Analog switch - glitch-

[Jester]

The circuit below is a squelch circuit, the analog switch is open when there is no audio and closed otherwise. The circuit works except for the nasty little spike that occurs when the switch is closed. I measured the voltage at LOUT with the switch open and it's close to ground (< 2mV). Not shown on diagram there is a weak pull down 100k at the o/p of the analog switch

It's not obvious to me what's causing the glitch.

Any ideas as to the cause and possible remedy?

 

[erikl]

Perhaps charge injection from the switch. Can you afford higher load capacitance? Try if this helps.

Or try compensation: Use the 2^nd switch inversely controlled , i.e. open it when the other one closes and vice versa. Input to GND (perhaps via 100kΩ, too), output connected to LOUT.

 

[FvM]

The observed charge injection is described as crosstalk between digital input and switch in the datasheet. Your circuit is generating maximal crosstalk because you bias the switch DC level near to ground instead of half supply. All crosstalk specification are referring to the advantageous symmetrical bias situation, so we can just say that crosstalk will be higher without a quantitative specification. Besides finding a better bias, you'll possibly want to use an analog switch with lower charge injection.

 

[dick_freebird]

There is always, except perhaps at one common-mode-point,
an imbalance between the gate charge (which includes Cgs
and Cgd) of the N and P switches, for a net charge bump on
any switching event. Nastiest is the turnoff, where you take
away one of the dissipation paths for that charge.

Some specialty switches advertise low charge injection
and anti-pop type design. Slowing the gate edge helps
at least with the perception and high frequency content.
Attaching shunt resistors (and maybe capacitance, so
long as normal frequency response isn't compromised)
can be a band-aid. If you feel a need to roll your own
you could, perhaps, use some of the ALD discretes and
a gate drive of your own construction (e.g. a controlled-
ramp op amp drive, or slow RC, or ...).

You could also consider instead of the simple shunts,
some sort of (say) FET shunt which blanks the audio
path for the duration of the switching event.

 

[Jester]

Lots of good help, thank you.

The part I originally selected is a NXP part that specifies cross talk at -56dB, and charge injection of 7.5pC

Fairchild FSA266 looks better with crosstalk of -70dB and charge injection of 5.5pC
TI TS5A2066 shows crosstalk of -66dB and charge injection of only 1pC (although they show o/p biased to VCC/2)

Worst case glitch I seem to get with the NXP part is -400mV, so if I assume Cl=18pF (7.5pC/0.4) = 18pF

Then the Fairchild part should reduce the spike to 5.5/18 = 300mV and the TI part would give 55mV if the 1pC is valid with the bias I have?

Which of these 2 parts would you try?

https://www.digikey.ca/product-detail/en/FSA266K8X/FSA266K8XCT-ND/3042800

https://www.digikey.ca/product-detail/en/TS5A2066DCUR/296-17877-1-ND/754593

 

[crutschow]

I would try the one with the one with the lowest charge injection.
Also why not add a small capacitor to ground at the output to further reduce the spike?

 

[erikl]

Also why not add a small capacitor to ground at the output to further reduce the spike?

Can you afford higher load capacitance?

If not, you could transfer the low pass filter R3-C5 to the output of the switch (I'd suggest: R3 before, C5 after the switch).

 

[Jester]

I would try the one with the one with the lowest charge injection.
Also why not add a small capacitor to ground at the output to further reduce the spike?

I will give the TI part a try.

I did try a 100pF as well as 470pFcapacitor and I did not observe any change, perhaps because I was connected to the o/p side of the switch?

 

[erikl]

... perhaps because I was connected to the o/p side of the switch?

Shouldn't matter when the switch is closed.

 

[KlausST]

Hi,

Surely there is charge injection in every cmos analog switch. But not that magnitude.

One thing is that the signal just before the switch is bypassed with a C and a R with reference to GND.
Therefore at the switch input the analog signal moves around GND. Positive and negative.
What analog input voltage range? Does it include negative voltage?

Another thing is that with the switch open, then the output is floating.
Maybe another 100k to GND at the switch ouput could help?
A 10nF in parallel is another option.

Klaus