I vaguely recall somewhat "lossless" snubbers that I implemented at GenRad (1977-87) that were diode-capacitor-resistor-inductor combos that worked so well that I made a fool of myself by grabbing the flag (un-heat-sunk) of a device to check its temperature forgetting that there was 300V P-P there... the technicians went hysterical ;-) ...Jim Thompson
| James E.Thompson | mens |
| Analog Innovations | et |
On Tue, 24 May 2016 16:41:44 -0700 (PDT), email@example.com wrote:
The loss-free snubbers depicted in the ON app and other locations, employing L, C, and D in varying quantities and orientation tend all to suffer from the same phenomenon, when using diodes that have a definable reverse recovery. As the reverse recovered charge is directly proportional to peak forward current and inversely proportional to rectifier area (ie size or current rating), the attempt to use rectifiers with sensible average current ratings in the snubber results in charge multiplication. More reverse recovery charge (and peak reverse current) develops in the snubber than is anticipated. If this peak current is passing through the primary winding, then leakage energy in the winding at snap-off will still require somwhere to go - typically to the negative supply through the parasitic body diode of the main fet switch. This diode is usually much slower than parts normally selected for switching. If the peak reverse recovery current in the snubber is preserved to pass into the parasitic diode, it will be returned to the primary for a third time....and so on. The behavior described may tend to defeat the purpose of the snubber, without intentional damping loss, or oversizing of snubber silicon. If the reverse peak current is less than the forward peak current, in each case, then the snubbing will eventually tail off. Effects may look logical on paper. Circuitry may even demonstrate a degree of functionality over a short-term and narrow line/load range - something that has no doubt spawned a host of learned technical articles to be written. I've got a host of them. Only testing over the full line/load and environmental range can reveal whether the components selected can always function practically, as intended.
This may sound like bullshit to many, but until you've actually scoped the current and voltage waveforms, I would at least advise caution.
The actively switched snubber, that returns energy to the flyback load (and can facilitate ZVS in both main and snubber switches), is promising, but requires a variable frequency and timing control that is not simple. Primary switch turn-on current may be negative, a behavior that causes many control methods to react adversely/ chaotically. This doesn't prevent it's use in some practical circuits, with varying effective advantage. In many cases, it is only required to function over a well-defined range of operating conditions, including the transient or temporary sub-optimal.
If energy is collected somewhere, it can always be re-converted by a separate circuit, synchronously or otherwise, if the cost of the aditional complexity is justified.
I'll have to go look at whatever clever idea I had. The circuit ended up not working but not, I think, because of my inadequate snubbing.
A circuit that I did not work on directly, but which I know worked well, was a fairly high-power flyback that just had a diode-capacitor snubber, but which then had a switcher from the snubber cap's voltage back to the primary DC rail. I can't remember whether it was called an active snubber or a regenerative snubber, but the circuit worked.
(The company went down the tubes -- it was a startup trying to make a super-efficient solar panel inverter that did not use electrolytics anywhere, and would, thus, "last forever". It was run by someone with a PhD in engineering, and I'm pretty sure it died due to lack of a sales and marketing network.)
Control systems, embedded software and circuit design
It's practical when LL is large and kinda-sorta-intractible. Like with high output voltage converters, where the cost savings (for a simple, sloppy transformer winding) outweighs the parts cost (a bootstrap gate driver and second transistor).
Such is used in forward and flyback converters, of the "two switch" (half wave) type. The transistors are operated in sync, and the excess flyback (including LL) or reset energy simply dumps back into the supply.
It can also be done push-pull, using a diode on the "far" side of the primary. Usually, the primary is bifilar, or the reset winding is made with smaller wire. This only addresses P-S leakage of course, not P1-P2 leakage, which remains. I've only ever seen a few of these; I expect they're all but historical relics nowadays.
Probably the best electronic engineer I know has a Ph.D. in Engineering - on a design study for an electric motor for oil tankers using super-conducting windings.
His nick-name around the lab was "Puker" - which can be translated as somebody who makes other people throw up - because of his habit of coming up with simple, horrible solutions (many of them one-transistor) for other peoples circuit problems.
EMI took out some 25 patents naming him as the inventor, or one of the inventors, which do commemorate rather more complicated innovations.
Nobody would have wasted his time getting him to run a business - you can hire people for that.
Jim's just being Jim. Button pushing is what he does when he's bored, like so many other denizens of SED. If he had a Ph.D., he'd be saying the same about "undereducated engineers" or people who think that a MSEE makes you a "master" like him. ;)
He used to rag on me like that, until we collaborated on a project. ;)
As far as competence goes, the worst engineer I've ever encountered had a Ph.D. in "Industrial Engineering", a very nebulous field that (as I belatedly discovered, to my cost) seems to be a dumping ground for people who couldn't hack real engineering grad school.
This guy was lead designer for a company that was hired to productize an instrument proto of mine, one that worked very well but didn't look very pretty. (*) Among other exploits, he coded up a digital lock-in--in LabView--using *least squares curve fitting* to pull a sinusoid out of noisy sampled data, instead of multiply-and-accumulate with N whole cycles of the sine and cosine of the (known) frequency like a normal person. (Turned out he knew zero signals-and-systems.) To make it worse, he had a DC restore function that masked the problem to some degree.
Like most crappy detectors, it kinda sorta worked at high SNR, but fell completely apart at noise levels that a real lock-in wouldn't even have noticed. The client (an angel-funded start-up) called me back in at the
11th hour to iron it out, but there were so many onion layers of crappiness in that project, and the crapmeisters had gone so far over budget, that they ran out of money before I could find them all. A pity--it's still a great gizmo.
(*) It also used an RC airplane servo to rotate a diffraction grating, which was fine as long as it had a preload spring. You do _not_ show the FDA an instrument made with toy parts. ;)