such as way that the gain crosses unity before the output filter kicks in with its 180 degrees of phase shift.

If you get this right you get a SMPSU that has excellent steady-state regulation to line and load variations, and it's output returns quickly to the correct value with no overshoot or ringing when there are step changes to the line voltage or load voltage. If the design is not right then the DC regulation can be poor, there can be AC ripple remaining at the output, and the output voltage may overshoot or ring when the load changes abruptly. The worst possible scenario is sustained oscillation where the output voltage flails around wildly between 0v and the maximum output voltage possible.

Do a search for terms like "Type-2 compensator". Type-2 is the simplest practical type that can work with a buck converter if it has enough open-loop phase margin. Type-3 is better if the output filter is underdamped as it provides a phase boost right where it is needed most around the double-poles of the output filter.

There are a few other things I would recommend for a 2.5kW SMPSU. Firstly, that is a lot of power to process from a single line with now active PFC. It will draw A LOT of line current, particularly if configured for 110V!!!

You don't need the additional 3900uF C24 capacitor. When you run the supply open-loop the line regulation is meant to be poor. So you would expect to see plenty of ripple at the 50V output when it is fully loaded up. Resist the temptation to put loads of capacitance at the input side to remove the ripple, that's not how it's meant to work. The closed-loop control will regulate out the ripple once you have it working properly. Putting massive capacitors on the HV side after the rectifier just wrecks the power-factor and makes the supply draw lots of current. It also stores more energy making the bang bigger if anything fails s/c.

When designing the transformer you need to make sure that you can still get 50V output at full load current even when the mains supply is at it's minimum value and the DC bus voltage is at a dip. i.e. You need to take the DC bus voltage ripple into account. The worst case conditions are full load, minimum supply voltage, minimum supply frequency (50Hz) and when everything is warm (maximum copper resistance.) If you can still achieve regulated output under these conditions then you have enough turns on the secondary of the transformer.

The switching frequency marked on the schematic (200kHz) is way to high for a 2.5kW supply. Something around 30-40kHz is in the right ballpark for a hard-switched converter at this power level. The phase-shifted ZVT conveter is common at this power level and would give better efficiency but is understandably more complex to design.

I'd also look to beef up C6 to 2.2uF, and replace D1, D6, D7 and D8 with schottky diodes that can support the gate-discharge current pulses of a few amps. Something like the 40V version of the 1N5818 diodes already used for the GDT clamping and CT rectifier would work better for gate turn-off speed up diodes too.

Finally the output current of 50A is quite high. You will probably find that you need to make up the output capacitance of C15 from many capacitors connected in parallel. Not to get sufficient filter capacitance but to get sufficiently high current handling capability, and to get the ESR low enough. Above the resonant frequency of the output filter the filter acts like a LR filter and the output ripple voltage is determined by the inductor ripple current multiplied by the capacitor ESR. Therefore it is important to keep the ESR reasonably low if output ripple voltage is an important design parameter.

If output ripple voltage needs to be really really small then you can always use an additional post filter, but put this outside of the feedback loop otherwise it's additional phase-shift will make feedback compensation almost impossible!!!

I hope this info helps,

-Richie,

KG7HF

Mon Feb 15 2010, 01:52PM

Registered Member #2683 Joined: Sun Feb 14 2010, 12:27AM
Location:
Posts: 14

Hi Richie,

Thanks very much for the design review and comments. I've looked a bit for "type-2 compensator" and have found some very insightful information. For now I'm off and researching. To your other points...

"...that is a lot of power to process ... It will draw A LOT of line current, particularly if configured for 110V!!!"

Yes, it is a lot. I included it in the design for two main reasons. Firstly, I only have a 110V isolation transformer and 110V variac. Secondly, I want to be able to use the hbridge board and controller for other smaller power supply projects, it makes the board more versitle. This is already implemented in pcb format, although rev .1 has a number of mistakes!

"You don't need the additional 3900uF C24 capacitor..."

That seems to be the case, regulation is somewhat acceptable without the 3900uF, I saw the 3900uf being used in another design which made me think I might need it. They are really expensive caps so I only included the footprint on the board.

"When designing the transformer ..."

I think you picked up on the documentation mistake. The posted schematic is 40:6 which would yeild about 51 V output, then with the diode drop etc is not sufficient. The actual implementation is 40:7 which is about 59 volts. I might go back to 6 turns and reduce the primary windings to about 34 turns. 40 primary turns is really hard to make fit.

"The switching frequency..."
200 kHz is what it is currently running at. Changing now to a frequency drastically lower would be a lot of work. The 200 kHz makes the magnetics much smaller. Of course for a 50V I could implement ZVS, however I also plan on using this same configuration for much higher voltages which I don't think could be implmented in a ZVS configuration. Also, the basis of this project came from a 100 kHz high power converter designed in the 1980's.

"I'd also look to beef up C6 to 2.2uF..."

Yes, I also thought the same thing.

"...replace...D1, D6, D7 and D8 with schottky diodes..."

Yes, actually another documentation mistake, they are 1N5818 diodes already.

"...capacitance of C15 from many capacitors connected in parallel..."

Yes, I too thought the same thing, finding a single cap of 100uF/100V with low enough esr is expensive or impossible. But finding 10 10uF/10V caps is easier and cheaper, plus they are small too so it is not a big deal on space.

Again, thanks Ritchie, the terms you used have put me on the hunt again. Plus, further analysis indicates the small output "rings" in the feedback look seem to make the system misbehave so being able to dampen them might be the key.

Paul

Steve Conner

Mon Feb 15 2010, 02:02PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Thanks Richie! :)

Also, you don't want C1. It'll add an extra 90 degrees lag to the loop and destabilize it. If you must have it to filter switching noise, it should be nearer 100pF than 100nF.

And most importantly, you don't have any compensation network. This is usually a series RC from the UC3526's compensation pin to ground, which I think is what Richie means by the Type 2 compensation. Or maybe there's a third capacitor in there. Anyway, the power supply will be disastrously unstable without one! You mustn't try to run it closed loop without one! Richie linked some documents that explain how to calculate the values, so RTFDS!

Or take the short answer, 2.2k in series with 1uF.

This should be massively overcompensated, so it'll be stable, but have lousy transient response. When working with current-mode DC-DCs, which are easier to compensate than voltage-mode ones, I just start with those values and work my way down until I get acceptable transient response. Talking of which, you had better get set up to do load-step tests, too! When you let go the PTT, you don't want the output voltage to spike up and blow your 200 dollar RF MOSFETs.

KG7HF

Mon Feb 15 2010, 02:20PM

Registered Member #2683 Joined: Sun Feb 14 2010, 12:27AM
Location:
Posts: 14

Hi Steve,

I've still got to read up on things, but you are saying 2.2k in series with 1uF from the comp pin back to the E- pin?

Paul

Steve Conner

Mon Feb 15 2010, 02:21PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

No, the compensation network is a different thing. It goes from the comp pin to ground.

GeordieBoy

Mon Feb 15 2010, 02:44PM

Registered Member #1232 Joined: Wed Jan 16 2008, 10:53PM
Location: Doon tha Toon!
Posts: 881

In my opinion 200kHz is still very high for a 2.5kW converter. It would even be high for a soft-switched converter at this power level. There's no point in having tiny magnetics if you need huge heatsinks on the MOSFETs and diodes to dissipate the resulting switching losses. But, different engineers design things differently. I like to keep the semiconductors as cool as possible.

Regarding the choice of output capacitor/capacitors, this has a big influence on the stabilisation of the control loop too. Low ESR is good for reducing output voltage ripple as low as possible, but can make control loop stabilisation more challanging. This is because with no load the output filter will have very low loss and high Q. Therefore there will be a large resonant peak and a very rapid phase shift at it's resonant frequency. Both of these act to reduce the margins in the feedback loop and push it closer to instability.

This document shows the two most common compensation networks for a voltage-mode buck derived converter:

The type-II compensation network is so called because it has 2 break frequencies. The lower break frequency is chosen below the loop crossover frequency where gain falls to unity. The integral action that results below the zero ensures excellent steady-state DC regulation, and the phase-lead of the zero itself helps to offset some of the phase-lag from the SMPSU output filter. This acts to keep the negative feedback loop stable. The rolloff above the upper pole helps to attenuate switching noise. The phase lag from this pole doesn't significantly degrade stability because it is well over the loop crossover frequency anyway.

However, if the output filter is very under-damped (very low output ESR) then the the Type-III compensator has two zero's which can provide up to 90 degrees more phase lead (180 in total) where it is needed to offset the lag of the output filter's double poles. That paper unfortunately doesn't show the phase response, but the important thing about the Type-II compensater for SMPSU use is that it has an "adjustable" phase boost in the mid-band of the frequency range. By juggling the position of the compensator poles and zeros you can dial in as much phase-lead as you need to make sure the control loop is unconditionally stable, even with the most horrible of output filter responses!

The part of that paper where they are talking about the open-loop response being either -20dB/dec or -40dB/dec at crossover depends on where the zero resulting from the output capacitor(s) ESR is located. If the output filter ESR is very low then the zero in the frequency response occurs at a very high frequency (much above the resonant amplitude peak of the filter) and you see the full 180 degrees lag of the filter kick in, accompanied by a steep -40dB/dec slope that stays at that slope (-2) If the ESR is large then the zero frequency is lower. In this case you see the resonant peak of the double poles: The amplitude slope starts out at -40dB/dec and the phase slews towards 180 degrees, but then the effect of the ESR zero kicks in and tames the amplitude slope to -20dB/dec and the phase swings back to 90 degrees before it ever got as far as 180 degrees. This later situation is easier to compensate!

As Steve said, you need to get prepared for doing at least three things to work through control loop compensation:

1. Characterising the real open loop response (i.e. plotting a bode plot if you have to!)

2. Working through the maths, or playing about on a circuit simulator to arrive at a design you think should work in the frequency domain.

3. Performing steady-state and load-step tests on an implementation of the real design to see if it meets your expectations!

Many things like permeability swing of powdered-iron cores, temperature dependent ESR of electrolytics and component tolerances can mean that the real life implemenation doesn't perform as well as the mathematical model. Good control system design if often an iterative process of design, test, modify, etc...

Where have you placed the continuous-discontinuous current boundary for the inductor current in your supply?

-Richie,

KG7HF

Mon Feb 15 2010, 03:18PM

Registered Member #2683 Joined: Sun Feb 14 2010, 12:27AM
Location:
Posts: 14

Thanks Richie, that's actually the first one I clicked on in my search.

"Where have you placed the continuous-discontinuous current boundary for the inductor current in your supply?"

Oh boy, I'm about to get it now. I have no idea what you are talking about. For 6 months, I have read just about everything on the web, looked at hundreds of designs, studied app notes, magnetics design notes, ferite and powder iron core information, any information on half or full bridge systems I could find and I have not run across the continuous or discontinuous current terms. Is it possible it's hidden in subtle other terms

Steve Conner

Mon Feb 15 2010, 03:40PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

KG7HF wrote ...

"Where have you placed the continuous-discontinuous current boundary for the inductor current in your supply?"

If you can't find it, maybe you left it in your other trouser pocket and it's been through the wash.

Have you seen some other DIY SMPS designs on the web? Ralph Hartwell's 1.5kV one is the best I can think of, but there are some scratch-built 13.8V types too. If you don't get the theory, maybe you can just copy the compensation circuit from one of these and hope it works. The circuit is basically the same, whether it's the TL494 or UC3526.

KG7HF

Mon Feb 15 2010, 03:53PM

Registered Member #2683 Joined: Sun Feb 14 2010, 12:27AM
Location:
Posts: 14

Yea, I know Ralph, we've exchanged a few emails and shared some ideas, before I got to this point though. Of course his circuit is for 2Kv and he's using a TL494 but yes, I've incorporated a bit of his design. I've also incorporatd a bit of Steve Wards design as well as a bit from Tim Hullick (well Tim was the initial inspiration), and I've stolen stuff from Ray Mack (Demystifying Switching Power Supplies), his book is falling apart now, but still good.

I thought I did a pretty good job of learning it, but of course, it is the first one. I re-read the chapter on transformers and inductors and didn't see once any refrence to continuous or discontinuous. Doesn't mean it isnt there though.

Sometimes I feel I've been through the wash and ringer on this project

, Since it's the first real big design I have done, that can be expected in a non cookie cutter project. The point is to actually try to understand it and become a better diy'er. I don't feel I need to know how the atoms move, but a basic grasp of some higher level concepts should be enough to get a working project and then drill down when and where necessary.

Steve Conner

Mon Feb 15 2010, 04:09PM

Registered Member #30 Joined: Fri Feb 03 2006, 10:52AM
Location: Glasgow, Scotland
Posts: 6706

Well, why would continuous/discontinuous mode be in the chapter on inductors? It's a property of the system as a whole, not bound to any particular component. CCM and DCM are defined here

but they're basic concepts of SMPS theory and should be in your book somewhere.

An easy way to explain it is in terms of ripple. If the current in your output filter inductor never falls to zero, that's CCM. If you try to calculate the ripple and come up with an answer more than 100%, then you know it falls to zero and you're in DCM.

When it comes to things like feedback and stability, you need to get used to working on the systems level using control theory, instead of thinking about individual components.

previous 1 2 3 next

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob