Transformer Design

[NickS]

...The ST Micro L6384-L6387 series aren’t controllers at all – they are an older family of high-voltage gate drivers – suitable for high or low-side gate drive of FETs run off a high-voltage rail.

Sorry about that. I am not sure where I pulled those part numbers from but I was actually
looking at ST Micro's setup for L6599 and L6598.

So all of the controllers I am finding for off line applications are resonant half bridge which seems to be only different in regard to an additional reactive component(capacitor or inductor) before the transformer. What are the advantages/trade-offs.

I have another controller to add to the list from linear Tech though it doesn't look like its technically a half bridge topology(LT1509).

One thing that has been bothering me came to light when reading AN2530 from ST Micro regarding the L6598(see attached plot). The converter efficiency takes a massive hit when PFC is added. Most remarkably in the range of 30-80W. In your experience is this typical? I would say that I usually would be using this lower range of power with an occasional spike up to the higher levels so this efficiency trend is really backward of what would be ideal.

 

[Militoy]

The main advantage of the resonant converter is the lower stress on the FETs achieved by forcing switching at the zero-voltage or zero-current point of crossover. An additional advantage can be lower EMI. A conventional half bridge is typically forced to commutate when there is considerable voltage or current across the switching elements – fast moving current in a circuit with natural parasitics can result in high H-field radiation; fast-moving charges result in high E-field radiation.

Some disadvantages are the nature of variable-frequency control, and the more finicky feedback control. When your supply shifts frequency for control – you must be aware of the possibility of interference at a range of frequencies and all their harmonics – as opposed to just one. More likely to find the “sweet spot” that affects some surrounding circuit, if you’re moving all over the map in op frequency.

The chart you have posted looks about right for the power range. An efficiency chart of this kind can be very misleading – if you don’t think it through. By definition, a power supply will have lower efficiency at lower output power. Efficiency is defined as Output Power / Input Power x 100%. When the numerator of this equation reaches zero (no load drawn) – the efficiency will be zero %. There will always be some fixed losses associated with just running the circuit and lighting up the magnetics. As the load approaches zero, these fixed losses will become a greater percentage of total power input. You have to look at the problem with some perspective – and realize that total power losses are decreasing as you decrease the output load. Since the supply and all of its components are designed to effectively shed all the watts lost as heat at full load – the situation will only improve as the load is decreased. The efficiency is ALWAYS rated, specified and certified at full output load. In the case of running a converter off of batteries – you can see the advantage of not specifying more output power capability that is actually needed in the system – unless power expansion is necessary to accommodate anticipated circuit upgrades.

 

[NickS]

I should clarify that I am ware of the natural decrease in efficiency as the rated output load drops. That aside the above plot is actually very good on efficiency at lower loads for the non PFC line, it holds nearly linear from 33% up to 100%. What I was actually trying to target with that question was the negative impact that PFC has on the low end efficiency.

I guess I was not expecting the PFC to reduce efficiency and I was not expecting it to exacerbate the already present low efficiency at low power usage. Perhaps some perspective would be useful here as well. It is doubtful that the efficiency hit in the exchange between the mains and the switcher is part of this plotted data. So perhaps with that factored in the two plots would flip flop on an overall efficiency scale.

Moving on. I am anxious to pick a controller. Do you have any additional input regarding that?

-Nick-

 

[Mitchekj]

Any conversion comes with an efficiency cost. The same holds true for a boost PFC stage. That alone will knock 5-10% off the top end, at best.

Part of the reason I'm liking passive PFC more and more... do you really need that 0.999 PF, or is something like 0.97 or 0.98 good enough?

 

[NickS]

Well depending on the controller we pick, some have built in PFC. Otherwise It doesn't have to be active. What sort of efficiency improvements do you see using passive? Do you have a good resource I can look at for passive PFC?

 

[Mitchekj]

If you were still worried about volume, then the normal passive methods would probably turn you off (bulky to deal w/ the low 60Hz.)

For the low-power flybacks we've done, it's been enough to get a 0.98 PF, and keep above the 80% efficiency mark just by using a critical conduction mode, but this relied on having small (really small) bulk cap values... basically forcing the FET to follow the input voltage. That doesn't impact efficiency all that much, less then a boost stage would anyhow. Thinking about it more, I don't know if that's doable in the forward or half-bridge.

Anyway, I should just shut up and let Militoy teach me something. Looks like he's already touching on it with the zero-voltage switching he's mentioned. I know enough to know that I don't know all that much yet.

 

[Militoy]

Nick - I'm a bit 'zonked' this evening travelling back home, but I should be able to jump onto the site with a clear head sometime tomorrow AM. I ALWAYS lean towards a design with PFC - but that's partly because 99% of our customers demand it - for every power level. In any case, we'll concentrate on one section at a time to keep focused. No problem just going forward with the LLC controller we've discussed - but if you want to go with a fixed-frequency 2-transistor forward - I'll let you know my favorite - the one I'm currently using in this power range is the UCC1801 (I think the low-temp versions are the 2801 and 3801). For PFC in this power range, I'm pretty satisfied with ST Micro's L4981.

Passive PFC will be "massive PFC" at 60 Hz - and will only be effective for a narrow range of load power. Set the power factor for near unity at a given load - then change to a lower or higher load - or change the reactive nature of the load - and PF will go into the toilet. At this level of power - you would be better off to just live with low PF, if you don't want the hassle or efficiency hit of the PFC section.

All for now - R.

 

[NickS]

So I know control frequency shift will impact the transformer performance. So what is a reasonable frequency range for a half-bridge? I see one of the parts is moving all the way from 100k to down around 20k(over load range). That seem to be very much non-ideal. How large of a problem is this?

 

[Militoy]

Back again -

Not a problem as long as the transformer is designed for the highest ET product (flux) anticipated - and as long as interference with other system components has been considered. It's more of a control issue than one that affects magnetics design. As I posted earlier - I usually pick a reasonable starting frequency - then lower it as much as possible while still meeting all design goals. Fifteen years ago - designers all seemed to try to push up frequency as much as possible - seeing high frequency as the "holy grail" enabling them to shrink the size of their magnetics and caps. Sometimes - running 500 KHz (or higher) can still make sense - but these cases are unique to particular purposes. In most instances - somewhere between 100 and 200 KHz is going to be the operating frequency where volume, cost, efficiency, EMI performance and flexibility of circuit layout all come together. Theoretically - running the power transformer at lower frequency forces the designer to select a core with more cross section - and to use more turns on the primary to avoid saturation. But there are other issues limiting making the core too small. Before getting too involved with frequency, we need to pick a topology. Both the half-bridge and 2-transistor forward will put us into a similar practical frequency range and power transformer core volume by the time we're done. We're going to be using off-the-shelf ferrite cores in either case. The advantages of the half bridge are unlikely to allow us to use a full size smaller core, in this power range. I'd advise, let's choose our poison - half bridge or forward - then settle on a controller. The next step will be selecting a core shape from the many available types. Once the process is complete, we can always revisit the original decisions - and push the design in any desirable direction.

 

[NickS]

Ok, I am back on track with this project. I am really liking what I see with this pair of controllers L6563 and L6599A for PFC and Half Bridge respectively. I am particularly excited that the two controllers are designed to work with each other to improve performance and reduce risk.(see attached). So unless there are objections I am going to run with this topology.

 

[Militoy]

Ok, I am back on track with this project. I am really liking what I see with this pair of controllers L6563 and L6599A for PFC and Half Bridge respectively. I am particularly excited that the two controllers are designed to work with each other to improve performance and reduce risk.(see attached). So unless there are objections I am going to run with this topology.

No objection - your selections make sense. I haven't worked with this particular combination before - As I mentioned earlier, I'm just winding up the first stages of a HV project using the L6599A - but I'm not really too familiar with the eccentricities of the chip yet. The L6563 is one I don't recollect working with - in any case if I did - I handled the magnetics only. So we can work through the design together - and we'll both pick up some new useful knowledge.

I will look over the data sheet for the L6563 - and pick a range of core shapes / sizes to select from. Or - would you rather take a stab at that part of the process - and we work from there? We should probably start out with the DC-DC part (the output), then do the front end. We can reverse the process, if you prefer.

In any case - I'll throw out a "bone to chew on" first. With this combination of topologies, for the power and boost cores, you should be thinking in terms of ferrite - period. There are other options to select from - MPP or MPP / ferrite combo toroid for the boost (if shoved into a tight space and trying desperately to avoid local EMI radiation) - or amorphous glass (downhole oilwell supply or space application), etc - but generally, we'll be working from the ferrite catalogs.

I would suggest we stick with a little better grade of ferrite to start out with. Select your favorite supplier - but pick one with a good range of materials and shapes. My first inclination is always to use Mag Inc cores - but I use a fair amount of Ferroxcube products, and some Fair-Rite, TDK, Laird, etc for various applications. Take a look at Mag Inc P material - then compare with F, J, etc - and do the same with Ferroxcube's offerings - or whichever supplier you may wish to use. Get a few datasheets in front of you - and we can start the selection process. For core shapes - PQ, DS, Pot core, and similar self-shielding shapes have some advantages - but look over the shapes in the online catalogs - and we can discuss a starting shape as we move forward.

 

[NickS]

...I will look over the data sheet for the L6563 - and pick a range of core shapes / sizes to select from. Or - would you rather take a stab at that part of the process - and we work from there? We should probably start out with the DC-DC part (the output), then do the front end. We can reverse the process, if you prefer. ...

I will try to get back tonight with the pertinent specs from the datasheet. Then we can address the important details to steer the hunt for cores.

 

[Militoy]

Sounds good - I've been travelling again, and haven't had access to civilian internet for the past few days. I should be back in the office on Wednesday. I find it much easier to work on a new design when I have direct access to my "Brain" - which is a collection of catalogs, charts and design notes that I've put together over the years. Our first step will be for you to decide on an exact output voltage and max current. I'll assume for now we're going with the higher input rail voltage, but it won't affect the core selection either way - only the primary turns - so you can go with either 120V or universal input. We'll work on the power magnetics first - then outward from there.

 

[NickS]

Sorry I have a lot on my plate at the moment.
So the PF controller makes use of a boost stage that takes the universal input up to 400V. I believe that that is just for uniformity with varying input levels and I am pretty sure it is adjustable. So from this post lets look at cores for two possible scenarios.
Here are the metrics for the transformer.
Case 1: Vin = 400V
Vout = 28V
Iout 3A
Sw Freq center = 250kHz

Case 2: Vin = 180V
Vout = 28V
Iout 3A
Sw Freq center = 250kHz

I am thinking that the size of the transformer could come down with a lower output voltage since it would mean less secondary side windings. If this will make a significant difference then I can live with a Vout of 20V.

I have spent the most time on Mag-Inc's website so unless they lack the selection we need lets start there for core hunting. Speaking of which here are some of the key points I would appreciate discussion on.

1) What cores materials are reasonable for this design? You mentioned ferrite before this is because it strikes a good balance of cost and size right? any other driving forces?

2) Size: What is the primary core metric we are concerned with in limiting our search within the bounds of a core material? ie It seems that we need to do some preliminary calculations at this point to narrow the field? If so this may be the dividing line of workable cores for case1 vs case 2.

3) What shape is most efficient/reasonable? Perhaps you could give argument for both hand wound and commercially would scenario.

Thanks

 

[Militoy]

>>>I am thinking that the size of the transformer could come down with a lower output voltage since it would mean less secondary side windings. If this will make a significant difference then I can live with a Vout of 20V. <<<

A lower voltage secondary of the same current will take up less winding space, and lower copper losses – but it may or may not be enough to put you into a smaller core size. Figure a savings in copper fill of a little over 28% by dropping from 28V to 20V (20/28 = .714 fill). 60W vs 84W might buy you a core size – depending on shape. A 28V supply will power a wider variety of devices than a 20V one will – so you may want to also consider versatility. Your choice.

>>>I have spent the most time on Mag-Inc's website so unless they lack the selection we need lets start there for core hunting. Speaking of which here are some of the key points I would appreciate discussion on.<<<

Mag Inc works fine with me – and I have a wide variety of their products available in stock to work with.

>>>1) What cores materials are reasonable for this design? You mentioned ferrite before this is because it strikes a good balance of cost and size right? any other driving forces?<<<

From about 50KHz up to maybe 4-500KHz – ferrite is typically the best choice. The main consideration is core loss. The high volume resistivity of ferrite limits losses from eddy currents in the core material – which increase dramatically with frequency. There are also amorphous glass tape-wound cores available that have similar properties at high frequency – but limited availability of shapes and cost considerations make them less attractive in a commercial design.

>>>2) Size: What is the primary core metric we are concerned with in limiting our search within the bounds of a core material? ie It seems that we need to do some preliminary calculations at this point to narrow the field? If so this may be the dividing line of workable cores for case1 vs case 2. <<<

We’ll start with a reasonable estimate of core and copper losses – based on how much power we’re willing to dump in the transformer alone. This is the start of our “power budget” – which we’ll work out for every dissipative component. We can then work out core surface area needed to shed those watts at a given limit of temperature rise. That will give us a first crack at picking a core size to start with. We can then work up a preliminary design – estimate turns based on core cross section and initial design goals; then check our initial figures by calculation – and adjust core size up or down as needed. We’re not building thousands of units to start with – so I would stick with a better high frequency, low-loss material like Mag Inc ‘P’ – then compare to a few different materials like ‘R’, ‘F’ or ‘J’ as a cross-check.

>>>3) What shape is most efficient/reasonable? Perhaps you could give argument for both hand wound and commercially would scenario.<<<

In military work we use a lot of toroidal cores. They are extremely efficient in terms of power / volume ratio; are self-shielding, which helps limit radiated emissions; and with the windings on the outside of the coil / core structure, shed heat very efficiently. The down-side of toroidal construction is the necessity to pass the entire winding wire load completely through the aperture of the core, each time a single turn is added to the winding. This can be a very labor-intensive process – or requires specialized winding machinery. It’s also difficult to gap a toroidal core in production, for a design like this one – where a gap is essential. A good compromise to this problem is to turn the toroidal core / coil system inside out – by reversing the positions of the core and coil. This is essentially what a pot core is. Using a pot core or any of the related shapes – PQ core, etc – allows winding on a bobbin by hand or using simple machinery; and affords some of the self-shielding attribute of the toroid. When cost is the primary consideration, simpler shapes such as the ‘E’ core can be used. When height is at a premium, low-profile cores like the EPC core might be a better choice. Unless there are compelling reasons to use an alternate shape, I would start this one on a pot core, PQ, or related shape. Lots of size choices in this style of core.

 

[NickS]

What I gather from your last response is that core selection can be iterative but the sets thus far are
1) select the output power requirements [ Vin = 400V, Vout = 28V, Iout 3A, F = 250kHz]

2) Ballpark core material based on trade off in losses size cost [ferrite based on frequency and availability]

3) Determine allowable copper/core losses ?? how is this point set? qualification requirements or heat removal?

4) Select core material [P, high frequency, low loss].

5)...
6)...

Please edit/amend this list as you see fit.

And next lets address the power budget in some quantifiable detail. To certify the power supply to various "green" standards I am sure this needs to stay low. But if that is not a concern is there space of cost to be saved in tolerating more core loss? Lets discuss tradeoffs for losses.

 

[Militoy]

1, 2 , 4 are good. I'll go over 3 with you as soon as I can sit down at the computer for a few without interruption (just leaving work now - and will likely be up late tonight massaging a machine). It's all about "blackbody emissivity"; radiating surface area and available convection.

I don't pay much attention to "green" standards - but in electronics, heat is always the enemy. Unless there is a very compelling reason - like absolutely limited real estate or weight budget - I would rarely consider trading for higher loss than what I consider nominal. I normally try to limit temp rise in a ferrite assembly to 50C - but would rather see 25 or 30, if I can get away with it. 50 degrees C (25 rise over ambient) is enough to be uncomfortable to touch. A lot of commercial ICs are limited to 70C operation. Why heat up the world around our core, if we can avoid it?

 

[NickS]

I don't want to dump any power in the core either. I was a little confused by your statement "..how much we are willing to dump in the core.." I understand that some must lost. And since you brought it up I assumed you had a tradeoff in mind thus addressed it. It doesn't seem like a driving metric but rather an acceptable evil when all the others come together.

So looking at mag-inc.com the first classification is
Powder Cores(DC apps), Ferrite Cores(high freq apps), Tapewound Cores(specialized apps)
So we pick Ferrite

Next reducer is shape (1)Torroid(common mode chokes, EMI filters, broadband transformers, pulse transformers and current transformers),
(2)Shapes(pulse transformers, high frequency transformers, and noise filters),
(3)POT core(power transformers, power inductors, converter and inverter transformers, switched-mode power supplies and filter inductors). So in agreement with your advice (3)POT core is logical.

Next are the letters C(900u), L(900u), E(2ku), V(2.3ku), R(2.3ku), P(2.5ku), F(3ku), J(5ku), W(10ku). I am not finding a good descriptor here but I see mu varries by the letter chosen and you suggested "P" which looks fairly middle road. Is this where we meet the cross roads with core loss vs ?

 

[Militoy]

I don't want to dump any power in the core either. I was a little confused by your statement "..how much we are willing to dump in the core.." I understand that some must lost. And since you brought it up I assumed you had a tradeoff in mind thus addressed it. It doesn't seem like a driving metric but rather an acceptable evil when all the others come together.

So looking at mag-inc.com the first classification is
Powder Cores(DC apps), Ferrite Cores(high freq apps), Tapewound Cores(specialized apps)
So we pick Ferrite

Next reducer is shape (1)Torroid(common mode chokes, EMI filters, broadband transformers, pulse transformers and current transformers),
(2)Shapes(pulse transformers, high frequency transformers, and noise filters),
(3)POT core(power transformers, power inductors, converter and inverter transformers, switched-mode power supplies and filter inductors). So in agreement with your advice (3)POT core is logical.

Next are the letters C(900u), L(900u), E(2ku), V(2.3ku), R(2.3ku), P(2.5ku), F(3ku), J(5ku), W(10ku). I am not finding a good descriptor here but I see mu varries by the letter chosen and you suggested "P" which looks fairly middle road. Is this where we meet the cross roads with core loss vs ?

Good start above. First core power dump - you picked up on my point correctly, right from the start. Dissipation from the core is BOTH a driving metric and a necessary evil. We’re just not talking about core losses here, though. The main thing that limits a core's ability to operate at a given power is its ability to get rid of heat. Say we wanted to build a 10,000W transformer in a 1 x 1 x 1 inch cube. No problem with voltage and frequency – just use smaller wire, so we can fit enough turns into the core winding area. We keep the flux density in the core to a reasonable level, by using enough turns. So core loss isn’t an issue – we set it where we want it. The obvious problem with this approach is copper losses – DC resistance loss (I^2R loss) from winding current; skin effect AC losses; and usually biggest of all, proximity losses. Remember – all the copper loss has to get out through the core – unless the winding is on the outside, like on a toroid. So – limiting the copper loss is usually the driving factor in core selection. We need to balance all losses – and get rid of them all over the core surface area.

This is why we start with temperature rise. There is a fairly elegant relationship between core surface area and temperature rise. The model was worked out in the 1920s, using blackbody emissivity as a starting point – and assuming 55% radiation cooling – and 45% free-air convection. Conductive cooling through the leads is ignored. This gives us a conservative starting point that works out OK, until we’re working with little PC or surface-mount transformers. The routine is to calculate surface area of the core – ignoring the bottom surface – we’re measuring the top and sides only. We’ll then figure on approximately .008 Watts / square inch / degree C temperature rise. So, a core dissipating 4W across 10 square inches would have about a 50 degree temperature rise in free air. Smaller cores, with larger relative surface area compared to volume, will push this relationship off the edge. But it provides a good first-estimation point to start with – and a means to compare cores, if we start with a known design.

P material has a good combination of attributes. I wouldn’t really call it “middle of the road” – so much as “optimum” for many applications. Look at the core loss charts at medium to high flux levels at 100 – 250 KHz. Several of the power materials that have lower core losses can’t be pushed as hard as P – or have other issues.

See if you can locate a copy of Mag Inc’s guide chart of cores / frequencies / power. Look it over – or, if you can’t find one online – I’ll scan one and put it up somewhere where you can access it.

All for now – R.

 

[NickS]

I am not finding the App. note with that plot