Connect with us

IGBTs are pretty fast

Discussion in 'Electronic Design' started by Tim Williams, Jan 31, 2006.

Scroll to continue with content
  1. Tim Williams

    Tim Williams Guest

    Since there were no objections (..or encouragements..) to the proposed
    arrangement, '719 is the bridge as I've reassembled it.

    '720 is the output voltage and current waveform. Note this is at 120-130VDC
    supply, and 29kHz or so (some amount before resonance around 21kHz).

    I like these transistors more than the MOSFETs in that you can see the
    discrete transistion as current passes through zero; the transistor won't
    conduct backwards, so the reverse drop will always be that of the co-packed
    FRED. Likewise, the forward bias is always a couple diode drops.

    '721 is the rise and fall time of the output. 2.5V/ns is pretty good for
    something this chunky.
    On the left of this oscillograph, you can see a small dip, which I suspect
    is miller capacitance. The gate fall respectively has a small blip on it as
    it passes around zero. About 280ns after, Ic falls and Vc swings
    completely in about 50ns.
    When the FRED turns on, it clamps with about 10V of bounce that rings around
    8MHz (t ~ 120ns), plus a lower harmonic. This sounds reasonable compared to
    the published resonance frequency of the closest bypass caps of ~10MHz.

    Any objections, comments, suggested measurements/conditions before I move on
    to 200V supply?

  2. Terry Given

    Terry Given Guest

    - can you show a picture of the gate waveform, measured at the igbt?

    - the output voltage waveform looks like it has a 100V spike on the
    rising edge. can you zoom in on that?

    - can u take a pic of your scope probe setup? that spike might not be these low voltages, use a coax scope probe tip adaptor, a
    BNC socket an a short length of tightly twisted wire, soldered directly
    across G-E or C-E.

    - how far away is your gate drive setup? put it as close as possible
    (creepage/clearance notwithstanding) to the IGBTs.

    - your DC bus inductance is on the order of 500nH or so. if you get a
    couple of pieces of double-sided Cu-clad PCB, and some 1mm nomex/lexan
    you can make a DIY multi-layer PCB. just folding your existing assembly
    flat would make a large difference.

    how fast is your scope? whats its rise-time spec?

  3. Terry Given wrote...
    That may be an attractive debugging tool, but it's far better to
    have the reverse current flow through a conducting FET channel
    than a diode with its reverse-recovery delay and snapoff after
    shutoff. Unfortunately at high currents and voltages, MOSFETs
    simply cannot keep pace with IGBTs in conductance capabilities.
    That's why you will find them in this type of application.
  4. Tim Williams

    Tim Williams Guest

    As in junction charge stuff? Wouldn't that just subtract from the load
    current, making it transition a little slower?

    Saturation and diode voltages are both within +/-2V, even 1V around current
    zero crossing. I see no harm in having a particularly slow recovery, as
    long as it's fully recovered by the time it has to switch off.
    Yeah, so, when reverse-biased with a good 50A or so, there's going to be
    diode conduction one way or the other, it doesn't really matter if there's
    [reverse] current in the FET junction.

    Which reminds me, if these IGBTs don't behave, I'm just going to buy a pair
    of 600V FETs from Digikey (STW70NM60 looks pretty reasonable, and has low
    enough Ron that I can even use my current desat circuit). So what if it'll
    cost me three percentage points efficiency, it'll *work*...

  5. Tim Williams

    Tim Williams Guest

    5V/div vert, 2us/div horiz, waveform before gate resistors.
    Ditto, waveform at gate.
    Spikes are:
    Same vertical; 200ns/div horiz. (at gate).

    Note the spikes coincide with output voltage transitions, but are of the
    wrong polarity with respect to voltage. (In #2, gate voltage is falling,
    but the pulse starts negative whereas output is rising.) Thus I took a pic
    of the emitter waveform, about 1mm from the package.
    (Same phase, time and voltage scales as #1 and 4.)
    Period is around 150ns for the pulse, i.e. 6.7MHz.

    Incidentially, though the pulse appears on the transistor end of the gate
    drive lines, the circuit side is clean. Alright, so I'll get some PCB and
    solder up the drive circuitses...
    That's from the current waveform. '721 is the output waveform zoomed,
    showing about 20V overshoot.
    The current transformer is a black ferrite toroid (probably high mu) with
    quadfilar wound 26AWG totalling 280T (4 x 70T in series), and a 2.8 ohm load
    resistor. Twisted pair leads back to the scope, with a few turns around a
    ferrite bead for good luck. (I know, the CT is floating, and hanging on the
    ground lead, so it shouldn't pick up much ground loop style hash that a
    ferrite bead would be used for, but so what.)

    The output waveform is measured with my 10x Tek probe clipped to the output
    terminal at the DC side of the coupling capacitor.
    Not very healthy then... about 12". I do have twisted pair, which is a
    Transmission Line(tm), though..
    How can that be? The 0.1 caps are an inch from any transistor, and all
    together should be on the order of 100-200nH.
    So the strips are flat with the heatsink y'mean?
    SFA. The output waveform has no difference vs. vertical bandwidth
    (switchable 20/100/200MHz). The oscillographs were taken at 100MHz. All
    the same, it is spec'd at 1.8ns or so.

  6. Terry Given

    Terry Given Guest

    then the pulse probably isnt "real" but it shows several important things:

    - your probing technique is picking up stray fields (or perhaps even
    oscillating by itself, but my guess is H). dont use a ground clip, take
    it off and use a BNC tip adaptor (the scope probe should have come with
    one, at least in theory), and a bnc with a short length of tightly
    twisted wire. that ought to make a significant difference. a simple test
    is to probe to your ground clip. If you see a spike, its the probe.

    oh yeah, make sure the wire you use has suitable insulation voltage rating.

    read Jim Williams LT AN47, and do what he says.

    - there is obviously plenty of stray field to pick up - invariably its
    due to the loops in your power circuitry. you cant do much about your
    work coil (other than ensure the current is sinusoidal), but contain the
    rest of your fields.
    its well twisted, but still....with care it should work fine, but making
    it work gets trickier as the gatedrive moves further away. and until it
    works, it destroys IGBTs :)
    I like to use coax terminated into 50R (I have some bnc thru
    terminators, but a T and an end terminator are about 10x cheaper) at the

    you can also place a faraday shield between the CT windings - suitably
    insulated Cu pipe, extending a short way from either side of the toroid,
    and one end connected to, say, earth.
    WAG. but the loop looks to be about 1" square, so a perimeter of 4". if
    we use 0.01" for the "wire" thickness,
    L = 0.00508*4"*[ln(4*4"/.01")-2.853] = 100nH

    OK, 200nH total :)

    so using the same IGBTs and bus caps, you can halve the total inductance.
    yes - they make parallel-plate transmission lines, which have very low
    inductance - Uo*spacing/width

    Ye Gods - look at the gate paralleling resistors. they form a great big
    loop,which (like your scope probe) *will* pick up any and all stray H
    field. remmeber the gate is just a cap, so a low Rg doesnt really help
    much here - and besides the inductance of the loop (and Tx line) also
    increases Zg, and the gate drive output impedance will also rise with
    frequency (it looks inductive).

    you have to be careful with your power paths, but *PARANOID* with gate
    drives. I have had loops smaller than this cause fatal problems, and it
    wouldnt surprise me if that is the case here.

    thats fast enough.

    an HP 17xx?
  7. Tim Williams

    Tim Williams Guest

    Well, the probe works fine- it only picks up a couple hundred mV when
    grounded as such. Little enough that I can ignore it on the 5V/div scale.

    I'll admit a forehead-slapping-worthy moment: the above readings were taken
    with the gate drive through a ferrite bead. I've had weird results in the
    past so I have the scope grounded to the circuit, and the circuit and output
    connected with the gate drive ground only.

    That means any current appearing between the output and circuit grounds will
    cause a voltage to appear across it. The gate drive itself is perfectly
    fine and the world is happy -- but since the scope *probe* wasn't going
    through the FB too, it got a nasty bounce!

    I now have an 8" clip lead ferrying the signal through the FB and the
    emitter shows a pulse of like 4V peak, 30ns wide. The gate OFF bias is more
    than sufficient to cover this, at this current anyway.
    Humm. Second hit on Google is quite attractive, but not quite an
    Application Note. :))~
    and one end connected to, say, earth.

    I'll keep that in mind.
    Yabbut, it's not a continuous loop- IGBT each side of the gate resistors, so
    the induced field would tend to bootstrap, no?
    Would you suggest, uh, Idunno- twisted pair for each IGBT? Maybe twisted
    pair off each, then tee them together after a ferrite bead, so as to allow
    some common-mode squishiness? The emitters should be doing the same thing
    so connecting lines together shouldn't be a problem ... but that says
    nothing of differing turn-on/off speed between the two.
    Hell no ;-)

    Tek 475.

  8. Terry Given

    Terry Given Guest

    yes it is. ignore the 2nd IGBT, consider a single gate circuit.
    hopefully you have a 0V plane on your gatedrive PCB, so we can ignore
    the loop at that end; similarly they are well twisted so we can ignore
    the contribution of the dangly wires. but you then open up into a bloody
    great (physical) loop with the gate resistor. At the very least, squish
    it right down. Ideally, join the emitters of a pair of paralleled IGBTs
    with a nice flat copper strip, with the gate drive 0V connecting to
    this, and the gate lead and Rg's sitting on top of it - IOW a "ground
    plane" around the IGBT gate connections/parts.

    it doesnt matter how well twisted or ground-planed the rest of the
    gatedrive is, this loop buggers it up.

    Firstly, it will pick up any H field and convert it into a gate voltage.
    Me and a buddy once spent 2 weeks tracking down such a loop-related
    problem in a drive - if we powered up the DC bus just right, it would
    turn all the IGBTs on, and blow up :) In the end a scalpel and a bit of
    wire-wrap wire cured the problem.

    Secondly, the inductance "softens" the gate drive response to an edge.
    For Rg = 10R, 100nH has the same impedance at 16MHz - a rise time of
    around 20-30ns. for negligible contribution, rise times need to be less
    than 200-300ns, which they are NOT. So when some evil dV/dt happens (say
    every edge), current flows into the gate, raising (or lowering) Vge....

    Its pretty easy to mess with an IGBT's gate if its got a nice big loop.
    Its a often lot easier to mess with the gatedrive circuitry itself, its
    really just an amplifier, so you need to be absolutely certain H wont
    cause you any problems. ground plane, ground plane, ground plane.

    thats why you have separate Rg's.

    the single twisted lead is fine, its what you do after the twisted lead
    that is the problem.

    can U show us a pic of the gatedrive PCB....
    I love my 7904 :)
  9. Tim Williams

    Tim Williams Guest

    Alright, so I "at least squish[ed] the gate resistors in".
    There's a strip of cardboard there to try to ensure they don't short out.
    The bottom left (high side) resistor runs across the transistor, placing it
    closer to whatever loops are in the transistor, and the collector (B+ rail)
    current waveform. This should at least be similar to the emitter current,
    no? The alternative is to route it down and around, which opens up a 1/4"
    loop, which is "bad".

    The upper right (low side) transistor has the same problem and solution,
    except there's only the transistor to "ground plane" it.
    every edge), current flows into the gate, raising (or lowering) Vge....

    Yabbut that's what the negative bias (-5V OFF state) is for, as I recall,
    isn't it? (That and emitter L flyback.)

    Current can fall at 50ns (IGBT t_off), but gate voltage falls in around
    200ns. It easily passes the linear region in 50ns though.
    It isn't.
    It's all mounted on an aluminum backplane, but I don't see a ground for it.
    Can always mount a screw for one, I suppose. The circuit seems to be
    working fine at the moment.

  10. Terry Given

    Terry Given Guest

    ye gods
    one single intermittent connection can destroy your IGBTs.

    such breadboards are great for prototyping, I've made gatedrives on them
    myself. And once the circuit operates correctly, re-build it on a piece
    of copper-clad PCB. that way bits wont move, impedances are far more
    controlled, stray inductance reduces (perhaps dramatically), and wires
    wont fall off.

    when it all works nicely, you can then build a little box around the
    circuit using more Cu-clad PCB, and solder up all the edges. that helps
    keep all the nasty fields out (and/or in), as well as clipped off leads etc.

    also dont forget the dangers of small components (eg nuts, washers),
    tools etc.

    make sure all the HV stuff is well secured (eg screwed to a large
    plank), with a shield overtop (1mm lexan is good stuff). that will help
    prevent blowups, contain the carnage and reduce the shock hazard. If its
    on the floor, dont slip and fall on the bus-bars while its live.

  11. Tim Williams

    Tim Williams Guest

    Sounding just a little stark there Terry...
    Well, it hasn't yet, and I've had a few instances, so I don't know what to

    I have, evidently, blown (open, not shorted) the PNP gate drive/follower
    transistor (which is still only a 2N4403, not the ZTX I purchased for the
    purpose). This results in a slow, constant turn-off time, since the current
    mirror is in effect through what's left of the Vbe (I guess the collector
    blew in this case). And, of course, the desat shoots, turning off at least
    one half of the bridge within 3µs.

    I have had one instance where something happened to the gate and it stayed
    high, resulting in the power transformer groaning *as if the bridge had
    shorted*. In reality, the battleship sized transistors were just owning it.

    Not very healthy... but if it happened with the old wiring I would probably
    be down another $40. The tighter bridge wiring works much better.
    myself. And once the circuit operates correctly, re-build it on a piece
    Inductance, sure, but I'm still not seeing how the breadboard is going to
    screw things up. Honestly, I've used longer air runs between components
    when messing around with the output of my 1ns pulse generator, and the
    pulses still plink around well enough (20% rule in effect) whatever I have
    hooked up.
    Hmm... fields...(yeah, "hummmm indeed", ha! ha ha!). Well the thing is,
    it's pretty much as bad as it's going to get, *as is*. But I'm not getting
    any trouble, even with the induction coil less than a foot away. So if I
    pack it up on a PCB, I'll have less to worry about, which means...I'll have
    nothing to worry about?

    All the spurious signals occur on edges, and the gate drive and all circuits
    already know what they are doing on the edges, plus I have power supply
    bypasses scattered about, so I don't really worry about it.
    That's not a bad idea. I almost peed my pants last time I had some MOSFETs
    go off like shotguns.

    The strange thing is, though, the last about 20 transistors I've
    burned...didn't. Fuckers won't even tell me who died!

    I guess that means I'm getting good at this solid state thing, but I'd much
    rather they just ooze the smoke so I don't have to lift and probe each damn
    lead to find the dead one(s)...

  12. Terry Given

    Terry Given Guest

    how can you be sure? intermittent connections tend to work except when
    you are looking.
    an open-circuit could do that :)
    heres a clue - "something" shouldnt happen once the circuit works.
    try running the bridge at full power, then bashing the side of your
    gatedrive mockup with the handle of a large screwdriver. wear safety

    the interconnects in prototype boards start out as cheap and nasty.
    Unlike fine wine, they do not improve with age. A diode-like
    interconnect once cost me, another engineer and a tech 3 days once - we
    narrowed the fault down to a resistive divider that didnt work linearly.
    replacing the protoboard fixed the problem; the old one got Widlarised.

    inductance causes three problems:

    firstly, the loops radiate H fields, making EMC compliance harder.

    Secondly, they pick up H fields, and can convert them to gatedrive signals.

    Thirdly they increase the output impedance of your gatedriver - trace
    the loop from gate thru Rg, npn or pnp (turn-on or turn-off loops),
    supply rail, cap, 0V, emitter - the loop is in series with the gate.
    didnt your IGBT bridge blow up? doesnt that count as "trouble" ?!

    now I've seen the gatedrive construction, I'd list mechanical problems
    at the top of the "why my igbts died" list.
    solve the mechanical issues first. while you do that, you might as well
    build it on a ground plane - it will take no longer than soldering
    together a rats nest, while minimising susceptibility to stray fields.
    its cheaper to not break them in the first place.

    power electronics is more about how you do things than what you do. the
    circuitry is often the easy part.

  13. Tim Williams

    Tim Williams Guest

    Er, so when I'm looking, they tend to be intermittent, thus, I would know
    about them?
    Screwdrivers aren't part of the design equation, you're changing the
    conditions! ;-)
    Yabbut, moot point as the loops aren't particularly wide and the only signal
    they recieve is, at most, in the milivolt range -- a pulse could cause a
    comparator to switch early, but only when it's about to switch anyway.
    I have 0.1uF ceramics at the transistors, so the output loop on the board is
    under an inch diameter. There's more before the lead turns to twisted pair!
    The circuit, before and after the faliure, tested fine. It was the bridge's
    fault, as near as I can tell.
    Afraid I have to disagree on this one. Heh, the IGBTs blew too fast for any
    mechanical fault to have a reasonable propability to show up. ;)
    As in those RF lashups?

    For GHz circuits I would take the time and tediousness for it, but for
    pete's sake Terry my edges are two and a half orders of magnitude slower,
    and even as nearby as things are, the inverse square law is with me on stray

    I can do point-to-point wiring on perfboard, or a step up from that, the
    perfboard RS sells that has individual copper pads. This doesn't lend
    itself to ground plane technique very easily. I don't see PCB happening any
    time soon since I don't have PCB design software, resist, etchant, or any
    reasonable way whatsoever to drill the holes lined up properly.

  14. Terry Given

    Terry Given Guest

    what, too chicken to try it? why not, you seem to like your gatedrive

    how do you know that?

    it is often quite easy to cause gatedrivers to switch several times on
    any given edge. that can be a great way of making switching losses much
    higher than you expected. I have tracked down several such problems in
    the past, that exhibited themselves as "random' failures during soak

    after enough time spent tracking down these sorts of faults, one learns
    to avoid them in the first place.
    ignore the twisted pair. that inch or so just added a hundred nH or so
    to Zg.
    you blew up the igbts without breaking the gatedrivers? thats a good
    trick, normally the collector shorts across to the gate, and trashes the
    gatedrive output stage (or more).
    you obviously dont understand my point.

    one single intermittent connection in a gatedrive/igbt assembly can
    *destroy* the power electronics. your construction technique (OK, I was
    going to say "hairy-assed mess") is just *begging* for such an event to
    occur. Hell, it may have done so already (and IMO probably has).

    whats worse, once you fix everything, the intermittent connection might
    not be obvious.

    the only practical solution is to build the damn thing properly.

    plus murphys law applies directly here.

    fancy pushing on some of the proto-board wires while its running? no?
    why not?

    AFAT inverse square law goes, look at the impedances too - if its a nice
    high-Z circuit node, it might not need much to push it around. and with
    enough wiring inductance, even a low-Z node can look high-Z to a nice
    fast edge.

    hell, once 50fF of capacitance buggered up a perfectly good circuit I'd
    designed. yep, 0.05pF. 5V square wave with 10ns edges, 50fF to the -ve
    input of an opamp, with Z = 10k or so.

    thats a moot point though - if you need to improve the mechanical
    construction, you might as well do it in as "RF" a manner as possible.

    why not think about it from a risk management perspective? them IGBTs
    aint cheap, it should behoove one to try not to break them.

    it does if you sit it on top of a ground plane.

    I don't see PCB happening any
    I dont do those things either, I build circuits on top of a piece of
    copper-clad board. some chips end up upside down, others dont. go read
    the two books edited by Jim Williams, and/or some of the analog devices
    & linear tech app notes on how to build a decent prototype.

    with a bit of practice, its no slower than using proto-boards.


    there are a wide variety of problems that can, and do, beset power
    electronics. physical construction is usually at the top of the list (my
    circuit looks OK, why does it blow up) - both from an EMC and a
    reliability perspective.

    If you deliberately avoid the really dumb mistakes (a wire fell off, a
    clipped lead got into the hardware, I forgot to fasten the IGBT to the
    heatsink, giant loops everywhere, DIY thru-plated pcbs etc) that leaves
    you free to focus on the real problems.

  15. Tim Williams

    Tim Williams Guest


    LOL. I should take a picture of... no heck, take a video, of me dropping a
    screwdriver on the circuit, with the scope watching gate drive outputs and
    watching what happens.

    Probably end up something like shorted power supply, or the signal just
    stops, or something. If I were to do this, I'd be more worried about the
    voltage regulators letting out smoke than anything else... simply because
    nothing else is designed to push as much current.

    Oh- FYI, the breadboard itself is actually in pretty good condition. Nice
    and stout springs, not even any melted holes! (yet)
    It would've done it by now? Idunno.

    How could I prove it either way? Lesse, I could use an air core coil say
    10" dia. for the series matching inductor, and wave the board through it.

    That should induce a pretty sick current in anything of note, eh?
    Ya tell me about it... that's why my BK 3026 needs a fix... :-o
    No I understand your concern, I just don't really see it happening in near
    probability (I can see you worried about once-in-a-decade events, like
    dropping a screwdriver on something ordinarily sealed in a chassis, but to
    me that's a freak accident and I certainly don't mind the down time fixing
    the circuit, sans expensive transistors of course).
    Eh? I don't get you. Of course I push on wires, being a low voltage
    circuit I often twiddle wires and resistors and capacitors while live. As
    long as the high and low power sections are seperate I can develop then
    test, in that order. If there were scratchy contacts, I would've tracked
    them down by now.
    Well, yeah...but that doesn't change the measured fact that the gate drive
    works (in lieu of catastrophic mechanical climate change, so to speak ;),
    and pretty reasonably for an LM393 and five 2N440x transistors. I have
    fault protection, albeit rudimentary (local desat would be better, but I
    would have to have three-way communication to shut down high side, low side
    drive and oscillator sections when either drive poops, plus reset them all).
    The only question remaining is mechanical rigidity (perhaps I didn't
    articulate this, but obviously this breadboard isn't permanent, it will be
    soldered some day -- when the circuit becomes *set in stone* mind you) and
    RFI concerns, which I have so far seen few symptoms of.
    Pointy underside bits with voltage don't really like flat conductors. I
    don't know what kind of an insulator you would recommend there, besides
    distance, which in that case I would call it shielding (like those tin cans
    on various TV and monitor boards) more than a ground plane.

    Speaking of shielding, the whole thing (if possible) will be inside a mild
    steel box or two, which should control coil EMI and switching noise
    reasonably. Plus a line filter..

    Alright, let me put it this way. If plane practice is better...
    ....Why does everything I take apart have printed circuit boards?
    Production aside.

    I'm not trying to attack your point of view, that's absurd- you're literally
    in the business. I'm trying to apply your view to my situation is all.

  16. Terry Given

    Terry Given Guest

    you raise a very good point - just how do you "know" that this is
    happening. self-interference can be devilishly hard to measure. The only
    sure-fire way I have found is with making changes and observing problems
    disappear; reverse the change, watch the problem come back. repeat until

    then examine the root cause of the problem, and resolve never to do it

    sooner or later, it will happen with your gatedrive - it may have
    already, there isnt really any way of telling.

    heres a wee tale - it involves a gatedrive design for a 20-drive product
    range. we never tested it with the biggest IGBT, and it lacked a bit of
    grunt. the poor gatedrive increased losses enough that the drive would
    die after 2 hours on soak test, so we didnt release the larger drives,
    and went back to the drawing board. time was short, so when we got a
    proto pcb laid out, our CEO went and ordered 1500, rather than the 10 we
    would have got. Sure enough, there was one mistake - 2 pins of a
    comparator were swapped. So we lifted a leg on a SOIC8, added a few
    dangly wires and off we went.

    except the damn drive blew up after a couple hours on soak.

    re-sounding bang.
    re-peat, with teeth a-gnashing

    eventually we tracked the problem down - an intermittent connection
    *within* the LM393. a heat gun could make the gate drive turn itself off
    and on - the pin concerned was the reference voltage against which the
    isolated gatedrive signal was compared. looks like we damaged the bond
    wire bending the leg. perhaps 10 times in a row (over a period of
    several days) using new chips each time. Hmm.

    so we re-did the layout, the circuit worked perfectly, and has never
    been changed since then (although about 50,000 drives have been made, so
    300,000 copies of that circuit).

    a few weeks later, the CEO lambasted my manager for "wasting so much
    money on prototype pcbs" :)
    I've learned to stay the hell away when its running, and fiddle with
    nothing. I also bite people who approach carrying cups of coffee etc.
    I'd consider it blowing up to be a symptom.
    sidecutters, and flip the PCB over so the Cu side faces away from the rest.

    its easier if you just learn how to assemble circuits on top of Cu-clad
    PCB. google manhattan method, there is a nice PDF and some truly lovely
    examples to look at.
    if done properly.
    what, you think that POS proto-board isnt ugly?

    besides, what self-respecting engineer trades functionality for aesthetics?
    several reasons. Firstly, most of what you dismember isnt a whopping
    great piece of power electronics. pull a few of them apart....

    PCBs can have ground planes too.

    most consumer gear is incredibly cost-competitive (HTF does one make a
    DVD player that retails for NZ$48 ?!), which is why they use the
    cheapest, nastiest PCBs known to man (single-sided phenolic paper, with
    many machine-inserted links). and a team of engineers to ensure the damn
    thing passes EMI (but only just, thats enough)

    a lot of consumer gear just doesnt work very well. its not uncommon to
    buy things that dont work at all, and nobody is suprised when mall-wart
    stuff falls to bits....

    power electronics is nasty stuff, and is what generates the EMI that
    designers must work around.

    also, if you know exactly what you are doing, ground planes are not
    mandatory. its just that they make life SO EASY....
    I make quite a nice living out of fixing other peoples EMI problems,
    usually by using a decent ground plane.

    one particular job, the PCB was large and a frightening mess. missed EMC
    by miles, and product regularly went bonkers. the solution: turn the
    artwork from a 2-layer PCB into a 4-layer PCB. Assign mid1 as 0V, mid2
    as +5V. delete all 0V & 5V traces on top and bottom layers. Voila,
    product now passes. perhaps 2hrs of work. the build volume was very low,
    so it was cheaper to add $20 to the PCB cost and spend almost no NRE.

  17. Tim Williams

    Tim Williams Guest

    Yeah, Occam's razor, change one bit at a time and all that...

    I think I'm going to put a few turns in series with Lmatch and wave it over
    the board, see what starts freaking out. Should be able to find quirky bits
    pretty easily and relatively non-destructively.
    Admit it -- it's better than some gnarly mess entangled through the air. ;-)
    Doesn't have to be a trade. Look at the P-52, Spitfire, etc.
    No doubt there, but what of 1980s monitors? The kind that are 19" or more
    and have BNC connectors for video? I've taken apart a few. Those were the
    days when they *cared* to put in perforated aluminum shielding.

  18. Terry Given

    Terry Given Guest

    I've done that sort of thing quite a few times (40W ham transmitters
    sitting next to drives, that sort of thing) in the past. and invariably
    the problems are:

    mechanical - interconnects, soldering, damaged smt parts, swarf, tools,
    hardware etc. IOW self-inflicted.

    electrical - H and E fields, stray inductance and capacitance are what
    separates simulations from reality.

    these "EMI" problems can be further divided as:


    any and all of which can be modified to solve the problem.

    stopping the noise at the source is best - control your fields. for H,
    keep loops small, use parallel-plate transmission-line construction,
    "ground" planes etc. E needs electrostatic shielding, and a ground plane
    is a good start.

    sometimes the path can be altered - eg orienting magnetics at right
    angles, placing a sensitive circuit well away from a noisy one etc.

    the victim usually needs to be "hardened" with power electronics, as it
    tends to sit in very close proximity to the source. ground planes solve
    a lot of the problems, but keeping impedances low is usually a good idea
    to reduce the effects of capacitive coupling to, say, power devices,
    busbars etc.

    if you read a few books on EMI, and fix a few problems, you pretty soon
    see how the physical construction of circuitry is so important.

    simulation tools are good enough that gross circuit problems can be
    sorted out before building actual hardware. The circuit effects of stray
    L and C are dead easy to simulate. its instructive to play "what if"
    with spice, and sprinkle L's and C's at various points in a circuit.
    not as weird as an MLC capacitor exploding when charged to half its
    rated voltage. because it was hand-soldered, develop thermal stress
    cracks and failed within a week.
    I've seen some masterpieces built that way :)

    attempted facetiousness. all the truly great stuff looks cool too. stuff
    that looks dreadful usually is.
  19. Tim Williams

    Tim Williams Guest

    Soldered the gate drives the last few days, taped a piece of cardboard under
    them and positioned over the heatsink. Aside from the arrangement of the
    buss strips and the yet-ungrounded heatsink, this should work pretty
    reasonable huh?

    The biggest gate loops are basically the transistors themselves. All
    external loops are under 1/4" the best I can tell.

    The gate drives alone test well: estimated output impedance 1 ohm, 500mA+
    source/sink capacity, 0.65us low-side propagation delay, 0.8ns high-side
    (the coupling transformer adds 150ns); output edge fall time 360ns (RC
    slope), rise time 140ns.

    I need to tweak the UVLO circuit, and the desat is untested (obviously,
    since I haven't tried the drives with the output circuit yet).


    Deep Fryer: a very philosophical monk.

  20. Terry Given

    Terry Given Guest

    Hi Tim,

    that looks great. does it go?

Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day