compensation modelling

[krish2487]

Hello all.

From my previous posts u can see that i am a newbie
and still havent changed much


i do have another doubt though

i was trying to understand compensation and came across several interesting papers.

As i understand it

1. The feedback mostly is taken only after the filter in the output(if secondary side control is implemented)
2. The filter causes the initial poles and zeros which we try to compensate

I did read about the various types of compensation
and did manage to get a hold of a spreadsheet on calculating the compensating components.

Say if i were trying to design a compensation network
Then i would have to initially find the response of the system right

Anybody has any tips on how to do this?
say
like simply running a frequency sweep and finding out the Opel loop response or such

any help is appreciated.

[darkfenriz]

SMPS control theory is basically the same as for linear ciruits. All you need to mentain stability is loop gain of less then 1 at 180 degrees phase shift.
Simulating AC sweep with linear circuits like op-amps and L, C, R is very helpful here.

Some common practical results are:
- LC filter causes 2 poles, which very quickly approaches 180 degrees. The major problem for SMPS compensation. For electrolytic capacitors the ESR creates a "useful" zero. For ceramic caps ESR is negligable and some form of phase-lead network is needed
- optocoupler, switching delay and parasitics create additional poles, typically above LC corner frequency and can further limit feedback gain*bandwidth
- current mode control cancels one pole, because it creates a virtual current source feeding a capacitor. This works as long as current loop bandwidth can correct (less then a half of switching frequency). But still current control of some form helps stability a lot and increases loop bandwidth.

Keep in mind, that one pole in the loop is actually a very convenient situation, because it creates only 90 degrees phase shift while it lowers the gain for higher frequencies. This is called a "dominant pole compensation" of feedback loop and is widely used wherever possible (op-amps, audio amps, PLLs,...).

So a compensation network is very application specific and very often needs a trials&errors approach.

[krish2487]

thank you

all the papers i read explained how to compensate a error amp in a given situation
but did not give any kind of info on how to find the systems properties itself


just a doubt

thank you very much.

[cerrem]

Originally Posted by krish2487

thank you

all the papers i read explained how to compensate a error amp in a given situation
but did not give any kind of info on how to find the systems properties itself


just a doubt

thank you very much.

What is your exact application..... Give details of your topology..
Control theory is a must to know...but real life converters have can have additional mess of things that can be sorted out with some extra effort not found in most textbooks...
For example most current mode converters have slope compensation, which is really is a mix of voltage mode with current mode, then throw in slave outputs with coupled inductors and steering inductors and then the transfer function is not straight forward like in a textbook...
The key is to first figure out your OPEN LOOP response on a bode plot, then you can superimpose the feedback compensation over this to get the desired CLOSED LOOP response...This is the most straight forward approach...Or you can derive the transfer functions if you have the time...
Chris

[darkfenriz]

My feeling is that intuitional understanding of control thoery is good enough. Calculating or simulating detailed transfer function is loss of time, one should instead 'feel' the control theory to make out a solution in a real life situation.
Compensating instability or snubbing is always a bit of trials and errors, the more you know and the deeper you understand the circuit, the less trials you need.

[krish2487]

sorry for the delayed response.
couldnt log on for a couple of days.

@cerrem
no specific application as such
as mentioned in an other post
i want to design, develop and debug a 100W push pull boost converter
the other parameters are defined but when it is the question of compensation i m stuck as to how to get the open loop response
I am using a 3525
i have read about control systems and specifically loop compensation techniques
but i did not understand on what basis did they ( the author of the papers) arrive at the initial open loop response
hence the question

@darkfenriz
i am by no means an authority on control systems and compensation but i disagree
its a rather tediousand intricate process not to mention trial and error but i believe compensation techniques have to be tailored to a specific application
a 'feel' of the system would by no means, IMHO, be a good substitute for a documented performance report.
maybe when one gets to the stage where they have enough experience, like you i assume, probably then a feeling might be a real close estimate to the optimum performance of the system.

But until then , we lesser mortals, are stuck with the pen and paper.

[darkfenriz]

If you design a push-pull design with two mosfets, center-tapped transformer and 3525, then a good starting points can be:
- assume transformer to be perfect in terms of no parasitics
- transfer is dominated by LC filter characteristics
- decide for switching frequency first, based on transformer design, calculate output L for a given current ripple (20% to 40% of DC current can be generally good fraction) and choose C for a voltage ripple you can accept, if you must use electrolytic, the ESR can well be dominant for impedance at switching frequency, calculate power loss in a capacitor from ripple current and ESR to be sure you don't overheat it
- calculate LC filter corner and C-to-ESR corner frequency
- so-called type II compensation with lead network is usually adequate
- 3525 has an op-amp in it, design a lead network and integration factor to get a combined transfer of L-C-ESR-phase_lead-integration, so that it never exceeds some 120-150 degrees
- open loop is basically gain of an opamp times gain of PWM stage (input voltage swing to ramp signal) times transformer turns ratio

I hope this helps,
regards,
Adam

[krish2487]

yes
thats quite a lot of help


that will keep me busy for the next few days...

thank you