Small transformers and LTSpice simulation

I have a design that could use three or four winding toroid transformer.

In the LTSpice notes they suggest keeping the transformer coupling at
1.0 or -1.0 otherwise there's lots of high frequency noise generated,
I'm not sure if that's real world, or the simulation going silly?

What I'm wondering is just how much coupling should I expect to see in
a real world transformer of around five to ten turns on a half or
three-quarter inch diameter toroid? Like the ones you see on PC mother-
boards. Or the larger mag-amp ones from PC power supplies that come
with two or three windings.

I have lots of toroids recovered from power supplies, no idea of what
inductance they have until I power up some circuit and try to match
LTSpice inductance with observed waveforms I'm guessing 33uH at
the moment for 40 or 50kHz operation.

I do notice that over-voltages start at close to ideal coupling, for
example 0.97 can give a nasty over-voltage spike on the leading edge
in LTSpice. I don't know how much of that to expect in a real circuit,
any guidance here?

What I plan to do is drive a transformer with a current limited latch
circuit, +ve edge turns on a 'hc74 flip flop, current sense through an
npn will pull 'hc74 reset line down. Should be safe enough to watch
the waveforms.

Frequency of interest is 20 to 100kHz, current up to 2A through N-chan
MOSFET driving the transformer, small snubber on primary as suggested
by LTSpice, secondaries are standard flyback, pair of schottky diodes
and caps.

I might even use a 555, as it has a -ve reset line? Save me building
a separate oscillator, add the voltage cutoff and it's done, cheap'n'nasty.

Can one make a 50KHz oscillator from half an 'HC74? and an R + RC? Is there
a odd numbered ring of inverters hiding in there?

Thanks,
Grant.

 

LTSpice shows such nice clean results with unity! I wanted to know how
far away to make the value.

0.99 or 0.98 looked more like real life for the thing I was investigating.

If was after a range to expect, I'll try it soon to see how close LTSpice
is to reality. I'm bouncing between two different topologies, perhaps I
should simply flip a coin to decide which one first

Grant.

 

The interesting thing about putting up ideas here is the brain bending
_simple_ questions some raise Because I didn't think of it? Plus
I'm trying to get an idea what to expect before wiring something up to
24V 100AH battery, things go bang, or hiss and weld themselves together
while the fuses are slowly deciding to open... I welded an Anderson
connector the other day, shorted out the battery, silly thing to do.

At least I didn't have dancing cables hissing and jumping about, inside.

Grant.

 

Actually I modeled the output latch to be like the comparator reset latch
that some of the power PIC chips have, with external level shifting to 10V
gate drive. Not used a current transformer for feedback yet, haven't looked
to see how big the signal is. I remember when you put that circuit up many
months ago. I recall rightly the PIC comparator reference can be varied
so the switching basically is happening in hardware, supervised by the
software. A timer setting the latch and output, the comparator terminating
power cycle?

What happens if your feedback coil wired backwards?

One reason I modeled the power circuit was to make sure it settled safe if
the controller stalled, that meant AC coupled gate drive so that open loop
step would not cause over-current. Very nice to simulate that little bit,
the rest of the circuit left to real world design, I wouldn't try to do the
lot in a sim. So now I have a failsafe hardware driver waiting for my to
catch up with PIC chip and an input sensor opamp.

Possibly we agree on this

I've yet to do PIC serial comms, bought some serial to USB bridge modules to
try for that. Keep getting distracted with other stuff though. No programming
PICs for months now.

Grant.

 

Yes, that's for the power supply driver type, one I'll be using soon

Bang! ??

Nervously watching a 30A converter from 100AH batteries?? Nope! Too nervous

For when the PIC is dead, waiting for watchdog reset -- when 'stuff happens'.

Someone commented recently about not using PIC below 16bit 24F* series, I'm
thinking about it, but I bought a lot of 16F and some 12F baby stuff to play
with, none of the in-betweeners like the high performance PICs you're using.

Yet...

Got some of the cheap TI 430 series intro kits too, not powered one up yet,
kit comes with a little USB interface, loads of software for download -- but
they're not getting a mention from anyone here? 'Cos they bad or just new?

 

The 16-bit MSP430s are squeezed from the bottom by 8-bit AVRs and from
the top by the multitude of 32-bit ARM7TDMI and Cortex M3 chips. There's
still a niche for the MSP430 series but it's getting narrower every day.

 

Usually I get 0.96 to 0.98.

Pretty easy to measure: Short a secondary and read the inductance again.
It isn't terribly accurate for a multi-winding transformer but give you
a ballpark number.

0.97 is quite realistic although I had 0.995 recently, with an EI-core
no less. You've got to deal with those spikes, they will happen in the
real world. There will also be ringout. I always try to tackle this
without snubbers because those kill efficiency. Better to dump that
energy back into a rail and get the covered Energy Star sticker

Why not use a real PWM chip for starters? You can always back down from
there to reduce cost once everything runs to your satisfaction.

 

Phssss ... tck ... bzzzt ... *KABLAM*

Seriously, I'd try this with a real PWM chip first. 30A is nothing to
sneeze at, if something goes wrong you could have molten solder
splattering about.

Then once it works you can always migrate to a uC or discrete solution.
Real PWM chips have some nice feature in there, such as leading edge
blanking for the current sense. A uC (normally) can't do that because
the granularity for such features is finer than their clock allows.
Although you could probably do it with a Cypress PSoC. The other issue
with uC is that they don't have enough gusto to drive a FET, so no real
financial incentive is left there.

Well, they run on 3.3V and the ports are kinda wimpy for driving a big
FET. Also not much direct HW-interaction with the timers is possible and
that can make a switcher risky. One little hang-up in the code ... *POOF*

 

Thanks for that, I don't know the AVR, I'm not particularly impressed with
the PIC from a programming PoV, in that stuff I found easy to do a couple
decades ago with 68HC05 family is near impossible (plus I'm terribly rusty).

For example, a simple LCD module display cache, only rewrite letters that
changed after a page update (in multi-tasking situation where 'app' writes
to virtual screen in memory, cannot directly access hardware to the display
as the hardware data bus is a shared resource).

Only had the PIC debugging support to go by, but I couldn't spot the issue,
gave up for a while. Well, the break is now ten months, embarrassing. I
got distracted by too many other things, this prototype been waiting for
software that long :/

Photo here: http://grrr.id.au/image/adc-card.jpg

PIC (40 pin) to run high resolution ADC (15bit) front end plus LCD module,
room on the foreground board for CPU with serial link. Later. Other board
will become a controller for a measurement project I been working on for
years

So it seemed I try to do too much with the PIC on the human interface side.

It's probably just right for power supply stuff I work with. Got the needed
latch onboard and so on.

Didn't want to go to ARM for my higher level stuff, seems too much? Maybe
it's not. But what of the entrance fee?

Depends on support tools, PIC wins there because entrance fee was like $50
for a PICkit II, plus the 28pin + 44pin cards have CPU with onboard debug
(limited ICE) support that makes it easy to view memory after a breakpoint.

Grant.

 

Nah, each 12V battery has it's own 35A fuse to guard against the most stupid
things -- like what I did last week, shorting out the 24V -- because I present
the protected batteries as two by 12V, then a connector puts them in series
or parallel, well I assumed the two centre contacts would be the natural
centre tap for series batteries.

Only as I plugged the old series whatsit into a new thingy I'd just made,
heard and watched that overloaded DC hiss welding, did I realise the old
whatsit joined the outer contacts to put the batteries in series. Oops!

Blew each battery's 35A fuse and weld up the mating Anderson connectors
where the arc happened. I'm so glad I decided on a 'proper' last resort
safeguard. Replaced pair of fuses, and that pair of welded Anderson
connectors

I don't like the idea of burning dancing wiring inside where I live!

Not flying molten solder... And the smell!

Ah, but some PIC chips have that latch in hardware, so the software is
setting the parameters, but not directly involved in the switch off
decision that is handed off to an onboard comparator and latch. So
it should be safe. Though one needs external RC to stop that leading
edge pulse.

I modeled that part of the power circuit as a 'HC74 latch with standard
MOSFET source current resistor feeding an npn, collector pulls 'HC74 reset
low to finish the output pulse on max current.

So the design aspect of interest for the simulation was the 5V to 12V gate
drive level shift with AC coupling, since the power circuit is half bridge,
active low pulse for the P-channel, active high pulse for the N-channel.

For that aspect, the simulation was useful, informing start capacitor and
resistor values around the level shift transistor vs output inductance.

You could be right there, in that I have to add the usual two transistor
plus diode level shifting drive. But then I'm not doing a commercial
high quantity design. So I'm not terribly fussed by extra components
at this stage It is interesting to consider commercial design reality
at times.

Hell, if I come up with a design where we had volumes to contract out
the assembly and so on, there'd be time to save every little bit. I work
in a niche area where if I did solve a common problem, the solution would
be copied in a flash -- but only if the commercial volume was there.
It isn't. Not in .au anyway. Not much of an industry left here, just
after-market add-ons to make stuff a bit better suit intended use.

Okay, the intro kit cost $4.30 includes a short USB cable almost worth it

They sending them out for free too, 'limited time only'... For months now.

But product design shouldn't go *POOF* if the CPU hangs, that's why there's
watchdogs and stuff -- perhaps I show my age here. Or that's why stuff
still has builtin fuses -- in the software doesn't lockup something might
fail anyway?

Safe uC design? I used to make stuff that had to work 24/7. That meant
watchdogs and undervoltage reset -- though the last only accepted by the
boss only after field failures were solved by the extra part cost. Grrr!

Grant.

 

Almost a car headlight, I have a heap of cheap 12V 50W halogen globes for
dummy load, as well as some big resistors, 5R 280W and some 10R 72W. But
my accidents here with 24V are usually bits of loose wire in the wrong place

Sure, that's what I meant by AC coupled drive, not fussed by the start state
of the port pins, or care if they're stuck hi or low when I stop the CPU to
examine memory content, which I'm used to doing and allowing for in hardware
plus software design. Old habits?

I agree with you, except I'm not good with any PICs yet, not done enough
with them, apart from select a few, buy a few and buy the cheap entrance fee
of tools (PICkit II plus the demo boards with debug support CPUs) to play
with.

I have my own style of multi-tasking OS mostly written, it's just a timebase
framework and interrupt round robin call list cooperative type, no magic,
no context save & restore 'cos there's no support for that. Sort of stuff
I did couple decades ago on 'HC05 CPUs. Bent and fitted into PIC reality.

Grant.

 

I don't have an inductance meter, seems it's time to make or buy one
since I want to play with inductors and transformers? Another PIC
project -- I think Jan worked on one? I already bought some 1%
polystyrene reference capacitors to make one. Yet another distraction.

Right. What about the fuzz LTSpice adds to the primary waveform? It
doesn't look real to me? I certainly believe the spikes, I've seen
that on hardware, but aggressively dumping it back to the rail killed
power transfer to the secondaries in my simulation. But that's where
I need to play with hardware as well, to see what's real and imagined

Balance how much energy to capture on the primary side, without snuffing
the life out of the system.

Because my favourite chip at the moment (NCP3063) is not a proper PWM, it
can be talked into PWM by injecting current onto the timing cap node, but
I didn't have much success with that yet in hardware*, and cannot find a
model for simulating it, to help me get into the ballpark.

* Drawback is one has to add a reference to complete external control
loop, so there's little incentive to go that way, too many components.

Got some of the common 2843 (?) series chips, could try them. Otherwise
I'm open to suggestions!

Grant.

 

Nah, come on, you Australians are supposed to be the masters in
improvising

(Saw Crocodile Dundee in L.A. yesterday)

Take a function generator and a scope. If no gen available the sound
card of your PC will do if it spits out a decent sine wave. Place a
resistor is series with the winding and feed at the top. Now measure the
amplitude at the top (where the gen connects) and then again across the
coil. This is like a voltage divider. The ratio will tell you the
impedance Z of the coil at, say, 15kHz. Now

L = Z / (2 * Pi * 15kHz)

In a pinch you can also use the sound card to measure the levels.

Not sure what fuzz you mean, maybe post the sim? You can copy it into a
post. As long as you don't use any fancy part models others can then run it.

Some fuzz is actually ok, like the minor ring-out that leakage
inductance causes during off times.

Well, essentially it's the flyback kind of energy in a non-flyback
converters that is often not healthy, needs to be dumped somewhere so
parts don't suffer. For example, if the primary FET can take 100V but
the spikes want to go to 150V that needs to be taken care of.

What architecture were you looking at? Why not pick one of the LTC parts
and then later after the simulation runs try out something else? For
step-up and SEPIC the LT3757 is pretty good.

 

Watchdogs don't do all that much in a switcher. By the time that comes
on a few clock cycles might have gone by. It can be a matter of
microseconds for a switcher to go from fully ok to kablouie. On a
current-controlled switcher things have to react within a fraction of
one cycle. Once the inductor goes too deep into core saturation all hell
breaks loose. Fireworks and all.

The UVLO is a smart thing you added there. Switchers should never run
without, else it'll kill the FET some day. All it takes is a slow brown-out.

 

Sorry, I never made uC controlled switchers before, just products that had
to run 24/7. Switching supplies I worked on decades ago were mains
commutated SCR monsters (up to 500A) charging industrial batteries -- so
one had a filter inductor about half the size of the mains transformer with
an air gap adjusted for no saturation pip at full current, high input and
lowest output. Battery model? Short circuit with a voltage offset!

I agree, one must switch off on current or not allow saturation by inductor
and cap selection -- what I did with the simulator was dangerous in the sense
I built what appears to be a safe power circuit without a current sense.

So I rewired a 100mR into the prototype so I could at least loot at the
current in operation -- but this particular whatsit is only a couple amps
and polyswitch or fuse protected.

Yeah, not supposed to happen, I had a visitor from interstate yesterday who
told of a brown-out lasting two hours! I was surprised, more than a few
minutes here and the power comes back or goes out completely.

Grant.

 

Maybe he is on one of those long SWER lines. Australian line power is
the ultimate test for any power supply. On the last one I re-designed
for your neck of the woods I made sure that nothing would ever saturate
up to 275V/50Hz. In the testing I gave it bursts of 300V for 15min. Then
I tested for the "eternal brown-out" where the voltage would hover
around where things draw max current and where the UVLO just hasn't come
on yet.

The other test would be islands with generators, where they sometimes
reach in and goose a recalcitrant engine. Vroom ... vrooooom ... *POOF*

 

I haven't seen it, apart from the 'that's not a knife' clip that was shown
all over the place.

No it's crap during the on time. . . I found it, it's actually the ringing
on the secondary while a diode there is reverse biased during the MOSFET
on time (flyback). I was simplifying the circuit to have something to post,
killed it, started over and found the ringing.

I guess with perfect coupling the ringing was absorbed by the primary's
snubber, and that's why it only showed when I set coupling less than unity.

Good way to learn.

I changed snubber to diode feeding an RC to +ve rail to recover some of that
energy, much better, and I have just as much control over the peak MOSFET
drain voltage as a straight snubber gave. This is where the sim 'what if'
is helpful for me.

Flyback converter. I was trying to ignore checking out all those LT parts,
I'll look at the LT3757, see what it's like.

I bought a heap of the 3063s for running LEDs, I don't have to use them
everywhere. I bought some PWM chips too, but on price, not for ease of
use with LTSpice -- this was a fair while back, before I started using
LTSpice.

Grant.

 

Today it showed me enough to learn that the particular approach I was looking
at is not worth building I've already built part of a more complex thing
that I'll continue with, that's a half bridge into the load, more parts, but
more likely to work as expected, plus there's a PIC chip in there to help do
some power accounting for the up-market version.

But I had to see how good the cheapie idea was for the job. Rather be able
to point out its shortcomings if someone asks why didn't I use that particular
topology in the future.

Grant.

 

I sat through it twice just to see the bint.

John

 

Hi, Grant -

Take a look at the LTSpice non-linear transformer in the educational
folder. Probably not what you're looking for, but it may be helpful in
some way.

John