Connect with us

Analog switch methods

Discussion in 'Electronic Design' started by Tim Williams, Dec 5, 2008.

Scroll to continue with content
  1. Tim Williams

    Tim Williams Guest

    In thinking about oscilloscope sections (see for
    background), I want a high bandwidth (out to about 20MHz) SPDT switch:
    something a little faster than a relay, maybe do a clean 1MHz with
    corresponding rise time (under 50ns would be good).

    So there's the classic diode switch, ala fast sampler stuff. Foward
    bias FWB with CCSs and the AC terminals become connected (within
    capacity of the CCS). Reverse and it opens up. Easy enough. So put
    in two (that's eight diodes), and alternate their current sources for
    the SPDT action. And to do that, I can use diode gates (counting 16
    diodes...) and a complementary square wave, which can easily be made
    with reasonable rise time (down to 30ns or so from mere 2N3904s). And
    with just a pF or so between the diodes, isolation won't even be too
    bad up at 20MHz (figuring a line impedance of 50 ohms I think). Takes
    an awful lot of diodes though, and I wonder about linearity -- it's
    that first 0.5V that needs to be beautiful.

    JFETs are classic switches. Not usually too conductive, so I'm
    thinking moderate impedance (500 ohms-ish?) at which point Cds would
    be worth making a "T" switch. And that's fine, the ground part is
    switched inverse to the signal parts, and there's two of them (a tee
    for each input) that are switched inversely, so I need the complement
    anyway. Switching hooks up about the same as for the diodes, maybe
    even using current sources (that's six of them...) to supply the gates
    for faster response (a current sink on the output compensates for the
    gate bias). MOSFETs are just as classic, but transistors with
    substrate taken out to a pin are harder to find.

    Or there's the BJT switch. Little known, and rarely used (I only know
    of one "invertible" transistor, the 2SC2878, of which I have two).
    But surprisingly effective, having MOSFET-grade "Rce(on)" specs (C2878
    says about 5 ohms at Ib = 10mA), low enough to couple 50 ohm lines.
    Capacitance is a bit high though, not really giving enough isolation
    at 20MHz, even at 50 ohms. Maybe some smaller transistors would work
    here (C2878 is some 400mA Ice(max)).

    Tim (damn googlegroups...)
  2. Guest

    SD214 if anybody is stll making it. The series ran from SD210 to
    SD214, but haf of them had protection diodes on the gate, which could
    be inconvenient.
    Any old transistor can be used as an invertign switch, but the 5V
    reverse voltage rating of most planar transistor base-emitter
    junctions is inconveniently low. Broadband transistors are even worse
    at 2V but some of them do have very low capacitances.
  3. Tim Williams

    Tim Williams Guest

    Looks lovely (although I'd like less Rds(on) -- I suppose it's a
    compromise between high Cds and high Rds though, so I'd have to cut my
    losses with a tee switch).

    Surprisingly SD214 shows up at Mouser, with "Lead-Time 27 Weeks" and
    the price, "48: $182.70". No, thanks...
    Searching Mouser, I see some transistors with Veb = 10V or so, which
    are rated for switching duty (Ron < 1 ohm at good Ib). I don't know
    how, but the 150mA type has 20pF Cout at Vce = 5V whereas the 500mA
    type has under 10pF...umm okay... And I suspect the 2SC2878 doesn't
    exist anymore.

    Come to think of it, there's a lot of unused base charge in one of
    these switches, isn't there? If you're using it at hFE < 1, isn't it
    going to take a while to stop being a switch? And which BE junction
    does the charge go into, anyway? It's kind of both, depending on
    which way the signal is... if it's swashing back and forth, that must
    make some awful time constants and nonlinearity, right?

  4. Guest

    Try a FST3125 or similar, if you don't need more than a volt or two of
    swing. These are cheap, fast, NMOS only switches with built-in gate
    drivers, and the substrate tied to a separate (negative supply rail)
    pin. (You'll have to provide a negative rail about equal to the
    minimum of the signal swing, and a positive rail 5V or so above that.)

  5. Tim Williams

    Tim Williams Guest

    Plenty fast, but looks like TTL or HCMOS- 7V max. So I can pass a
    +/-2V signal, which is fine, but I'd need a +/-2.5 or 3V supply to do
    that. I'd rather run from +/-5 or, better yet, +/-10V. And there is
    startlingly little data in either Fairchild or OnSemi datasheet - I
    mean, no isolation data? No V/I curve? The 4066 makes me feel more
    comfortable than this thing... too bad it's just not good enough.

    No comments on using BJTs? Didn't one of the Johns here do something
    with those? John? Anyone? Bueller?...

  6. Guest

    Actually the FST3125 ron will get higher with swings close to the
    positive rail and it will turn off completely if both the source and
    drain of the NMOS switch reach within a threshold voltage of the
    positive rail. The threshold voltage under this condition is
    relatively high because the back-gate (substrate) is biased
    permanently to the negative rail AFAIK. Nevertheless, some amateur
    radio people have been making quite good mixers out of these cheap
    digital bus switches (google H-mode mixer). You would probably use
    +/-1V swings, with a -1V and +4V power supply. I would not recommend
    using a digital bus switch in an analogue circuit for a commercial
    design unless you can make a lifetime buy of a qualified version of
    the switch so that you are immune to the manufacturer changing the
    process or design in a way that is only a problem for analogue
  7. I've always been a happy user of the SD210 series MOSFET
    switches. I've also used the NXP BSD22, which is similar,
    but cheaper. Alas, they are all discontinued.

    NXP still has the BF1108, which has a rather lower RDS-on,
    but also much lower voltage ratings.

    Jeroen Belleman
  8. cassiope

    cassiope Guest

    How much isolation do you want? At what bandwidth? If you need more
    than a single SPDT switch can provide, have you thought of using two
    layers of switches to defeat blowby? [i.e. with a standard CMOS
    switch, and using the first layer to switch between ground and that
    signal, and the second layer to select between one signal and ground

    Sure, I like discrete switches - I designed a few into some Tek
    both junction devices and SD21x's. But if your bandwidth is as modest
    as you say, there's no reason these days to do that unless you want to
    learn more about the old method.

    Have fun!
  9. Tim Williams

    Tim Williams Guest

    -40dB would be good. DC-20MHz. 50 ohms switching would be good, so I
    can just run everything low-assed-impedance straight through (with the
    understanding that some signal will be lost across the switch, which
    can't be a nonlinear response!). Peak amplitude about a volt, so it
    should handle up to 20mA.
    I have, take for instance these examples;

    I don't think I can find JFETs small enough (2N3819 grade capacitance)
    and conductive enough (J105 grade conductivity) to do 50 ohms in one
    pair, so they'll have to do it in two -- a tee switch. And it's two
    tees for the A/B action. This would have to be followed by a follower
    (possibly with another JFET for high impedance input, though right
    into a diff pair probably wouldn't be too bad anyway) to get back down
    to 50 ohms.

    MOSFETs, like the SD214, are pretty close, but I'd have to hunt for
    them too. 1N914s I have in my parts bin...
    Old methods? Ha! Just be glad I'm not using tubes in the output
    stage! ;-) The low capacitance of a 6DJ8 is tempting, though.....

  10. cassiope

    cassiope Guest

    You need to be clearer on exactly what you are trying to do. 50 ohms?
    Why? What sort of voltage/power range? You cannot expect to get
    sensible answers if we don't have your context.
  11. Tim Williams

    Tim Williams Guest

    Oh, real specs? Sure...

    A/B switch for ALT/CHOP function on (mostly) discrete oscilloscope
    design (I think I alluded to this earlier). Source impedance 50 ohms
    (or something else, like open collector or emitter follower (Zo < 10
    ohms), if need be). Load impedance whatever, 50 ohms (50 ohms
    straight through) preferred, but if I need for example a FET follower,
    then I need a FET follower. Isolation as mentioned. Switch time
    under 20ns, without much glitching if possible. What else?

  12. Guest

    I still say try the FST3125, though you will need the awkward +4V/-1V
    supplies. I have used these switches to handle +/-1V or so signals in
    50 Ohm systems up to a few MHz (though they would ba OK for higher
    frequencies) and found them very satisfactory, especially for the
    price. If you want to have one with proper analogue specifications,
    then you can buy something similar (but maybe a bit slower switching
    time) from Analog Devices, but the price will be higher since it is
    not marketed or specified as a logic device.

  13. Fred_Bartoli

    Fred_Bartoli Guest

    "Tim Williams" <> a écrit dans le message de
    Oh, real specs? Sure...

    A/B switch for ALT/CHOP function on (mostly) discrete oscilloscope
    design (I think I alluded to this earlier). Source impedance 50 ohms
    (or something else, like open collector or emitter follower (Zo < 10
    ohms), if need be). Load impedance whatever, 50 ohms (50 ohms
    straight through) preferred, but if I need for example a FET follower,
    then I need a FET follower. Isolation as mentioned. Switch time
    under 20ns, without much glitching if possible. What else?


    I was wondering what you really wanted because channel switching is almost
    never voltage switching but rather current switching.
    Think current steering...
    Have a look at any reasonnably modern Tek scope service manual.
  14. Guest

    Needed to switch a 50ohm, 50MHz, 2V sine via two LC quadrature
    networks (DPDT), feeding a NE602 mixer. Hugely optimistically I
    ignored the datasheet and wired in a '4053 with predictable results.
    Desperate for some numbers, started putting together a discrete
    version and 6 hours later had a breadboard stuffed full of bits,
    massive signal loss and virtually no isolation. 3 hours further on and
    a second breadboard dissed me with mediocre signal level and 20dB
    isolation. I know when to call a halt, so the next day bought a cheap
    Fairchild FSA450 video switch from Farnells. Sorted in one!.
    I'm a fanatical believer in cheap discretes but occasionally packaged
    Silicon offers the best option.
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