Connect with us

High voltage push-pull output stage

Discussion in 'Electronic Design' started by Stefan Heinzmann, Dec 22, 2003.

Scroll to continue with content
  1. Hi all,

    I'm trying to come up with ways to build a DC coupled amplifier that can
    produce +/- 400V output voltage swing at up to 100mA. Please don't
    lecture me about dangerous voltages and the risk of killing myself. I
    know. I just want to check out what the alternatives are and their
    relative merits.

    I think a push-pull ouput stage is called for, however there are no pnp
    or p-channel transistors around that can withstand 800V or more. The
    output devices therefore need to be either npn BJTs, n-channel MOSFETs,
    IGBTs or tubes/valves. So you need some form of phase splitting driver
    stage that ensures proper bias for the output stage. The amplifier
    should have good pulse response, but needn't have very good distortion
    ratings or wide bandwidth. The step response should settle reasonably
    well within about 30µs. Amplification is a fixed factor of 10.

    I have experimented with MOSFETs and a traditional phase-splitter in a
    spice simulation but got no reasonable results. I tried to use the two
    outputs of a differential amplifier, but that didn't work right, either.
    The problem always seems to be the bias for the MOSFET whose gate swings
    up and down with the load. I finally used a pair of optocouplers to
    couple the differential input amplifier to the output stage while giving
    me freedom in shifting the voltage. That sort of works. I need some
    locally stabilized voltage for each MOSFET gate driver, however. For the
    moment I use a zener diode together with a constant current source made
    from a depletion mode MOSFET. I also thought of using tiny DC/DC
    converter modules, but they're not cheap. And it is not really elegant.
    I don't like it.

    Are there any other ways? Or information on phase splitters that work
    properly down to DC? How would you approach the problem?

    Cheers
    Stefan
     
  2. Stefan Heinzmann wrote...
    Traditional push-pull circuits with N-type devices can work if
    you're using an output transformer. But in that case you can
    use a step-up ratio and don't need to use high-voltage parts,
    or stick to one polarity type for that matter.

    Given that 1200V MOSFETs are readily available, it's easy to
    make up to +/-600V dc-coupled amplifiers, using a totem-pole
    output stage like we show in AoE, fig. 3.75, but with an added
    negative rail and level-shift stage. Here's the basic idea:

    .. basic high-voltage MOSFET dc amplifier
    .. by Winfield Hill
    ..
    .. + supply rail -----------------+-------,
    .. | |
    .. R11 |
    .. | D
    .. +--+--G
    .. C2 | | S
    .. ,--R4--||-, | C |
    .. IN | | | B--+
    .. o--R3--+--|+\ | | E |
    .. | >---+--R5--, | | R10
    .. ,--|-/ | | | | out
    .. | E | '----+-+--R12--o
    .. | gnd -+- B | D1 |
    .. | | C +--|<|----+
    .. | | | | |
    .. +---------- | -- | ------- | -+--R2--+
    .. | | | | | |
    .. R1 R13 | C3 | '--||--+
    .. | | E ,-||--+ C1 |
    .. gnd +- B | | |
    .. | C R8 | R9
    .. | | | D |
    .. R14 +---++--G C4
    .. | | | S |
    .. | R6 C | gnd
    .. | | B--+
    .. | | E | Add a zener across
    .. | | | R7 each FET's gate.
    .. | | | |
    .. - supply rail --+----+----+----'

    The circuit is inexpensive and works well at low frequencies
    but it can be further enhanced for a -3dB bandwidth over 100kHz.
    As you can see the circuit includes short-circuit protection.
    My finished designs generally have more parts to add features
    such as HV-output monitoring, reduced cross-over distortion,
    increased slew rate, improved power-supply rejection, etc.

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  3. Yes, sure. That's what the tube/valve guys have done all from the
    beginning. A center tapped primary on the output transformer allows the
    usage of one polarity type very easily. But as I need DC coupling, that
    wasn't an option. You are well aware of that of course.
    Still haven't bought AoE, I have to admit. Last time I stood in front of
    it I thought I would be rather annoyed to find the next edition to be
    available a few months after I bought the old one. So I decided to wait
    some more. But it is getting longer and longer...

    (But I don't want to urge you. A hurried book often is a poor book).
    Thanks a lot! I'll put this through simulation.

    The transistors either side of R13 are PNPs you stacked in order to get
    the required Vceo rating, correct?

    The proper value of R11 seems critical to me, as there will be a
    compromise between output rise time and power dissipation. Do you think
    a constant current source would be worth it here? (It would probably
    need to be built from a high voltage depletion mode MOSFET, and there
    aren't many...)

    R9 and C4 are a snubber network, right?
    I noticed the current limiting. I agree with you that this is a
    necessary feature. Not sure what you mean with HV-output monitoring. Do
    you think of a second error amplifier to shut down the circuit in case
    of output overvoltage?

    Cheers
    Stefan
     
  4. John Larkin

    John Larkin Guest

    Nice circuit. D1 can be a zener, and you can bootstrap R11 for a bit
    faster pullup.

    This is a true class B widget, which works but sort of always bothers
    me mentally.

    John
     
  5. John Larkin

    John Larkin Guest

  6. Stefan Heinzmann wrote...
    Sadly, I can say don't wait; it's going to be a while. Furthermore
    the next edition is probably going to have masses of cool stuff in
    the present edition excised to make room for new stuff and to allow
    for a reduction in page count to keep the price under $100. In the
    end you'll want a copy of all the editions. :>)

    Yes, that's a cascode stage necessary for voltages above +/-300V
    or so, depending on what you choose for Q1. Examples are mpsa92,
    2n6520, ztx558, bf493, ksp94, fzt560, fmmt560, nzt560, ksa1625,
    stx93003, 2n5416, 2n5657, etc., depending on desired slew rate.
    Indeed. It's the Coss of Q4 and the Crss of Q3 plus Q4 that
    you're fighting (if high speed is desired, be careful to choose
    a small FET type). The R11 resistor(s) need to withstand the
    full rail-to-rail supply voltage, 800 to 1200V, etc. I use three
    or four power resistors in series to stay well under the maximum
    voltage ratings and to allow dissipating more heat.
    There are several issues: to increase the pull-up current (and
    thus slew-rate) for outputs near the positive rail, to reduce
    total power consumption near the negative rail, etc. The former
    can be handled in some cases by splitting R11 and bootstrapping
    the junction with a capacitor to the output, as John mentioned.

    As for a current source, you'd want to make it from multiple PNP
    cascode transistors, like Q1 Q2, because a high-voltage FET has
    too much capacitance. Generally it's easier to just use bigger
    resistors instead. One possible enhancement relates to R11; the
    amplifier's loop gain decreases at high frequencies, decreasing
    the positive-supply PSRR. Filtering R11 fixes this issue.
    A big issue is high-frequency compensation, especially into a
    capacitive load. C4 adds a permanent minimum capacitive load,
    simplifying the scene, and R9 adds a helpful feedback-loop zero.
    It's a pain to always find a HV probe for measurements (ordinary
    probes aren't rated above 250V or so, and at any rate one hates
    to stress expensive scope probes). The R1 R2 node can't be used
    as an accurate monitoring point because of C1. There are several
    good possibilities to solve the problem.

    John suggested using a zener for D1, saving one component. This
    is reasonable for low-frequency amplifiers, but the high zener
    capacitance would damage performance at even moderate frequencies
    by increasing the crossover distortion. In fact, some kind of
    magical network is needed at point X to isolate the desired fast
    crossover from the Q5 FET's pullup capacitance.

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  7. The highest standard one I know of in DPAK is 600V (Japanese).

    Best regards,
    Spehro Pefhany
     
  8. John Larkin wrote...
    The PA97 opamp is nice, but lacks an output current limit.

    We needed to replace APEX's incredibly-noisy PA42 mosfet-
    input opamps, so I created a small "daughterboard" PCB,
    mounting a PA97 but with a PA42 pinout. My PCB includes a
    current-limit circuit, plus my favorite LDN150 as a current-
    sink pulldown to reduce the PA97's high crossover distortion.
    We've made more than 25 of these little PCBs already.
    No, they use an internal cascode circuit, probably with two
    each Supertex VP1550 and VP2450 dies for Q3 and Q10, resp.
    We ordinary mortals can order the VP2450 parts. :>) Their
    schematic doesn't show other critical items. For example
    the input JFETs are 2n3955 (which is why it's quiet), so
    clearly they haven't shown the necessary cascode stage to
    work with Q6 and Q7. Their schematic shows 22 parts, but
    their website photo shows 40 or more. :>)

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  9. John Larkin

    John Larkin Guest

    I have a similar problem. I need a small, cheap, low quiescent power
    linear amp to drive an e-o modulator (maybe a 50 pF load) with
    risetime roughly in the 100 usec range, 800-1000 volts p-p.

    One cool thing to make is a fet-optoisolator cascode. The fet can be a
    depletion mode thingie

    |
    |
    d
    g---+
    s |
    | |
    c |
    ===> b |
    e |
    | |
    +----+
    |

    or, if the parts aren't available, use a regular HV mosfet and bias up
    the gate with a second P-V optoisolator.

    The obvious refinements are, well, obvious.

    John
     
  10. Hmm, you almost talked me into it. Need to sleep on it..

    I got a lot of crap in simulation until I realized that the OpAmp inputs
    are the wrong way round. Amazingly, I got situations where the output
    still toggled! Shame on me for needing so long to realize this!
    I had picked the first one already, which probably is THE standard high
    voltage small signal PNP.
    I tried it with the IRFBG20, as I could get a spice model for it. But I
    didn't get far enough yet to tell whether my design goals are met.
    It was precisely those issues I was worrying about.
    Thanks a lot for those tips! When cascoding PNPs for a current source, I
    realize that 2 transistors won't be enough. I haven't got any experience
    with the dynamic behaviour of such constructs. Do I need to worry about
    uneven division of the voltage drop during transients? Or will that sort
    out itself automatically?
    The load may be reactive in my case, so the provision is welcome.
    Ah, ok, I didn't think of this. Do you typically include such circuitry
    as a permanent part of your design?
    You meant Q4, didn't you?

    Cheers
    Stefan
     
  11. John Larkin wrote...
    That's not a hard requirement, i = C dV/dt = 1mA max, including
    50pF of FET capacitance. Really wimpy parts are allowed. :>)

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  12. John Larkin wrote:
    [snip...]
    What is it that bothers your mind? Is it the distortion problems?

    Cheers
    Stefan
     
  13. Stefan Heinzmann wrote...
    The BG20 is not a bad transistor, a bit old, but well suited
    because of its modest capacitance. I purchased 200 last year.

    You may or may not get a Spice model of the schematic to work,
    but don't rely on it to tell you much about the real circuit.
    That's because most power MOSFET spice models are completely
    unsuitable for low-level linear use. They fail to model the
    sub-threshold region at all, and most totally screw up in the
    critical low-current unsaturated-operation region. One needs
    accurate g_m and Id-vs-Vgs modeling at 0.1 to 100mA currents,
    and good capacitance-vs-voltage modeling as well.

    For example, see the g_m curve in AoE page 132 and the Id vs.
    gate-voltage curve on page 123. Power MOSFET models should
    show g_m leaving the g_m = I/Vt curve with a new sqrt Id slope
    in the Id = 1mA to 50mA region, but sadly many models show the
    FET complete off with the appropriate gate voltages!

    We've discussed this issue here on s.e.d., with my recent
    power MOSFET measurements used to illustrate the problem.

    I'll check the BG20 models I have, but don't hold your breath!

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  14. Stefan Heinzmann wrote...
    The transition from pullup to pulldown is one sad issue, but a more
    serious issue is the differing pullup / pulldown operating modes.

    In pullup it's operating as a common-source stage with low drain
    currents (1mA, etc) with a high-value load resistor (i.e. high gain)
    followed by a low-Z output follower. By contrast, in pulldown it's
    operating as a common-source stage, at higher currents (when high
    negative dV/dt is present), with the common-source load impedance
    given by the actual load, which may be reactive. Intuitively this
    is not a very appealing situation. But like John says, with some
    effort and a good understanding, it can be made to work quite well.

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  15. Winfield Hill wrote...
    Sorry, old chap, but the model is very poor for linear use.

    Below Vgs = 2.9V the current is nearly zero, with no log region
    as it should have. Between 2.9 and 3.0V it rises rapidly to 5mA
    (very poorly modeled), from 3.0 and 3.5V it increases to 120mA
    (not too bad, but probably lower g_m than a real part) and above
    4.5V it slowly meanders above 1A, finally reaching 5A or so.
    The spice model is not too bad above 50mA.

    ..SUBCKT IRFBG20 10 20 40
    * TERMINALS: D G S
    M1 1 2 3 3 DMOS L=1U W=1U
    RD 10 1 5.22
    RS 30 3 0.276
    RG 20 2 107
    CGS 2 3 483P
    EGD 12 0 2 1 1
    VFB 14 0 0
    FFB 2 1 VFB 1
    CGD 13 14 218P
    R1 13 0 1
    D1 12 13 DLIM
    DDG 15 14 DCGD
    R2 12 15 1
    D2 15 0 DLIM
    DSD 3 10 DSUB
    LS 30 40 7.5N
    ..MODEL DMOS NMOS (LEVEL=1 LAMBDA=2M VTO=3.1 KP=0.833)
    ..MODEL DCGD D (CJO=218P VJ=0.6 M=0.68)
    ..MODEL DSUB D (IS=5.81N N=1.5 RS=0.536 BV=1K CJO=151P VJ=0.8 M=0.42 TT=130N)
    ..MODEL DLIM D (IS=100U)
    ..ENDS

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  16. John Larkin

    John Larkin Guest

    I guess it's the idea of closing a feedback loop around a zero-gain
    deadband, especially when driving a capacitive load. A tiny bit of
    noise could be conjectured to flail the drive node like crazy. I
    suppose the residual AC feedthrough (fet g-s and the output diode
    capacitance) keeps the gain from being truly zero.

    John
     
  17. John Larkin

    John Larkin Guest

    Fairchild has an optocoupler with a 400 volt output transistor, which
    makes a cool 400 v p-p amp using just a pair of optos as the output
    totem pole. CTR is a little low, but usable. Too bad nobody makes a
    kilovolt photofet! All the optomos SSRs I've seen have Schmitt action,
    darn it.

    John
     
  18. John Larkin wrote...
    Most of my designs have a single conveniently-placed resistor
    that eliminates this reasonable but apparently-bogus fear.

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
  19. John Larkin

    John Larkin Guest

    That's OK if it doesn't steal too much drive; then it just makes
    things worse, but at least in a different way. It's is a nice way to
    kill crossover distortion in a complementary power amp and still run
    the output parts 'class B' which is thermally nice.

    I sometimes build transconductance amps, which is a whole different
    kettle of squids. The crossover situation there is tricky and also
    philosophically disturbing. But heck, I'm not a philosopher.

    John
     
  20. John Larkin wrote...
    It's especially appropriate for high-voltage amplifiers used to
    drive piezo elements, which look like a big capacitor. After
    the hard work is done slewing the output to a new voltage
    (charge/discharge the piezo capacitor), the amplifier has no
    further job except to maintain the output to a fine degree.
    The added resistor (a large value, as you say) helps to insure
    the viability of the amplifier's fine control at dc, where the
    output load appears to be an open circuit. Another application
    with similar dc "open circuit" load properties is creating the
    voltages for ion-beam lens, focus or steering electrodes. One
    of our systems has nearly 50 electrodes, so a simple compact
    inexpensive amplifier circuit is appropriate for 50 amplifiers.

    Thanks,
    - Win

    whill_at_picovolt-dot-com
     
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

-