CMOS switching to multiple outputs

Branching off from the discussion of CMOS switching in Nishmaster's console thread, but I decided to start a new thread as not to hijack his thread.

I can see how putting the switch at a summing node (for switching inputs) or in a feedback loop (for switching gain) is advantageous in terms of performance, but I'm struggling to see how you can switch 1 source between 2 outputs, without having to have 2 seperate opamps and switching the gain on each (so the switch changes opamp A from unity to -XdB, and opamp B from -XdB to unity simultaneously)

Also, does anyone have any experience with the Silonex Audiohm optocouplers? From Silonex's website its seems that you can get very good specs using an opto in both series and shunt simulatneously, however at basically 2 optos for an SPST it seems like it could get expensive, probably better suited to muting functions than routing. Pros and cons of these for switching when compared with CMOS would be very cool.

Again, I plan on picking up Self's Small Signal very soon, but was also interested in the articles on switching in Self On Audio, I don't really fancy buying the whole book just for 1 section though, does anyone know of anywhere online its available?

Cheers
Elliott

ej_whyte said:

I can see how putting the switch at a summing node (for switching inputs) or in a feedback loop (for switching gain) is advantageous in terms of performance, but I'm struggling to see how you can switch 1 source between 2 outputs, without having to have 2 seperate opamps and switching the gain on each (so the switch changes opamp A from unity to -XdB, and opamp B from -XdB to unity simultaneously)

Click to expand...

Well, that's what it takes...

Also, does anyone have any experience with the Silonex Audiohm optocouplers? From Silonex's website its seems that you can get very good specs using an opto in both series and shunt simulatneously, however at basically 2 optos for an SPST it seems like it could get expensive, probably better suited to muting functions than routing. Pros and cons of these for switching when compared with CMOS would be very cool.

Click to expand...

The big advantage of these optocouplers is you don't have to worry about THD, there's virtually none. The main drawback is they are rather slow, particularly at turn-off. Going from 1Megohm to 1Gigaohm may take 10 seconds. Not a major issue, but it must be taken into account. Also, their ON resistance varies from one to the other, so if you are critical about gain, you have to make sure your circuit is tolerant enough. Their power requirement is also higher than CMOS gates; it may become a factor if ther's a lot of them.

ej_whyte said:

I can see how putting the switch at a summing node (for switching inputs) or in a feedback loop (for switching gain) is advantageous in terms of performance, but I'm struggling to see how you can switch 1 source between 2 outputs, without having to have 2 seperate opamps and switching the gain on each (so the switch changes opamp A from unity to -XdB, and opamp B from -XdB to unity simultaneously)

Click to expand...

Yeah, I'm not picturing any other scenario that would work here. I'm not sure why you'd need to change the gain, though. I'm seeing something like this:

You'd need a CMOS switch with 2 NC/2 NO. Set your gain with R3/R4 for each different type of output. You'll get even better distortion performance from putting S1 and S2 inside the feedback loop. S1 and S4 would be closed at the same time, and S2 and S3 would be closed at the same time.

-Matt

Matt, from your pic you probably want the left side of S2 and S4 being connected to the right side of S1 and S3, else your virtual ground is no longer virtual.
(What's shorter than a short? Two shorts in parallel.)

Harpo said:

Matt, from your pic you probably want the left side of S2 and S4 being connected to the right side of S1 and S3, else your virtual ground is no longer virtual.
(What's shorter than a short? Two shorts in parallel.)

Click to expand...

Neither that nor Matt's arrangement. In Matt's case, the signal is always shorted to ground via a CMOS gate so there will be some output, but highly distorted and noisy.
Harpo, connecting as you suggest shunts the unused output's inverting input to ground, which makes the output extremely noisy.
Matt, you'll have to think it over; it may be as simple as adding one input resistor and then you play with the different configurations. Check attached schemo.
But in fact, considering the performance of CMOS switches today, I wouldn't hesitate to use them in voltage mode.

abbey road d enfer said:

Neither that nor Matt's arrangement. In Matt's case, the signal is always shorted to ground via a CMOS gate so there will be some output, but highly distorted and noisy.

Click to expand...

Erm, well, that was boneheaded, wasn't it? :-[ Of course that won't work at all.

To be honest, I don't think shunting anything to ground will be necessary. Just remove the ground shunts and connect one NO gate to the inverting input of IC1a and one NC gate to the inverting input of IC1b. Turn the whole thing into a SPDT arrangement. The Roff of the gate will form a voltage divider going to the "off" amp, making the level at the that amp very, very tiny. You could use the extra two gates to mute the outputs of the buffers, but that puts them in a voltage mode situation, which even with high performance parts seems to be 1 or 2 orders of magnitude worse in the distortion department.

-Matt

I have lost track of exactly which schematic you are referring to, but there are several considerations to consider when using cheapo CMOS TGs.

They don't support full +/- 18V rails so you want to prevent the body diodes from conducting if audio signal exceeds their rails voltage.

With multiple switches inside one package you don't want a deselected inputs to crosstalk (capacitively) into the other switches.

JR

John, the boneheadedness I was referring to was my own.

There are modern parts designed for rail-to-rail usage, but you're right, the less spendy options will need to have more care taken with the surrounding circuitry.

-Matt

I was not responding to bone headed, I was responding to not back grounding.

Do the expensive parts specify crosstalk between them? I guess you will find out, since it can depend on circuit values et al.

JR

They do provide crosstalk figures between gates, yes.

MAX312/3/4 says -85dB at 1Mhz, better than -96dB at 20kHz.

-Matt

abbey road d enfer said:

Harpo said:

Matt, from your pic you probably want the left side of S2 and S4 being connected to the right side of S1 and S3, else your virtual ground is no longer virtual.
(What's shorter than a short? Two shorts in parallel.)

Click to expand...

Neither that nor Matt's arrangement. In Matt's case, the signal is always shorted to ground via a CMOS gate so there will be some output, but highly distorted and noisy.
Harpo, connecting as you suggest shunts the unused output's inverting input to ground, which makes the output extremely noisy.
Matt, you'll have to think it over; it may be as simple as adding one input resistor and then you play with the different configurations. Check attached schemo.
But in fact, considering the performance of CMOS switches today, I wouldn't hesitate to use them in voltage mode.

Click to expand...

What is the purpose of the 100k to ground in your voltage mode drawing? A bit annoying that it takes twice the amount of opamps, I guess thats just the way it is though  :-\

Hmm I'm still trying to weigh up between relays, CMOS or opto switches at the mo. I guess I need to actually commit each option to paper so i can get a clearer idea of all the support circuit, costs, component count and spec of each option. The application is a rack unit not a console so not a huuuge amount or switches (mainly 4PDT), making me think that current draw from relays wouldn't be too bad, however I'm still pondering their longevity.

Cheers
Elliott

ej_whyte said:

What is the purpose of the 100k to ground in your voltage mode drawing?

Click to expand...

Without the 100k the non-inverting input would have no bias/voltage reference when the CMOS gate to ground is un-connected.

-Matt

Nishmaster said:

They do provide crosstalk figures between gates, yes.

MAX312/3/4 says -85dB at 1Mhz, better than -96dB at 20kHz.

-Matt

Click to expand...

That seems good enough if it is measured as used.

JR

> annoying that it takes twice the amount of opamps

Opamps are 13 cents each. $2 if you get fancy ones.

You have already spent more time-value than that by fretting.

Opamp per output is not extravagant. Series gate per opamp input. Inverter or gain-amp if your gate won't take full swing.

If you like bending-up tiny light-tight boxes, unselected photoresistors with LEDs can be very economical gates with large swing and vast control-side isolation.