Connect with us

Should I use a SG3525 ?

Discussion in 'Electronic Design' started by Pooh Bear, Nov 16, 2004.

Scroll to continue with content
  1. Pooh Bear

    Pooh Bear Guest


    I'm designing a forward converter. Probably be full bridge as opposed to
    half - at the several kilowatt power level - probably @ 100kHz or maybe
    a bit more.

    Plan to use dedicated gate drivers btw.

    For the controller, the SG3525 seem to be a popular choice. Multi
    sourced and a classic workhorse. Used in some current products I've
    looked at ( in the pro-audio market ).

    It's been around a while though. Just wondering if I'm missing an
    alternative that is superior in some ways.

  2. CBarn24050

    CBarn24050 Guest

    Subject: Should I use a SG3525 ?
    Hi, if your using a full bridge then you can use a phase shift control
    stratergy which makes your drivers much simpler, especialy if you plan to use
    drive transformes. Unitrode make a chip, the number escapes me, or you can use
    a PIC there is an application note somewhere on the microchip site.
  3. Pooh Bear

    Pooh Bear Guest

    Would you mind elaborating on that ?

    I've also considered one of Philips' super fast 8051 variants. I'm a bit of an
    8051 fan / small-time expert but have zero experience of PICs despite considering
    them from far back ( and dislike single source too btw ).

    Thanks, Graham
  4. Yes, a PIC at a buck and a half each plus S/W development time or a 3525 at 10
    cents each and *no* S/W development time.

    Why is it when people don't know the real answer they always say "or you could
    use a PIC" ?

  5. Pooh Bear

    Pooh Bear Guest

    Quasi-religious faith ? Looked at them - single source - uhuh !

  6. Phil Allison

    Phil Allison Guest

    "Pooh Bear"

    ** The folk at QSC do seem to like them a lot.

    ** At the the "several kilowatt" level you ought to be considering using a
    PFC SMPS design - gives you regulated DC rails at well.

    Might pay to get a SMPS expert to do that one for you.

    ................ Phil
  7. Terry Given

    Terry Given Guest

    Stick to peak current-mode control, rather than voltage mode control.
    That peak primary current limit is a wonderful thing. Especially when
    you fit a large saense resistor at the beginning, limiting the current
    to a nice small amount (say 3x expected mag current).

    UC3846 was the original part for this job, but unitrode have a much
    nicer version now with better rise/fall times, faster current limit
    comparator, more GBW opamp, much less current etc. I forget the number

    If you go voltage-mode and something goes wrong, KA-BOOM. A good CM-smps
    ought to run forever into a dead short on the output.

    Unitrode, Cherry etc all have nice app notes on how to do it. When you
    have made it work at 50W, then start cranking the power up. Then when
    you have made it work at several kW, go build a phase-shift version and
    throw out all that primary switching loss (hint: same power stage etc.
    just a different controller, and perhaps a transformer with more leakage
    inductance). But beware the drop in duty cycle. And lookout for > 180
    (or < 0) degrees too :)

  8. John Walton

    John Walton Guest

    You can get 1.5kW out of an outlet for a toaster, but not for an SMPS power
    supply. (that is to say, a standard 117 VAC residential outlet in the U.S.)

    Under certain circumstances, both halves of some controller chips can
    conduct --
  9. Phil Allison

    Phil Allison Guest

    "John Walton" <
    ** Pooh in in the UK - ie 230 volts at up to 15 amps continuous.

    Using a PFC corrected SMPS design up to 3 kW of continuous DC power is

    ............... Phil
  10. legg

    legg Guest

    A phase modulator normally drives it's switches with a ~constant 50%
    duty, allowing reduced drive transformer and ancilliary circuit

  11. Pooh Bear

    Pooh Bear Guest

    Ok - I'm with you now. Seen the method described but never seen it in practice.

    Since I'm planning on using gate driver ICs - the transformer issue doesn't arise.
    Not sure I see how 'ancillary circuit complexity' would be reduced though.

  12. Pooh Bear

    Pooh Bear Guest

    Can you suggest ?

    Not a mention of the 3525 here ( apart from the specs of the improved B
    version parts )

    I'll look at the 3846 though.
    That's pretty much what I'm planning. Very much bit by bit.
    Because audio amplifers have very rapid changes in load current beyond the
    capability of the feedback loop to respond, popular practice seems to be to
    run the controller @ max duty cycle anyway and accept that the output volts
    will refelct variations in ac line ( PFC sorts this of course ).

  13. John Walton

    John Walton Guest


    Quite an interesting article in the November/December issue of QEX in which
    an ATX supply is modded -- used in a ham transceiver which is operated both
    in phone and CW modes -- for a widely ranging output current (1 to 20 amps
    in this case) the most critical element is the inductor in the LC filter.
    The value of the inductor is inversely related to the minimum current drawn.

    At any rate, it seems that Class A amplifiers would like to mate with SMPS.
  14. Pooh Bear

    Pooh Bear Guest

    That figures. A nice steady current draw.

  15. Terry Given

    Terry Given Guest

    Hi Graham,

    with an open-loop full bridge converter, a judicious choice of Dmax,
    transformer leakage and mosfet D-S capacitance (perhaps in parallel with
    added caps) the bridge can be Zero-Voltage Switched (ZVS). The
    upper-lower switch dead time is selected to be 1/4 of the period of the
    resonance between the leakage inductance and FET capacitance. That way
    one switch turns off, an LC resonant transition to the opposite rail
    occurs, then the other switch turns on.

    This starts to get tricky when duty cycle changes (hence the use of
    phase-shifting between legs, to keap dead time constant), but open loop
    it can and does work very well. And is generally used after a PFC, for
    the reasons you state.

    You ought to be able to find some papers describing this topology/

  16. Pooh Bear

    Pooh Bear Guest

    Hi again Terry.

    I had been thinking simply along the lines of 'hard switching'.

    Sounds to me like you're talking about something that uses some resonant
    converter type thinking.

    I'm following you part of the way at least. The trick is to ensure that the
    mosfet/igbt that's turning on does so at a low Vds / Vce it seems. That then
    reduces turn-on losses.

    You'd have to control leakage inductance pretty carefully to do this surely ?

    You could reduce turn-off losses if the load current decayed during the power
    transfer cycle too I guess.

    There's a practical example I have that may be doing this. It has an LC in series
    with the TX. I thought it might be a fully resonant converter but testing it on
    the bench suggested not. I was hoping to extract the 1 page of 5 from a pdf
    schematic to illustrate. I can probably do this toimorrow. Would welcome your

    Regds, Graham
  17. Terry Given

    Terry Given Guest

    Hi Graham,

    When you do a hard-switched full-bridge converter (FBC), you (should)
    organise a small amount of deadtime between the upper and lower
    gatedrives of one half of the bridge, to avoid shoot-through (both
    switches on simultaneously, shorting out the DC bus). Often this is done
    with asymmetric gatedrives, eg a diode across Rg for slow turn on, fast
    turn off. Good PWM controllers allow you to set this dead time
    A FBC driving an inductive load always (somebody will bite me for that
    statement :) turns on at zero-voltage: when the lower switch turns off,
    mag current commutates to the diode of the upper switch. Then the upper
    switch only sees a diode drop before it turns on - voila, ZVS. Alas the
    lower switch gets a hiding at turn off. Often the leakage inductance
    alone is enough to make this happen.

    The idea then is to increase this dead time (say from 200ns to 1us), and
    slap a big(ish) cap across each FET. The Lmag + Lleak then makes the
    voltage on the lower FET slowly (thanks to the biggish cap) rise (LC
    resonant circuit), so the lower FET has time to turn off while Vds is
    held low by the cap. You can make it work with the FETs own capacitance,
    but swamping it with an external cap is a very good idea in production,
    as FET capacitance is not generally a controlled parameter.

    Of course this resonant transition occurs even in a FBC (there is always
    some L & C) but its usually way too fast to help turn-off. At very high
    PWM rates (MHz), no external caps are necessary (see above caveat).
    yep. or add an external L. Leakage is controlled entirely by winding
    geometry, so if you can make that constant, leakage will be too.

    Often simply changing the winding topology is sufficient, but beware
    proximity effect - for example moving from a sandwiched winding to one
    atop the other will significantly increase leakage (4x or so, IIRC
    leakage is proportional to the square of peak MMF), BUT proximity effect
    may get a whole lot worse (the number of effective layers just doubled;
    with a sandwiched winding 0.5S-P-0.5S the effective number of layers in
    the Primary is half the actual number). At high power its more efficient
    to design an optimal, low-leakage transformer, and throw in a little
    series L. Usually its quite small.

    on proximity & skin effect:

    proximity effect = skin effect, but caused by currents in adjacent

    With skin effect only (say a 1-turn winding on a choke) if you increase
    wire size, DCR goes down. But the ac-dc resistance ratio goes up, and
    the total resistance remains constant (ignoring the optimum point, which
    isnt a great deal better). OTOH more copper = better heat conduction, so
    dT will drop a bit.

    With proximity effect its very non-linear, and strongly depends on how
    many layers in the winding. I once re-engineered a 1500W 400V - 24V
    dc-dc transformer. It had 12 layers of 0.6mm Cu foil, running at 100kHz
    (delta = 0.2mm) so the foil was three skin depths thick. The windings
    were interleaved, so the effective number of layers is 6. Using Snelling
    fig. 11.14 the Fr (ac-dc resistance) ratio is about 80 - i.e. the AC
    resistance is 80x the DC resistance! These transformers set their
    windings on fire (the bobbins didnt burn, but were totally destroyed).

    I dropped the Cu foil down to 0.1mm, ie DCR increase 6x. But Fr drops
    from 300 to 1.3 (told you it was non-linear), so AC resistance = 7.8 x
    original DCR, compared to 80x - in other words the AC copper losses went
    down by a factor of 80/7.8 = 10x. The >> 200C temperature rise at full
    load dropped to around 45C. And all the doubting thomases who had said
    "rubbish" to my solution were forced to eat a decent size piece of
    humble pie. Being the modest chap I am, I gloated mercilessly.

    reductio ad absurdum: remove the load completely, turn-off losses will
    plummet :)
    will be interested to look.

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