Coil losses

IIRC the Hobbs converter is a bit different. One switcher chip that drives two consecutive FET-inductor-diode-cap groups. That should be quite efficient.

I usually got north of 85% out of flybacks even without recuperating winding. But they can create more EMI concerns than boost converters.

Good strategy.

It seems to work for John Larkin, but it still strikes me as tieing one hand behind your back. If John Larkin ever learned to design his own transformers, and worked out how to get small quantities made for him, he'd probably make even more money.

Consultants have to take the silly ideas of their customers seriously, but it isn't wise to let the customer's delusion block the route to a good solution.

If you had ever run a company or a large part thereof with P&L responsibility you would have swallowed that statement.

I was quite close enough to profit and loss decisions to be well aware of t he way you could lose money - and credibility - by taking a customer's sill y ideas too seriously. Finding a way to ease them into a more sensible poin t of view could be very tricky indeed, and we didn't bother when it didn't matter, but the last thing you want is for customers to find out that there 's a better way of solving their problems than the one they paid for, no ma tter how insistent they were on a particular solution before they became co nscious of a better one.

One of our products is licensed to a big company to manufacture, and the custom dc/dc transformer can't be made any more. It's a pot core on a surface-mount 10-pin header, and the header is no longer made. BigCo uses contract manufacturers, and the CM knew that the header was going EOL, and could have bought a thousand cheap, and didn't. A CM has no incentive to stock EOL parts. So now I have to redesign the dang thing, starting with the header as a tiny PC board maybe. What a pain.

We avoid custom magnetics and contract manufacturers. The guy who designed the power supply was a former employee of Signal Transformer and liked to design magnetics. I should have stopped him.

There are tons of SIP and surface-mount DC/DC converters around now, cheap and multi-sourced. And many thousands of standard inductors and transformers. Not much point in designing expensive custom magnetics unless the application is really exotic.

I wonder when Sloman will fire up his famous custom-magnetics oscillator. It's good to check up on that every 5 years or so.

If you mean the bobbin that went EOL maybe you could talk to Renco.

There are only few things where full custom makes sense: Beer brewing, pizza, bean dip, and such.

I do not like DC/DC blocks. Many have EMI issues and they can go EOL out of the blue. I prefer a fairly popular switcher IC and then catalog discretes for everything. Sometimes I even go sans switcher IC, especially when the plan is to keep something in production for more than two decades.

:-)

Guacamole. Pizza for sure.

Enthusiasm for winding wire on cores is some sort of disease or addiction.

I have considered making pulse transformers from spiral traces inside multilayer boards, with kilovolts of isolation. But one would still have to hassle with the core.

Like this:

A totally unexpected problem?

Components are taken out of production all the time - it's not a problem th at is exclusive to magnetic parts.

The rest of your crew know about integrated circuits (which also get taken out of production). Maybe you should stop them designing in integrated circ uits, and concentrate on discrete component design (not that discrete compo nents aren't taken out of production either).

There may be quite a few 1:1 transformers. Getting the turns ratio you ough t to be using is trickier, and getting the exact inductance and resonant fr equency is trickier yet.

There are a lot of design variables in a transformer, and the chance of fin ding exactly the one you ought to be using is rather lower than it is with inductors, resistors and capacitors.

You aren't paying attention. The custom-made transformer/inductor was going to inject harmonics about 105 dB below the fundamental, so it got dumped a bout three years ago. I've installed a Linux partition on my desk-top compu ter (SuSE 13.2) and installed KiCad - it's even got a page of the schematic of the latest incarnation, but I've not done anything on it for some six m onths now.

No prospect of a customer, so no motivation.

And transformers.

Rather like designing things right, rather than tinkering with them until they more or less work.

There are plenty of planar cores around.

A few years ago somebody - I think it was Spehro Pefhany, but that might be because the link was to a Canadian firm - posted a link to a printed circuit house that specialised in printed transformer windings.

Things like 10 ounce copper on really thin substrates, so you can get close to the sort of fill-factors that you can get with wire windings.

At an exhibition once in the Netherlands I talked to Dutch firm that did that kind of work, and got the same kind of story.

It didn't look as if was going to make sense to make a whole board that included printed windings, so John's links do suggest that his thinking didn't get as far as talking to anybody who might have made what he had in mind.

My first action in cases where I need a custom transformer is a call to a custom transformer maker. I have once re-wound a huge speaker to make it work with a tube power amp I just built and could not afford an audio transformer for. Once, as a teenager. Never again because I think I let off all the choice words I knew while doing that and my fingers hurt for a long time.

I think Tektronix uses that method in their diff input scopes. If you have one you could take a look.

We have one of the TPS-series isolated-input scopes. It's fabulous for probing power stuff; you can hook the probe ground clip to anything. I think the signals are FM'd to get across the isolation transformer.

ut

y_exploded.png

_board.jpg

mikeselectricstuff milling trough the layers of a Sun cpu module

-Lasse

Or PWM-ed. That's how I usually do it.

It is 100 MHz bandwidth, which is hard to PWM.

Depends on what granularity you need. 8-bit precision would still be feasible at that BW now that we have cheap comparators that allow to discern tens of picoseconds. FM could be cheaper though and won't need a servo path.

For many people the magnetics up to the GHZ range are daunting. But it's not so bad because the core has checked itself out of the game much before the spectra reaches up there. I learned the ropes more or less from really old tech: TV splitters. In Europe they split a coax input into VHF and UHF symmetrical at 240ohms and the UHF section was all microstrip. On phenolic so they could sell at rock bottom prices. I still have one somewhere. When people scoffed at such ideas during design reviews and hinted that it may be unproven voodoo I passed it around and told them that it was from the 60's.

One of these days I'll lay out a test board with a variety of embedded transformers. Some applications could use no-core structures, for fast pulses and such, but some, like power transmission or brutal gate drivers, would need some sort of core.

The only real reason to do this would be for high voltage isolation. There are tons of cheap surface-mount transformers around otherwise.

One of my cow-orkers tried it for a 5KV power supply. He scrapped the idea after the first (miserable) failure. I don't remember what the problem was (I'll try to remember to ask tomorrow) but it didn't go well at all. I think he's using something from the lighting industry now.

For high-current, high-power boost, a two-phase controller, like the LT3782, works well. Cuts peak input current in half, halves and averages the load current delivery. E.g. a 12V to 42V, designed for high-environmental-temp operation,

Interesting thought. You'd want a 3nsec period, and if you used a couple MC100EP195 parts in alternation

you could get roughly 10psec increments - roughly 300 steps, or about 8-bit resolution. The actual delays are temperature dependent, so it would need frequent self-calibration.

The 196 has an additional FTUNE pin for vernier adjustment, which wouldn't be much use in this application, and NB6L295 has a serial data path, which would make it even less useful.