When you design a power supply, how deeply do you apply control theory? :x

[AlienCircuits]

When you design a power supply, how deeply do you apply control theory?

I am thoroughly confused by power supply designing. I hope people here can help me clear my head and understand what is important before I try to do my first designs. I feel like everyone who I try to learn from is either playing tricks on me, unintentionally withholding information, or they are in the dark and don't want to admit it and may actually be more ignorant than I am. I think some people design feedback systems without knowing anything about control theory, and this frustrates me a lot when I'm trying to learn how to design something with the proper engineering approach.

Maybe I can express the confusion by saying that I have two control theory text books, one over 900 pages and another over 600 pages, and neither gives any information on how a control system works in a power supply design. And I have a 480 page textbook on power electronics that never mentions any controls strategy for regulating outputs. It often gives formulas for computing ripple or output voltage, but it never says how this all works as a feedback system. I don't think it even has terms like PID in its text, and this is supposed to teach me power electronics that supposedly use feedback? All the application notes, tutorials, etc. seem to focus only on one aspect, such as component selection or the fundamentals of how an inductor spike can be used to boost a voltage level. They never really seem to go into how stability and response are designed other than just mentioning of their existence.

Anyway, I wanted to ask experienced designers and experts about this. How much of control theory do you actually use in designs like SMPS or linear regulators? Is it more just the "art" of knowing what a capacitor will do at the output, and you know that in controls it is a compensator, or lag, or lead component in your circuit. Or maybe, do you model your voltage converter system as a linear circuit and then design a transfer function that gives you the desired output with known values of phase margin, gain, etc.?

Also, how are such non-linear devices like SMPSs able to be controlled with linear techniques like negative feedback? How can you even apply proper control theory techniques to such non-linear devices without approximating them to linear models?

- - - Updated - - -

Some other problems and confusion I have:

I can pickout LDOs and I even picked out a buck converter control IC, and I can build these things based just off of what the datasheets and application notes tell me. The problem is that these devices seem to hide all of the details on how to actually design a proper voltage regulator. For example, I know that some LDO datasheets say to use a tantalum cap on the output because its ESR helps to offset a pole . . blah blah blah. . but it never gives insight to how the suggested values were derived.

 

[BradtheRad]

Since you mention switched-mode power supplies...

I was helped greatly by the Falstad simulator, in order to understand their workings (buck, boost, buck-boost).

https://www.falstad.com/circuit/

It is animated, and provides moving oscilloscope displays for whatever component you desire.

His list of circuits does not have a switched-coil supply. I had to create them. It took me several hours of playing with different components.

I found that for a given load, and a given coil, I need to adjust frequency, duty cycle, etc., before things worked properly.

I have not yet tackled automatic voltage regulation. It can be done using an op amp and a voltage controlled oscillator. These must be tailor adjusted to provide the right pulse train to the switching device. I don't believe there is a single setup that will provide a set constant voltage for all combinations of load, coil, supply V, etc.

There are the commercial IC's that control a SMPS. No doubt they contain proprietary circuitry which the manufacturer wants to keep a secret. It's conceivable that authors are in compliance by not telling us all the secrets as to how they achieve voltage regulation, etc., when they write their books.

 

[RF_Jim]

Re: When you design a power supply, how deeply do you apply control theory?

I am thoroughly confused by power supply designing. I hope people here can help me clear my head and understand what is important before I try to do my first designs.

...
Anyway, I wanted to ask experienced designers and experts about this. How much of control theory do you actually use in designs like SMPS or linear regulators?

Seriously? We depend upon the vendor of the part we are using; in the real world, there isn't time to reinvent the wheel ... many vendors offer interactive PSpice simulators to allow one to easily design-in their parts. It is important to understand some control theory though, but one does not need PhD familiarity with the subject these days.

RF_Jim

 

[AlienCircuits]

Well, I have done some research on the web today.

I found a great article by a man named Bob Mammano:
**broken link removed**

That is what I was getting at. I then found lectures from a course:
**broken link removed**

He also shows how a non-linear switching circuit can be modeled. Now I am getting somewhere. I am going to buy the book he uses since his notes are directly from that, and since my book "Power electronics" by Hart has no information on the control aspect of power supply design.

 

[mtwieg]

Read every single app note on this site: https://www.venable.biz/tr-papers.php
You need to submit some info to access the papers, but it's free fast and I've never gotten spam from them. These are easily the most straightforward explanations of SMPS control theory I've ever found. I'll link the first one directly, as a small sample: https://www.venable.biz/tp-01.pdf

 

[D.A.(Tony)Stewart]

I learned from Keith Billings who designed SMPS for Hammond who supplied to us at Burroughs.. I believe he moved form Canada and is a Prof in the USA now and teaches electronic design.

His fundamentals get into ever aspect of every critical part and their effects which include but are not limited to;

Design using torroidal cores to achieve minimum core size without saturation, maximum wire size vs temperature rise, core losses, effective inductance loss with DC current (swinging chokes vs linear chokes), effects of permeability, winding techniques, (minimum loss full wind, single layer windings, or winding to a spec'd temperature rise., insulation & core effects on transformers etc

... Chokes which carry high currents have most of the energy in the air gap , not the core.

He uses the fundamentals of applying Faraday's law, Ampere's law, the BH magnetic relationship, using continuous mode, dis-continuous mode switching/

In my opinion, the SMPS convertor needs more understanding on the analog non-ideal properties, than the control aspects which are also importance with feedback on primary winding current, secondary voltage and secondary current.

These basics of magnetic storage, in addition to the real properties of capacitors with ESR, inductance and leakage. copper utilization factors, the real properties of diodes * transistors with capacitance and storage charge, non-linear gain and transconductance in addition to the magnetic core losses are critical before thinking about control systems gain, phase margin with pre-load stabilizer, and multiple tracking regulators. ... derivation of area product of core in transformer design, window utilization factors, current factor, topology factors, winding area factor, skin effects, harmonic losses, proximity effects.

He uses nomographs that he has developed for every choice in values.. I recommend his book on Switchmode Power Supply Handbook.

He also made amazing products.

With all these analog factors to learn,, no wonder you were confused about it simply being a control system. It is the control of each part that counts.

 

[AlienCircuits]

Thanks for the information and links! I'll try to digest as much as I can today, so I might have more questions if you want to answer.

- - - Updated - - -

With all these analog factors to learn,, no wonder you were confused about it simply being a control system. It is the control of each part that counts.

That is what I have had the suspicion of. I am fairly comfortable with analog electronics, and I am fairly comfortable with control theory, but I have never seen them applied (I'm a newb) together in a straightforward way with voltage regulators. I think the new generation of electronics people get spoon fed so much design that they never need to consider the design process at much detail anymore, and that makes me upset. No way am I going to let National instruments webdesigner pick out my parts for a buck regulator and tell me how stable the design will be, even if its right. That would be against my engineering philosophy.

 

[RF_Jim]

No way am I going to let National instruments webdesigner pick out my parts for a buck regulator and tell me how stable the design will be, even if its right.

One more reply, if I may ...

As schedules and money get 'tight' investors and bosses during 'product review' can, with little doubt, change one's mind. At some point everyone can use every bit of 'help' and assistance that the part's vendor through their app engineers and automated design aids can provide.

See, there are other issues, such as parts procurement (actually GETTING the part one wants, in quantity, from 'approved' sources as allowed by parts procurement ppl in the company), on-going 'quality assurance' issues that the right parts are actually being procured by the parts specialist (all parts on all reels are not necessarily equal! We have gotten SMPS FETs in that had Gate Vth enough outside of spec on most parts that the SMPS would not start!), and then on-going issues with manufacturing and variances with build quality including solder reflow processes and board surface contamination as well as parts 'aging' and moisture absorption (which becomes an issue as parts outgas that moisture during trip through the reflow oven) and no one above I see has mentioned meeting the various EMI and RFI specs including conducted as well as radiated EMI (usually these issues crop when product is 98% complete and the first EMI preliminary 'screening' tests are performed) - all these issues come back to the cognizant design engineer sooner or later and at that point one can use all the outside 'help' from automated 'tools' and design aids as one can get one's hands on ... (and I haven't mentioned the perils of environmental testing of product over temperature, physical shock (shaker table) or ESD and 'lighting' discharge testing!)

Unless, of course, this is just an academic exercise ... then we can concentrate on the 'purity' of our design processes and form the basis for writing a strikingly-good white paper on PS design ...

Regards,
RF_Jim

 

[AlienCircuits]

The problem to me is that if all of these auxiliary issues of procurement, manufacturing, and compliance are coming up, and I only have a webdesigner tool to fall on, then I'm really in trouble. I would rather know how the design is working, why the components are used, etc so that I can quickly and accurately fix problems. I don't want to tell my boss "well, the webdesigner tool told me to use this inductor, I don't know why its radiating so much", because that would just show incompetence. I also have another engineering philosophy that theory is much more relevant than what some die-hard hand-on engineers like to distinguish themselves from. Sure you can call it an art, but that doesn't mean its mystical like some pretend it is.

At an even more motivating level, I want to know how these things work so that I can create new designs and ideas that progress the field.

 

[D.A.(Tony)Stewart]

inductors are noisy radiators unless they are torroids or shielded. Use a shorted turn probe or simply a probe ground on tip as antenna to sniff for EMI leakage and consider a torroid vs cylindrical pr E type. Dont be afraid to admit technical problem that you are resolving, you may need better tools so ask. I dont know all your symptoms, but if EMC, then you need to characterize the symptoms better.

 

[AlienCircuits]

inductors are noisy radiators unless they are torroids or shielded. Use a shorted turn probe or simply a probe ground on tip as antenna to sniff for EMI leakage and consider a torroid vs cylindrical pr E type. Dont be afraid to admit technical problem that you are resolving, you may need better tools so ask. I dont know all your symptoms, but if EMC, then you need to characterize the symptoms better.

Sorry if I mislead you from reading my last post. I have not designed a SMPS yet (although I did build one like a monkey following instructions before), so I don't have any of these problems. My post was in reply to the person above me to explain why I'm not comfortable using webdesigner or other tools that do the design for me.

 

[ZekeR]

Anyway, I wanted to ask experienced designers and experts about this. How much of control theory do you actually use in designs like SMPS or linear regulators? Is it more just the "art" of knowing what a capacitor will do at the output, and you know that in controls it is a compensator, or lag, or lead component in your circuit. Or maybe, do you model your voltage converter system as a linear circuit and then design a transfer function that gives you the desired output with known values of phase margin, gain, etc.?

Also, how are such non-linear devices like SMPSs able to be controlled with linear techniques like negative feedback? How can you even apply proper control theory techniques to such non-linear devices without approximating them to linear models?

Yes, when designing nonlinear systems (SMPS's, op-amps, LDO's, ... just about everything's nonlinear), you "linearize" the operation about a certain point (or at the corner cases). Performing an AC analysis, you tweak the controller compensation so it's stable over the entire range of desired operation, which means it may be optimal compensation at one corner, but sub-optimal (highly conservative) at another corner. Stability's most important, after all, so it's better to be conservative at some points rather than unstable at others. A rule of thumb is to ensure you have at least 60 degrees phase margin over all measured conditions, as component variation will certainly affect your stability.

Generally, the act of design requires some knowledge of how things will behave before you start. And I don't (necessarily) mean that you need to have everything perfectly calculated or all of the analysis done before you start; you just need to know that, for example, a high impedance behaves like a current source, and when you feed a current source into a capacitor (such as the output capacitor), the current-to-voltage transfer function will be an integral (i.e., a pole), with a scaling factor dependent on the capacitance. When designing voltage mode bucks, you should know that it's a second-order system and it's already unstable before you start because you have 180 degree lag as soon as you close the loop, so you need to introduce a zero to get enough phase margin; current-mode bucks behave as a current source output, and therefore the transfer function from control node to output voltage is a single-pole system. With boosts and flybacks, you should know that the right-half-plane (RHP) zero is a limiting factor in stability. When designing R-C compensation, you should know how to introduce poles and zeroes as appropriate to increase/decrease phase and gain as needed. etc. etc.

Equipped with the right mental tools, you can build a linear model (with values from your textbooks, or you can derive it if you so choose), plug the values in, and tweak your components until you get the desired phase margin. You could go further to derive equations that give you the optimal components, but you'd spend a lot of time fiddling with algebra (and component variation would make them non-optimal anyway), so it's generally more pragmatic to let your simulator tell you the results and make reasonable adjustments. For example, if you have no phase margin with a 2nd order rolloff, you can introduce lead compensation to bump up your phase. We know the maximum phase boost happens at the geometric mean between zero and pole frequencies, and you can do the exact calculation if you're so inclined... or you can simply insert a component that's in the right ballpark, run the ac analysis sim, and see if the phase boost was where you wanted it... and then iterate to walk it in until you get there.

Once you've figured out what compensation values to use from the linearized model, it's time to try the real thing. Plug the components into a real switching regulator, and perform a load step transient on it. If it appears well-behaved (any ringing dies down within one cycle) over all corners of operation, then you're safe. Some people would insist that you take another Bode plot, but it turns out that a loop meeting the Nyquist criterion during a frequency analysis actually does not guarantee system stability; the transient response is the real test. The primary thing Bode plots give you is a convenient design tool which lets you know what components to tweak and how.

So, yeah... Designers use control theory, but we try not to be too theoretical with it. If that makes sense. The long mathematical derivations can be left to Ph.D.'s in ivory towers. In real life, a ceramic 22uF capacitor will derate to 10uF at its rated voltage; your inductor will vary by +/- 20%, your switching frequency will be off by 15%, and the amplifier gain will be 50% high. Taking all of these things into account cannot be reasonably done with analytical analysis, but by generous margining, rules of thumb, and building an actual prototype (with the flexibility to tweak anything if needed). Performing parametric simulations (such as corners simulations, monte carlo, etc) can help verify that your margins are enough, but many experienced designers will just get a switching regulator, attach some variable resistors and capacitors, and tweak them until the step response looks good.

A really good designer I know once told me, "Rocket science is easy: you just need to make it close enough, and be able to tweak it when you get there."

Hope this helps.

 

[D.A.(Tony)Stewart]

If you want some prudent advice, learn the non-ideal characteristics of passive parts. There are many. I learnt from reading the library on MIl-Std Handbooks for passive parts and they served me well.

- You can also setup a test jig to look at the impedance in a bridge circuit plotting V vs I on a XY scope and sweep the frequency with DC bias and change amplitudes to see how the values change. Change the reference resistor in the bridge to near ESR values to show ESR more clearly.

- Use heat guns and cold spray to quickly change its values.
- Tap the part to see how microphonic it is and check to see how inductive capacitors are by sweeping past its Self Resonant frequency.
- Learn more about the BH loop and the ET product and the LI orstead values. See how losses are non-linear with frequency.
- Get parts from a good Japanese company that characterizes impedance vs freq.
- One thing designers often fail to recognize is how margins affect yields until they go into production.
- Monte Carlo and Design of Experiment (DOE) with Taguchi methods changing many parts by +/- rated tolerance to test the outcome.
- Some do not even know what tools exist to do this in simulation.
- Many do not realize how to build in analog DFT circuits cheaply to self test or make automated functional test better by indicating margin to failure.
- There is a statistical parameter every design should know and that is Cpk and how to measure it.
- It basically n-sigma with upper-lower control limits to give a figure of merit. Rather than yield on prototypes, Cpk is the better measure of a good design.
- It predicts how many will failure with a sample number of units for every test parameter, based on tolerances. Cpk>>1 is the goal
- Random parts are best, not from the same batch.

- In your case I would use differential amplifier to monitor current and voltage across your magnetics to ensure they are operating in a fairly linear zone and not saturating the core. You can use the XY scope for this if you put the sense circuit close and use short twisted pair to the sense circuit for V & I across the magnetic part.
- most failures in production are solder related or wrong parts, not bad parts, but that can happen. the reliability of parts normally far exceeds the reliability of the design. Get used to it until you get good at it.
YOu have a lot to learn, no time like now to start.

 

[ZekeR]

If you some to prudent advice, learn the non-ideal characteristics of passive parts. There are many. I learnt from reading the library on MIl-Std Handbooks for passive parts and they served me well.

...

- Use heat guns and cold spray to quickly change its values.

Ahh yes, forgot to mention that part. Probably the most important thing that experience gives you is not the knowledge of what is possible and achievable, but rather the knowledge of what can possibly go wrong and how to avoid it. In school you learn lots of neat theory on exotic techniques (using estimators, finding eigenvalues to determine the ideal control law matrix, and doing exact pole-zero cancellation for example), but in life you learn that, for the most part, only the most basic methods are practical because you have many other practical challenges to deal with.

As for cold spray, it's not the best thing to do, but I normally use compressed air (turned upside-down), because I'm a cheap like that. Heating and cooling your circuits to see how they respond is ESSENTIAL. And you should probe your inductor current too, because some inductors will saturate early when you heat them up (and when inductors saturate, things tend to blow up).

 

[D.A.(Tony)Stewart]

I guess cold spray is old school but I used it sparingly by only letting it go in short spurts or drips.
For heating IC's I would pulse modulate a soldering iron and learnt how to calibrate my fingers for leakage capacitance to simulate dust to look for spurious signals. COrnformal spray is a good idea but will add capacitance coupling of a many pF between pads but may protect it from environmental leakage from moisture, if done right. Use polyurethane globs to secure large magnetic parts from vibration and shock.

 

[AMSA84]

Hi guys,

I read this topic and the comments and something came up.

For those who use the ICs that are in the market to do the control part of the circuit, this control theory stuff really doesn't matter at all for the design, right? (apart the fact that we need to know which IC to use, of course)

Now if we want to design a linear or a non-linear controller from scratch, using for example OPAMPs to do PID controllers or whatever, taking into account the converter model, the feedback loop, etc, this theory stuff is important.

Do you understand what I mean?

I am right?

Well, I have done some research on the web today.

I found a great article by a man named Bob Mammano:
**broken link removed**

That is what I was getting at. I then found lectures from a course:
**broken link removed**

He also shows how a non-linear switching circuit can be modeled. Now I am getting somewhere. I am going to buy the book he uses since his notes are directly from that, and since my book "Power electronics" by Hart has no information on the control aspect of power supply design.

 

[mtwieg]

For those who use the ICs that are in the market to do the control part of the circuit, this control theory stuff really doesn't matter at all for the design, right? (apart the fact that we need to know which IC to use, of course)

There is practically always some control design to be done, even on specialized control ICs. Most will still have error amplifiers to compensate, at the very least. But they will often guide you in designing that part so you don't have think about it too much.

The only cases where you don't have to manually do some control analysis would be in some linear controllers where the compensation is fixed internally (I've seen this in some LED controllers) and in controllers with very simple and primitive control schemes, like hysteretic control or bang bang control.

 

[AMSA84]

Hi there again.

I see the point. But grabbing on the first topic, that control theory is only important then for those who design THAT kind of specialized ICs, right?

 

[AlienCircuits]

There is practically always some control design to be done, even on specialized control ICs. Most will still have error amplifiers to compensate, at the very least. But they will often guide you in designing that part so you don't have think about it too much.

The only cases where you don't have to manually do some control analysis would be in some linear controllers where the compensation is fixed internally (I've seen this in some LED controllers) and in controllers with very simple and primitive control schemes, like hysteretic control or bang bang control.

I built a buck-boost power supply with a national IC, and the datasheet pretty much told me the formulas to pick out all the parts. I did not use 1 thing I learned in control theory to design my SMPS then and it worked, at least in the lab. This is where my problem lies - I didn't really learn anything by building it because it already told me everything to do.

 

[AMSA84]

Hi there again.

I see the point. But grabbing on the first topic, that control theory is only important then for those who design THAT kind of specialized ICs, right?