CMOS switches / distortion, noise

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

baadc0de

Well-known member
Joined
Jul 9, 2009
Messages
797
Location
Europe
Hi,

If I were to route audio through high-voltage version 4066 CMOSs (20V for example), say, four of them cascaded, would there be distortion and noise buildup? What if I did that to 24 channels of audio, still acceptable or negligible?

Cheers,
B.
 
The datasheet specifies the distortion as 0.12% at 10kHz. Of course it depends on the circuit you are going to use. With 4 analog keys in line the distortion will increase. Can't you do it with passive mini-relays? Or minimize the switching matrix to only 2 keys in line?

Mark
 
Douglas Self wrote some interesting stuff about this topic in his books "Self on Audio"and "Small Signal Audio Design".
Not only distortion is a problem with multiplexers. Control voltage feedthrough is much more ugly and needs a lot of effort to keep it out of your audiopath.
In a conversation with a console designer he told me that the usual 4053 multiplexer is not even good enough for audio. There was a Philips version of the 4053 which had less CV feedthrough which he was using in his designs. Same chip from other manufacturers didn´t meet his specs.
 
Calrec gets away with using 4053's in large amounts without compromizing signal integrity a lot - they switch only into inverting inputs of opamps (virtual ground). That works quite nice imo.

example: http://www.gyraf.dk/schematics/Calrec_UA8000_mic_input.gif
and: http://img.photobucket.com/albums/v172/gyraf/Calrec_AC1521.gif

(sorry about the poor image quality)

Jakob E.
 
gyraf said:
Calrec gets away with using 4053's in large amounts without compromizing signal integrity a lot - they switch only into inverting inputs of opamps (virtual ground). That works quite nice imo.

example: http://www.gyraf.dk/schematics/Calrec_UA8000_mic_input.gif
and: http://img.photobucket.com/albums/v172/gyraf/Calrec_AC1521.gif

(sorry about the poor image quality)

Jakob E.

Yup, that was the new idea.. and switching into ground via resistance.
 
;D  I actually used CMOS transfer gates (4016/4066) in a console to switch audio paths electronically back in the late '70s-early '80s.

If you look at how they work, the distortion mechanism is based on the TG's on resistance changing with terminal voltage, divided by the load impedance. I came up with a way to not only switch the audio in series with an opamp inverting input so the terminal voltage across the TG was almost nothing, but I also located the TGs inside the opamp feedback loop, so the current through the TG was also near zero. Even if there was distortion caused by the TG with almost no terminal voltage driving the very easy load of an opamp input, that distortion would be reduced by the loop gain margin of the opamp. Used this way, I was not able to detect any distortion from the switches, using the best bench equipment typical of the day (Sound Technology).

IIRC I ended up using a pair of TG per audio line I switched, and instead of just leaving a off input open circuit, I shunted the off input to ground to reduce any possible crosstalk. In fact the voltage and current inside the TGs is so small that the PCB layout is more influential to affect crosstalk than voltage crosstalk inside the TG packages. Another benefit of shunting the unused inputs to ground is it provides additional protection against input signals larger the +/- V/2 causing CMOS input diodes to conduct.   

It was a minor PIA to come up with +/- V/2 rails just for the CMOS, but in the context of a console with 16x or 24x lines being switched it was a minimal expense. I also had to dial in timing and other tweaks to not make clicks while switching, but it worked well and was very clean.

JR

PS: It seems we have discussed this here also years ago... IIRC Brad also mentioned using TG in products to switch audio.
 
I always try to spell out acronyms at least once when I am using something uncommon, and I did for this. TG means "transfer gate" and is their name for 4016/4066 CMOS switches, which inside are simple P and N channel MOSFETs in parallel, with their gates alternately driven high or low for on/off as appropriate for their polarity. The distortion increases as signal swings closer to the + and - supply as one polarity MOSFET gets turned more off and the other harder on, since these changes in conduction are not symmetrical. The ideal case for low distortion is to keep the terminal voltage of gates at a constant voltage, which is literally impossible, but clever design allows us to keep terminal voltage changes small.

  JR
 
There are much better CMOS switchers/multiplexers than the 40xx.
I've used a lot of DG308/411 with success, the latter being much better than the former. They are at least an order of magnitude better than the venerable 40xx.
You must remember that distortion across CMOS depends on the voltage across them; you must make sure that the current is the lowest possible.
Receipe: low impedance source (typical opamp output), very high input impedance of the receiver (typical non-inverting input of an opamp), make sure that the input of the receiver is never left unloaded (an xPST should not exist, only xPDT).
 
jensenmann said:
Douglas Self wrote some interesting stuff about this topic in his books "Self on Audio"and "Small Signal Audio Design".

These are interesting articles indeed, sure worth reading.
Or check the EW+WW magazine versions if you don't have the books (vv).

IIRC he came up with a pretty blameless result (to use 'Self-speak'), be it at the expense of component-count: opamps with JFETs & timing-caps.

But as said already in this thread, the 40XX-ICs can be put to decent use as well if you need a high but not the highest level of performance.
 
Not to be repetitious, or argumentative, but the way I used the CMOS TG inside the opamp feedback loop I was unable to measure (with Sound Technology analyzer) any nonlinearity visible above that of the unity gain inverting TL074 opamp (actual noise gain was more like 2-4x but still quite clean). This was several decades ago, so I was probably bumping into the analyzer residual and/or opamp noise floor at that.

This audio path was so far superior to professional tape media, and typical consumer playback of the day, that I would characterize it as very high performance (I was using this inside a recording console after all..).

If I had to revisit this today, the opamps and distortion analyzers would be better, so I might be able to measure something, but with modern digital technology why bother with the complexity?

JR

PS:  I probably should have been reading EW+WW, and D Self books, so I wouldn't have to come up with this stuff on my own.  :eek:
 
Hi John,

Hope it wasn't my previous posting that triggered you  ;) since I well believe that even the 4016 (with its highest RDS_ON) can be made invisible for all relevant situations.

I recall that the Self EW+WW articles started with CMOS & then went on for higher level of performance + complexity.

I don't recall the exact reason offhand though, might have been addressing mainly the clickless switching side of things.


And then there's this... (attached)  :)
 

Attachments

  • BJT.pdf
    85.5 KB · Views: 81
Ewww  you tricked me into looking at a Behringer schematic...  :-[

I was forced to do some clean up design on a 4TR cassette deck and that used some crude bipolar mutes and switching. FWIW there are actually transistors designed and optimized for such applications (like using transistor saturated switches upside down for lower saturation voltage).

Yes i was reacting to your comment that 4016/66 could not deliver "the highest level of performance", while all this analog switching is mostly invisible to modern designers, if we ignore A/D convertors with multiplexed inputs, DPOTs, etc.

JR
 
Gents,

Again, forgive me - I'm still rolling sleeves up with discrete circuit design. Just looking at the Behringer schem...

Wouldn't you get a 0.04VDC voltage drop across those BJT's when they were saturated?

Looks like it's performing a gain switch? (switching a resistor into circuit).

Finally, wouldn't that voltage drop cause problems?

/r
 
JohnRoberts said:
PS: It seems we have discussed this here also years ago...

http://www.groupdiy.com/index.php?topic=24460
http://www.groupdiy.com/index.php?topic=22233.0

(sadly some of the interesting stuff was in Wayne's posts)

abbey road d enfer said:
I've used a lot of DG308/411 with success, the latter being much better than the former. They are at least an order of magnitude better than the venerable 40xx.

They're also an order of magnitude more expensive, so that about evens out.

JDB.
[but that's less of a big deal for DIY]
 
Just an FYI:

Most of the customers I deal with in home audio systems (Japanese, Korean, Chinese) all use CD4052's as ADC input mux's.
However, they run them at max voltage with lower voltage signals, so that the signal doesn't approach the gate voltage.

Given that some of them are happy with a system SNR of 70dB, maybe they don't worry too much about 0.000000% THD.. :)
 
Rochey said:
Gents,

Again, forgive me - I'm still rolling sleeves up with discrete circuit design. Just looking at the Behringer schem...

Wouldn't you get a 0.04VDC voltage drop across those BJT's when they were saturated?
Perhaps roll them up a little farther...

Seriously how do you come up with that 40 mV saturation voltage? It depends on several factors, but yes it will not be 0V.
Looks like it's performing a gain switch? (switching a resistor into circuit).
yes, shunting across the R
Finally, wouldn't that voltage drop cause problems?

/r

Yes, among other things...  

For one, the CM rejection of that circuit is only roughly in balance at low gain, at high gain CMR goes to hell.

But looking at the brand name on the schematic explains a bunch...

Now please stop making me look at this poop.... (I will play along some as this is educational, but mostly in a don't do this...  way.)

JR
 
Back
Top