Strange idea..

at about 200 mA.

Give this a try:

I've also been playing with this recently. Pretty much ideal behavior = in just six transistors (no feedback winding). Who even needs a UC3842?

(any toroid),

That can be a tricky approach. If you do a stupid blocking oscillator = thing,

Put half and half together, and you get the equally stupid chopper = thing, which looks like a multivibrator, but maybe isn't (since timing = comes from the tank, you don't need RCs). This time, at least, it = saturates in both directions, so you can drive low Q loads, and maybe = biasing is a little easier.

The amazing thing is, the stupid chopper is only one inductor away from = a surprisingly smart Royer oscillator. With 180 degree conduction = angle, a Royer is very efficient, has little switching loss, and uses = most of the VA capacity of your transistors. The downside is, the tank = looks like a capacitor against the series inductor, so you can get = overshoot, necessitating excess voltage ratings on your transistors, and = you need a big fat inductor, presumably high enough inductance that = current varies less than maybe 20% during a cycle (i.e., a half cycle, = because you get sine wave cusps at the CT).

the

Because, coupling stinks. And if you use a resonant tank to compensate = for stray L and C, then you don't have regulation, you get an impedance = match system. So the whole thing gets a bit uglier.

On the plus side, if you're only ever driving the same LED, you only = need to match for one load.

to 900MHz,

AFAIK, LEDs look like capacitors in the MHz, and probably make excellent = bypass capacitors in the 100s. And in turn, in the 1000s, they probably = look like hunks of wire, i.e., inductors. So you'll need SMT diodes if = you want to drive them with AC, if they even go that fast.

LEDs are junction diodes, so they'll be a lot slower than schottkies, = but not nearly as slow as silicon, since they're made of the same things = as fancy RF parts, i.e., GaAs, InP, etc. As far as I know, they look = ideal up to a few MHz, so they're at least that much better than Si = junction diodes.

I don't think class C is a good idea, because you can't cover the range = of input and output voltages and currents while maintaining high = efficiency. Class C is what, 80% tops? At this voltage, a proper = synchronous MOSFET switching circuit will top 90%. What you lose in = physical size you make up for in power savings -- you'll probably spend = as much circuit area on heatsinking and resonators as you will on a = controller.

You could certainly look at using the same transistors for switching at = maybe 1/10th the frequency, or less, but good luck finding a controller = that will do it -- you're pretty much stuck rolling your own (not a bad = idea... if you can sell 100k of them) or doing it all discrete (=3D way = bigger than either).

Tim

--=20 Deep Friar: a very philosophical monk. Website:

capacitors in the 100s. And in turn, in the 1000s, they probably look like hunks of wire, i.e., inductors. So you'll need SMT diodes if you want to drive them with AC, if they even go that fast.

nearly as slow as silicon, since they're made of the same things as fancy RF parts, i.e., GaAs, InP, etc. As far as I know, they look ideal up to a few MHz, so they're at least that much better than Si junction diodes.

I've measured various small LEDs that have capacitances from about 5 to around 100 pF. There doesn't seem to be a predictor for that. LEDs also show step-recovery-diode like charge storage, so are probably not good RF rectifiers.

Skyworks makes cute little (as in, you-can-barely-see-them-SC79) schottkies that run around 0.25 pF, and they are *very* fast. Probably not very good at 200 mA.

Some early "switching" power supplies were LC oscillators followed by rectifiers, toobs yet.

John

On a sunny day (Sun, 11 Jul 2010 14:35:32 -0500) it happened "Tim Williams" wrote in :

Thank you Tim, it is a LED controller, but the 3.3V is for the 3 micros (PICs) and the Ethernet chip :-)

I think in the long ago past we had a discussion about the meaning of 'blocking oscillator'. Although the diagram of mine looks the same, and CAN be used as a blocking oscillaor, it is not... it is a simple class C sine wave oscillaor.

------------------- +12.5 | | Scottky 6k8 |------ ------- a k ----> +3.3 |___ ||( | ( | | | ||( 10t === ( === | 1t ) ||( | 150 nF ( --- | | | ----- --------------| | | c | --- - b /// === e BC547 | 1u | /// ///

The waveform at the collector looks like a pure sinewave, except the bottom is flat, where the transistor switches on.

But I do not need regulation, as the 12.5 V comes from an external switchmode. I would need some regulation if I connected it to a full 12V battery,

In the long ago past I cheated with this system by adding a series regulator on the input, it dropped only a little bit of voltage so got hardly warm :-)

blocking

That's a blocking oscillator, operated in a non-blocking mode. Assuming = "blocking" refers to "it's not always oscillating".

With big C on the load, when the "flyback" is done discharging through = the load, voltage goes down around zero (or +V at the collector), but = because of the capacitor, it undershoots, which forward biases the = transistor again. Resonant or quasi-resonant circuits are hard to make = blocking; if you underbias them, then you'll get squegging instead. The = trick then is to make C small enough so the RC time constant roughly = matches the LC time constant, so the squegging rate is close to the = cycle rate and it runs as a throttled sinewave oscillator.

I made a blocking oscillator for a high voltage supply,

:-)

LDOs can be very handy for cleaning up ridiculously noisy SMPS.

Tim

--=20 Deep Friar: a very philosophical monk. Website:

mode, with slightly clipped peaks due to the diodes.

I would be interested to know what your thought process is when you design a blocking oscillator such as the one in the above supply. Do you use trial-and-error, computer simulation, or a particular set of arcane equations?

I have never had much luck finding a "design guide" for blocking oscillators. The one thorough analysis of such a circuit I've seen was in a journal about a modification to the circuit to generate a "broad band chaos" by adding a resistor and capacitor in the emitter leg. Their preliminary analysis consisted of setting up the system of nonlinear differential equations that described the circuit and doing a simultaneous numerical solution. I don't know how much practical use doing that would be for designing a power supply, though...:)

Correction: I believe the capacitor may have been from collector to ground, rather than the way I described it above. I'll have to find the paper and take another look.

On a sunny day (Mon, 12 Jul 2010 02:39:23 -0500) it happened "Tim Williams" wrote in :

But I cannot follow you there, sure it works, but you lose power in that snubber? I did the same for a 300 MHz transistor scope, had some more turns, but used the above tuned circuit. But I went up from 12V DC to a couple of kV IIRC. No power losses in snubbers, no hot components! Also have a look at the old TV CRT H output stages, boost diode configurations. Anytime I see a 'snubber' my alarm goes off :-) Here a nice very old HV supply with build in rectifier: ftp://panteltje.com/pub/HeNe_laser_hv_supply_img_2070.jpg side view: ftp://panteltje.com/pub/HeNe_laser_HV_supply_sideview_img_2069.jpg

This thing produces great sparks, it is a 555 timer switching a TIP140 power transistor. The grey thing holds the HV transformer and HV diode. Runs on 6 or 12 V (do not remember), bought it in the eighties for a helium neon laser I had. Lost the diagram... But you will note the absence of any 'snubbers', in fact nothing gets hot. Snubbers indicates low efficiency :-)

I did that once in a design for the army, they did acceptance testing and could not measure the ripple... LOL

These waveforms may be of interest to you, I connected it to 6V and 12 V, note the parabolic waveform. ftp://panteltje.com/pub/HeNe_laser_supply_collector_voltage_img_2079.jpg

Base drive to the TIP140 from the 555 timer: ftp://panteltje.com/pub/HeNe_laser_supply_base_drive_img_2080.jpg

Output good for a few kV, and short ciruit proof: ftp://panteltje.com/pub/HeNe_laser_supply_arc_img_2078.jpg

The parabolic waveform is what you want, I soldered out the grey object, it turns out to be just a potted voltage multiplier, no HV transformer inside (measures infinity), and when it is out then it shows that same collector waveform. The small transformer is a the step up, it has a many turns secondary that measures about 500 Ohms DC.

The first two. As for the third, only crude approximations, the ones = you'd use in any switching circuit, nothing specific to this.

I learned by trial-and-error, so I have an intuitive grasp of things; I = can pick a few values, calculate a few other, and get close on the first = try. Usually bias and timing capacitance are the problems (since it's = easy to get everything else right).

The critical part of finding the bypass capacitor Cbb

You can control the power output in Discontinuous Current Mode (DCM) = (this circuit) by varying bias; the transistor stays on for constant = time (i.e., until inductor current reaches the peak value, then the = transistor turns off), and in the process, Cbb (C3) gets discharged, so = you just charge it back up at a rate determined by power consumption. = It's an electronic hit-n'-miss engine.

The other way is to modulate the turn-off current, which is how you do = BCM (Boundary CM, the next cycle starts when inductor current goes to = zero):

In all of these circuits, the next cycle starts when the switching = transistor is forward biased. In Cbb-timed circuits (the DCM example), = it's normally biased off, then charges up, then switches on (assisted by = positive feedback). In continuous circuits (BCM, resonant and = quasi-resonant), the undershoot caused by the LC resonance starts the = next cycle.

Of note, core type is important. Ferrite has low loss, so it tends to = resonate very easily. This is good for resonant/BCM circuits. Powered = irons are very resistive and well damped. Fast_DCDC uses a powdered = iron core, and behaves as an ideal DCM circuit.

If you put a ferrite core and lots of capacitance in a DCM circuit, = you'll end up with squegging, because it keeps cycling until Cbb has a = voltage lower than the undershoot peak. This is usually a few volts = negative.

Note there isn't much seperating DCM from BCM, and over a wide range of = line, bias or load conditions, you may get squegging from an otherwise = well-behaved circuit.

was=20

a=20

Hah. It's definitely a complex system and quite capable of generating = chaos. I don't remember what conditions I had, but I've definitely had = the transformer making what sounds like white noise.

Tim

--=20 Deep Friar: a very philosophical monk. Website:

that snubber?

Well, I'm making 2kV here, and at my line voltage, it was making 3kV = before I installed the regulator:

I have learned it is a Good Thing to be paranoid about the voltage specs = on things. I put peak snubbers on everything. Probably not necessary = in this case, but better safe than sorry. It only burns a fraction of a = watt so it's not a big deal.

They did that by adding extra windings. I don't mind winding, but if = it's just for a snubber, what the hell, I'll just leak it off in a = resistor. When inductance is high enough, the loss is small.

I recently took apart an iMac power supply that had a UC3842 based power = supply: transformer, FET, clamp diode, the works. But it also had a = ferrite cored filter inductor. I don't know why, but they went with a = half wave forward converter, instead of the 3842's textbook flyback = circuit.

It was only a 5W resistor, so it's not like it was costing much = efficiency. Hell of a lot better than destroying your 800V MOSFET, too.

Not bad, though mine makes greater sparks :)

helium

hot.

Snubbers are OK to skip where there's a lot of capacitance. This = effectively makes a "lossless" snubber. The downside is, it gets very = frequency sensitive (think induction heater), or draws lots of reactive = current (bad for caps), or something like that.

For instance, your transistor sucks a big startup surge. It's only a = microsecond, so it doesn't matter, but if it were doing it every cycle, = in hard switching instead of ZVS, it would be dead. So it's not a = generally applicable approach.

Tim

--=20 Deep Friar: a very philosophical monk. Website:

On a sunny day (Mon, 12 Jul 2010 12:17:38 -0500) it happened "Tim Williams" wrote in :

OK, I was getting the wrong impression perhaps because you marked it as '1 W'. As for the high voltage, I had (or have?) a lot of those TV H output transistors, BU208 and things like that. They are great, 700V Vce or more.

Cool. I once was repairing a color set, and forgot to put the HV cap back on the tube, it was hanging of the workbench net to my leg. I got zapped with 25 kV. Was not so bad, quick to power down.

But much better to tune the thing? Square waves radiate a lot...

as

Oddly, I recently tried making an absurdly high ratio boost converter = with a 2SC5404 (same old 10A 1.5kVcbo thing). It was avalanching at = only 300V. I think this was because I wasn't driving the base negative, = only "off" (shunted to GND with a power MOSFET).

I replaced it with an 800V MOSFET, which dutifully went right up to 820V = peak. Not a bad ratio for a 16V supply.

=3D

reactive =3D

Tune?

BTW, you might want to check your newsreader's settings. In previous = posts, you've mentioned an interest in the workings of Usenet so I'm = guessing you'll be interested.

Take the above quote for instance, it has been erroneously appended with = equals signs. I'm guessing the encoding isn't set correctly. If it = were, either the equals wouldn't be placed, or my newsreader would = remove them appropriately.

In other passages (like the second line of your first quote in this = message), your random-length lines wrap at inopportune points, resulting = in ugly paragraphs.

I don't have a clue how my posts (quoted-printable encoding) appear to = other newsreaders; in my own, they wrap to the window width, which is = perfectly readable. AlwaysWrong seems to object as if they don't wrap = at all; maybe he has to read them with a horizontal scroll bar. In his = case, I consider that an advantage... but I wonder about others.

I do know that some newsreaders (including OE) have trouble quoting = "quoted-printable" (which seems rather ironic). The "> " marks don't = get added automatically, which is annoying.

Tim

--=20 Deep Friar: a very philosophical monk. Website:

I believe the usenet standard is to wrap lines (i.e. insert linefeeds) at ~60-70 characters (to allow for quoting on a 80 character screen I guess). It is also a convenient width to read, long lines are tiring. Mine warns me if I try to post ones much longer than this, so I wrapped yours (above).

Your long lines are a bit awkward to read IMO, but nothing too obnoxious:)

On a sunny day (Mon, 12 Jul 2010 13:02:39 -0500) it happened "Tim Williams" wrote in :

Yes, it needs a drive transformer, and some circuit to drive it strongly so to remove the charge.

I really have not tried any HV MOSFETS other then BUZ44A (400V) many years ago.

OK, let me try to explain my way of thinking, maybe bit vague for some, but it goes like this: It is much easier to keep a pendulum going then moving it fast left to right or forward backward, takes less energy. I like resonant, way of least resistance. As you have capacitance in the transformer windings plus lots of other places, why not make use of it and tune things.

Yea, this is true, the point is some newsreaders (specifically yours for example), a Microsoft product, seem to use that new encoding system. Mine does not and I am not in the mood to add it. So, in this post I have removed the '=' by hand. My suggestion to the maker of your newsreader (not that they give a shit) is to keep with the lowest denominator (right word?), but then again they dumped html on Usenet too. I cannot find anything in your newsreaders header that says I should process its text differently, except : MIME-Version: 1.0 Content-Type: text/plain; charset="Windows-1252" Content-Transfer-Encoding: quoted-printable X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.5931

No clue what 'OLE' is, sound like some Spanish football cryout.... So, anyways, why make something simple complicated, no MS dumps its latest shit on the world. Do not worry about it.

"Jan Panteltje" wrote in message = news:i1g15t$l60$ snipped-for-privacy@news.albasani.net...

Odd. A shame, I guess? 500 and 800V FETs are handy for = around-the-world supplies, 500V for PFC and forward converters, and =

IGBTs are better than FETs above 300V, and are cheaper per amp, but good = luck finding any inbetween 600V and 1200V, those seem to be the most = popular voltages for some reason. They aren't all that popular for = lower currents, where FETs still dominate AFAIK. They also seem to be = more fragile, not tolerating excess voltage in forward or reverse = (co-pack IGBTs, with diode, solve the reverse problem for the most = part). It's kind of weird, they're rated for avalanche breakdown = voltage (at a current of like 1mA), but I have never seen an energy = rating on them (like FETs have). As far as I know, more than a few mA = causes destruction for some reason.

some,

right

places,

Ah, but how much energy do you really burn in switching? In general I = mean. And note that, in general, "tuning" (which I guess does refer to = resonant or quasi-resonant snubbers, tuned circuits, that sort of thing) = is not possible, and obviously enough it increases reactive current, = which may be troublesome. A generalization can't be made, but it might = be anywhere from the miniscule amount of q =3D C * deltaV or Phi =3D L * = deltaI parasitics, to several times load current typical of LLC circuits = and etc.

The thing about tuning, and something you absolutely must remember: it = doesn't *help* you. If it's there unavoidably, you should take = advantage of it; but don't be under any illusion that adding reactive = components will make a circuit work better.

Two extremes make a great example. One, a high voltage generator with = huge capacitive secondary, and the other, a forward converter with = essentially resistive behavior.

For the HV generator, you could drive it with a single transistor (as = your laser supply does), with the disadvantages of: turn-on transient = switches full capacitance =3D big current spike, frequency and duty = cycle must be set to maintain class E operation, leakage inductance must = be low enough (or primary capacitance high enough) to minimize overshoot = at the primary (otherwise you need voltage peak snub), and the extra = reactive current causes increased heating in the supply filter cap, some = amount of current in the transistor and now-required damper diode, and = additional heating in both the primary and secondary due to that = capacitive reactance.

Further advantage could be taken by driving the transformer with a = [dynamic] current source, i.e., an inductor. This doesn't work very = well in half wave, but it does work well in full wave, e.g. PP or full = bridge. You get a Royer oscillator, where the voltage swings in = sinusoidal humps, both transistors operate in ZVS, with no reactive = current flowing through the supply (it's all squashed by the series = inductor, which can be arbitrarily large), and the output is a fairly = clean sine wave, limited only by Q and switching speed (during zero = crossing, when both transistors are on simultaneously, a flat spot = occurs).

The basic point is, since you have capacitance, you might as well use = it. This might involve canceling the reactance with inductors, or using = a capacitor-friendly approach, like CC drive. Keep in mind that, by = increasing absolute current and voltage, you are increasing losses, = maybe not as much as an explicit snubber, but some nonetheless.

The forward converter, on the other hand, works like this: the = transistor turns on, current shoots up proportionally, then it turns off = and current goes to zero. There is nothing to "swing" and start the = cycle, nor anything to end it; timing is arbitrary. Whereas the = mechanical analogy of a pendulum seems to be easier, this example is = more like dragging a weighted block across the floor: voltage (velocity) = doesn't go anywhere until you apply a certain amount of current (force, = static coefficient of friction). When you apply that voltage in either = direction, a proportional current (including direction) flows. It = doesn't help you to apply reactive forces (vibration, hammer taps, = etc.), because friction is friction (note: the dynamic coefficient of = friction is always lower than static, but in this electronic analogy, = they are equal). So either way, you have to apply the work, and yes it = takes some effort to get up to that level of force/velocity, but you = don't have anything to help you, and it's a completely lossy system = anyway, so it won't even 'feel' right. You'll have to apply so much = reactive energy that you'll definitely increase losses trying to reach = an underdamped (feels-kinda-resonant) system.

Now, this is all obvious to us, but it's amazing that it escapes the = overunity types. They are quite fond of two things: = rotating/oscillating machinery and circuitry, and anything that seems to = operate on little enough power that it "feels" lossless to the hand. = One silly example uses several miliwebers of seriously strong = supermagnets, arranged without any pole pieces whatsoever, in a ring = glued to a disc, such that the magnetic field on the periphery seems to = ramp up, going around the disc. The result is, when another magnet is = held near the disc, the magnets in the ring push away, and as the field = ramps down, the ring is propelled. Obviously, it won't be propelled a = whole revolution, because there's a step where the ramp begins again. = The astute operator just happens to nudge his wrist at this moment, thus = apparently making a "magnetic motor" that rotates without "any" energy = input whatsoever; the magnets seem to push themselves around! I expect = the two effects at work apparently convincing this operator of the = energy arises from 1. the force being almost entirely radial (=3D force = that does no work), except for an imperceptible tangential force (doing = the actual work), and 2. because the amount of force required to spin a = disc on a bearing is small, the energy input (that nudge of the wrist) = is also small, perhaps being imperceptible as well, giving rise to the = illusion that it happens of its own energy.

And apparently Tesla himself was convinced of this fallacy. Reactive = power is energy, not power. It is fundamentally impossible to = accumulate energy in a resonant system -- it will always come to = equilibrium for constant power input! Such belief has even been = professed on this newsgroup from time to time -- Jim Thompson once spoke = of an LC sinewave oscillator he constructed which required no "gain = correction", like the classic HP Wein bridge oscillator has. This is of = course another fallacy, since a true linear oscillator will grow without = bound, but given a sufficiently high Q, a small constant-current nudge = (of constant amplitude -- not a linear system!) to a parallel resonant = tank will appear as linear (low distortion) as anything else. The same = is true of a series resonant circuit and small constant voltage = stimulus, but series resonant tanks are harder to implement on the small = signal level; for a resonant voltage of 5V and Q =3D 100, you have to = drive the thing with a perfect voltage source of only 50mV, whereas the = parallel resonant, carrying 10mA reactive, can be tweaked with only =

Tim

--=20 Deep Friar: a very philosophical monk. Website:

To clarify things some more, for example IIRC in the TV the HV coil is tuned to the 3rd harmonic of the H frequency, 3 x 15625 Hz.

Not a bad ratio for a 16V supply.

you've mentioned an interest in the workings of Usenet so I'm guessing you'll be interested.

signs. I'm guessing the encoding isn't set correctly. If it were, either the equals wouldn't be placed, or my newsreader would remove them appropriately.

your random-length lines wrap at inopportune points, resulting in ugly paragraphs.

newsreaders; in my own, they wrap to the window width, which is perfectly readable. AlwaysWrong seems to object as if they don't wrap at all; maybe he has to read them with a horizontal scroll bar. In his case, I consider that an advantage... but I wonder about others.

They wrap to screen here if I turn on wrap option, right now they're going past end of screen since I set 250 char line width for posting to a place that requires no line wrapping sometimes ;)

"quoted-printable" (which seems rather ironic). The "> " marks don't get added automatically, which is annoying.

I'm curious how this will look without my usual manual newlines ;)

Pardon the interruption...

Grant.

Not a bad ratio for a 16V supply.

you've mentioned an interest in the workings of Usenet so I'm guessing you'll be interested.

equals signs. I'm guessing the encoding isn't set correctly. If it were, either the equals wouldn't be placed, or my newsreader would remove them appropriately.

your random-length lines wrap at inopportune points, resulting in ugly paragraphs.

newsreaders; in my own, they wrap to the window width, which is perfectly readable. AlwaysWrong seems to object as if they don't wrap at all; maybe he has to read them with a horizontal scroll bar. In his case, I consider that an advantage... but I wonder about others.

end of screen since I set 250 char line width for posting to a place that requires no line wrapping sometimes ;)

"quoted-printable" (which seems rather ironic). The "> " marks don't get added automatically, which is annoying.

There should be a wrap setting in your newsreader, in the same section where you set quote (>>>) marks. ...Jim Thompson

I'd like to read the his views on that. He doesn't seem the type to be easily duped. I know he felt there was an 'aether' permeating the universe, but that isn't such a big sin.

mike