Cleanest Transistor for Audio ON/OFF Switching

[jonnyeye]

I had a little epiphany today while re-reading this thread.

R.G. brought up using two JFETs as an L-pad (one series and one shunt) to ensure very good muting. This has a new set of problems; either you use one P-channel and one N-channel and have simple control circuitry (but getting the right devices is more difficult) or you use two N-channel (and have to build control circuitry). Here is where my eureka moment comes in: A CMOS SPDT switch is an L-pad, if wired correctly. How do? Here are two options:


(The chip layout was stolen shamefully from Maxim's datasheet)
When CV is low, the switch connects IN to OUT (with a small and somewhat non-linear resistance) and opens OUT to GND (or rather connects it with a very large resistance). So the input voltage passes through essentially unaltered. When CV is high (mute), there is a very large resistance between IN and OUT, and OUT is also connected to GND through a low resistance. This ensures that the signal is always attenuated greatly in the mute position, no matter what impedance you are working into.
Note that +-12V is just a suggestion; it can run on any bipolar power supply from (say) +-5 up to +-15V (check the datasheet), but this changes the threshold voltage for switching if wired as shown. However, you do have the option of adjusting the CV threshold voltage (as alluded to in the image); by bringing the VL pin to any voltage up to V+ (a resistive divider should work), the threshold is set to 2/3 of the voltage on pin 5 (VL). The DG419 is not exactly a cheap device but should be about $2 in unit quantities. If that seems too rich for one 8 pin device, or if bipolar power is not available, then check out the following:


(Sorry RG)
Again, 9V is a suggested voltage to run this from, but anything from 5 to 20V will work (although the threshold voltage changes again with applied voltage, and there is no easy way to alter it here). This version is conceptually identical to the previous version, except that we need to set up a local ground Vr for this circuit to do its thing (the DG419 version side-steps this because of its bipolar supply). The two resistors to generate Vr are equal value, anything from 10k to 100k should be fine. Distortion may also be somewhat higher than the DG419 version due to the reduced working voltage and the different specification/construction of the underlying devices, but is likely to be tolerable in most cases. Total parts count is about $1 or so, although you could replace the electrolytic in and out caps with film if you have a disdain for electros.

(Note that these are untested (I don't even have a DG419 in my stock!) and may need some tinkering to work properly. No refunds on free advice.)

[anotherjim]

I did think of 4053, however...
Size seems critical to the OP.
Buffered control input means you can't slew the switch action. An RC would only introduce a time delay before it switches fast. Same goes for the DG series switches (didn't know they were still available).

[PRR]

> anything from 5 to 20V will work

You are pushing luck with 20V on an 18V-rated chip.

> the DG series switches (didn't know they were still available).

Many of the DG-series switches were used in huge quantity in factory and hospital gear. Some of them will be available for a long time to come.

[tempus]

Is it possible to ramp the control voltage on a TLP222 like it is on an H11F1, or will this simply delay the time it takes to turn on very quickly?

[anotherjim]

It seems not, at least not perfectly...
http://stompville.co.uk/?p=423

[amptramp]

Is it possible to ramp the control voltage on a TLP222 like it is on an H11F1, or will this simply delay the time it takes to turn on very quickly?

One more reason to go with the CD4007 or MC14007 - the control voltages are brought out to pins and are not buffered.  You can delay turn-on and turn-off and tailor the R-C networks or use diodes to accelerate either turn-on or turn-off as the requirements require.

[anotherjim]

Could we perhaps use the idea in that TLP222 of inverse series MOSFET channels with, say, x2 2N7000's? This puts the parasitic substrate diodes of the FETS in opposition. Solves that problem with single MOSFET's?

[bool]

So you'd make a simplistic (yet convoluted) mosfet-relay ...

That could be done; but I think you'd have to come up with a clever-ish bi-phase switching scheme if you wanted to use a series-shunt config (with added RC smoothing)...

[anotherjim]

Is was thinking the on resistance could be low enough, you might get away with just a shunt.

[tempus]

One more reason to go with the CD4007 or MC14007 - the control voltages are brought out to pins and are not buffered.  You can delay turn-on and turn-off and tailor the R-C networks or use diodes to accelerate either turn-on or turn-off as the requirements require.

But the H11F1 or TLP222 give you the advantage of isolation from the control signal. In the article quoted, the author found that he couldn't effectively ramp up the control voltage on the 222 - is there a way to tell this from the device's datasheet or is it only ascertainable through trial and error?

As a side note, the CD4007 is an attractive option, with 3 switches in a 0.63 package, provided it can be set up to switch noiselessly. I've tried to use BS170s for audio switching, but they distort as a result of the protection diodes. Will this not be an issue with this device?

[anotherjim]

Well now, in the CD4007, it is a bit different with those pesky diodes...


A MOSFET is formed on a bulk/substrate which ends up with that diode effect across the channel. You can call it a protection diode but it's really a parasitic that can't be avoided. It isn't necessarily suitable for sole flyback protection from a big inductive load.
Anyway, having more connections, the 4007 MOSFET's turn the substrate diodes to the nearest power rail.  Signal would have swing beyond the power supply rails to clip those diodes.

I have used the 4007 successfully as voltage controlled resistor and signal switch and so have many others. I think its use becomes a good idea when all of its parts can be found a use. Like JFET's, variations in thresholds means you need to tweak the relationship between control level and signal bias to get the right control response with every build. I personally do not find this onerous. For self & boutique builders having to do a set-up should not be a deterrent.

With the photo-mosfet, it seems linear control is more difficult. Possibly the gates are so sensitive that they switch fully almost the moment the LED emits some photons?
That's why I thought you might at least borrow the series N-channel MOSFET idea from the TLP222, since you then have access to the gates. For an RC slew control, the last gate that switches on fully affects the shunt mute and the first gate that switches off removes it.

[amptramp]

Finally! Some love for the 4007.  If you make one input the signal and the other the Vcc/2 ground, this circuit:



shows you how to make a switch with series and shunt switching elements.  You can add R-C lag networks from pin 6 to pin3 and pin 13 and 8 to pin 10.  Pin 13 is the drain of the upper transistor and pin 8 is the drain of the lower transistor and 8 and 13 are connected externally.  The economics of dealing with IC's is skewed; you get six MOSFET's for the price of one and the CD4007 / MC14007 IC is available anywhere CMOS is sold.

[bool]

Is was thinking the on resistance could be low enough, you might get away with just a shunt.

I think not only that, but if you pulled the both sources down with a high-mohm bleed resistor down, you could simplify the feed the gates control circuit immensely. You would just need to "isolate" the both gates with a two high-mohm resistors and feed them via a "nice" RC constant. (for the complete circuit; five resistors, one cap and two mosfets).

Otoh, good luck gents - I had a working noise gate with a series-connected BF960 (!) and a 555/741 control circuit in 80's - and it wasn't exactly easy to make it work "as expected". Fets are bitches to get right, but they do the job when you get them to behave.

[anotherjim]

x2 series 2N7000 shunt does work. "Linear" control for fade is narrow, just about 1v - 2V relative to ground. During fade the channel diode cancellation is lost, since (I suppose) one or other MOSFET is bypassing it's own diode, so there is distortion during the fade. This would probably be ok in guitar use with a reasonably fast fade. I would expect TLP222 would suffer from this also.
Isolation from signal (ground referenced by 1M) when the mute is off appears good, but I didn't feed it a large test signal.

Channel On resistance isn't low enough for a good mute against a 1k input resistor. 10k is better. Output probably ought to be buffered although 10k is ok into normal Hi-Z inputs, we don't always have those.

[bool]

To get a better fade window, imho you'd need "perfectly matched" mosfets. At least that's what I "ass"-umed when I simulated something along these lines. Meaning that in the world of computer models this will have some desirable performance, but when there's real-world device mis-match involved, the linear "fade window" will become much narrower.