Is that a charge pump at the gate for isolation? Can it be driven by microcontroller PWM on one pin and ground on the other?
It is; note that common mode between driver and power path will turn it on just the same. And that intermediate drive levels will partially bias the transistors, leading to high dissipation. It's reasonable for limited applications, but most of those will have the driver isolated, in which case, err, it doesn't need to be cap-coupled at all and can be tied common source with the switches...
Or the caps could be dimensioned small enough that AC mains for example doesn't trip it, in the mean case -- but can't be made small enough that the exceptional case (EMI or transients) won't. So it's not very safe for that reason.
Also, speaking of transients, there's nothing limiting gate voltage, which a transient could easily overdrive, and then your switch is always-on which is rather inconvenient... (Semiconductors tend to fail shorted.) Easy solution there, use 12V zener diodes for the charge pump.
A more refined version could use an LC resonant tank, to filter out most of that junk (mind, it's still susceptible at the resonant frequency); in which case the coupling capacitors can be made quite small thanks to the enhanced power coupling of a double-tuned resonant filter. Drive of course also needs to be fixed frequency, but that's rather easy to obtain from an MCU with crystal (or even the internal RC oscillators are usually precise enough for this) and a timer-counter.
Still better would be a balanced LC resonant tank, using a centertapped choke to couple +/- balanced into the network, so that common mode interference is cancelled out. Now you're using a transformer anyway, so I don't know that it matters much at this point... pulse transformers aren't terrifically expensive and the left circuit is simple and easy.
The one advantage to the resonant method is, if you don't mind the frequency being a LOT higher (alas, out of the range of an MCU + timer), you could use coreless planar PCB structures at zero parts cost (just design time and board area). Run it at say 433MHz from a gated crystal oscillator (use small schottky diodes for the low capacitance and no recovery loss). The isolation can also be extremely good (thickness of FR-4 is good for many thousands of volts). Another advantage is, the keying can be quite fast now (a modest fraction of Fo), better than most optoisolators and maybe even competitive with digital isolators (some of which use the same construction internally!).
But I digress; there's a fair amount of network theory and poking around to prove out something like that, and even in an ISM band, it may not be easy to meet emissions limits. More just as a flavor of what's possible, not necessarily feasible.
If I understand this right, if Vgs is being driven as described in my previous post (between -10 V and +10V) then connecting common source or common drain will not make a difference?
Yes, for Vg(on) - Vgs(max) < Vcm < Vg(on) - Vgs(on)(min).
For Vgs(max) = 20V and Vg(on) = 20V, the CM range is only 0...15V say (for a minimum Vgs(on) of 5V).
Like I said, it's practical for small voltages, a few volts.
SI1539CDL is a MOSFET pair I had in mind. It has Vgs +/-20 V, Vds +/- 30 V. Dont see it to be a problem for this circuit.
I meant in the range of +/- 10 Volts for low voltage AC. I wanted to distinguish from AC mains voltage or similar when i wrote low voltage.
Yeah that's not gonna happen, the span is already as much as Vgs(max) and that's already distorting as it runs out of Vgs(on) at the peaks.
BTW, another way to use the common-source switch, is to bias the gates from an external supply. Say the source and drive are common ground, and drive is +/-30V current limited to say 0.1mA, something small like that. (Easily enough done with a handful of transistors and resistors, as level shifting and CCS.) Tie this to the gates
and add a S-G zener so that Vgs(on) and (off) are limited, regardless of Vcm. Now you can run Vcm up to +25/-30V peak and have reliable operation. Note that the drive injects some current into the source or load, so this won't be an acceptable solution for every situation; for something like an audio amp, it's fine.
BTW, AC/DC (MOS type) SSRs are just the same thing over again, using a small photovoltaic (solar panel) stack powered by the LED itself. This delivers minuscule current (some uA), hence they switch quite slowly (some ms). They're quite handy -- just look at how much circuitry they save!
Tim