mux-friendly opamp

For moderate speed ADC driving, I like the AD8605. Nice beefy output.

Cheers

Phil Hobbs

John Larkin wrote in news: snipped-for-privacy@4ax.com:

It is single output.

Unless what you mean is how friendly its output is it to being MUXed with other signals. Why would it distort upon downstream injection?

IOW what defines MUX unfriendlyness? On the in or out side.

I always thought that the amp would not know how it got fed its complex waveform.

And it (this op amp) has some kind of ripple dampening (less ripple sensitive) feedback. How does that not distort?

With Mux friendly, do you mean how well it behaves when a shift in capacitance is injected into the output?

Cheers

Klaus

Both. This TI part handles input spikes.

What's nasty is that most mux's and multiplexed ADCs act like a charge storage node that's switched between channels, and most opamps freak out for a when when a slug of charge is injected into their outputs. Some ring badly, some just jump and take a while to recover.

All cmos mux's add extra charge injection, from the gate drives to the cmos switches, from

That could be caused by output impedance too high; for an LM324, the usual treatment would be a pulup or pulldown resistor (so the output is never unloaded). There's some multiplexers that have inbuilt load/termination resistors, too, so the solution doesn't have to include offchip components.

TI seems to consider the input; the data sheet doesn't mention charge injection into the opamp output.

Fig 53 of the data sheet is interesting. 10uF caps! Opamp and ADC data sheets seem to want to keep the charge injection hazard a secret.

The angled tee connections are annoying. I hope that doesn't become a trend.

I did include 1K resistor packs between my opamps and my ADC, but that wasn't enough.

I saw the chopper architecture and was too chicken to use it in a current design. Maybe I shouldn't have been.

Choppers sure do a nice job making 1/f go away.

Cheers, James Arthur

Need to make a composite amplifier with a chopamp for DC and LF, and a wideband for RF. The problem will be at the crossover. The wideband will have much less prop delay so there will be a phase glitch near the crossover.

Anyone know how to solve this?

Not knowing what charge injection to expect (but probably negative, because P-channel capacitance during turnon is likely higher than N-channel for a CMOS analog switch), it matters if the resistors are pullup or pulldown. To keep emitter impedance on the output low, you want to ensure a healthy emitter bias current from the NPN emitter in a complementary class A/B follower output stage, to minimize the glitch with a transient sink of current. That'd be a pulldown resistor, to some supply below your signal level.

The double-damned modern datasheet on the OPA2189 doesn't give a clue of the output driver nature; maybe it's CMOS? It could be tested, of course. Feed a square wave through a small capacitance into the output pin, and see if the edge that best approximates a MUX channel turn-on transient is improved by load resistor to +V, -V, or ground.

Series resistance, and a bit of load capacitance, also decouples the op amp output, but utilizing the internal compensation capacitor seems like the cleaner solution.

Lots of "zero drift" opamps do all of that for you.

I'm thinking of chips like the Hittite HMC1022ACHIPS, 48 GHz broadband amplifier. The output is AC coupled and I would like to add an additional amplifier to restore the DC component. See Fig. 47 in

OK, we've thought of that. You need a splitter ahead of the amp pair and a combiner, basically a bias tee, at the output.

I don't think the crossover is terribly difficult, at least in theory/simulation. The crossover could be low, KHz to low MHz. But the details will be difficult at 48 GHz.

Are you going to wire-bond that thing?

You could also maybe DC couple that chip, with appropriate supply offsets and such, and avoid splitting/combining. There would be one offset step to manage somewhere.

I don't see that. The RF path is straight through. The DC path just biases the output.

The SI-LIST talks a lot about launching multi-GHz signals and how to simulate them. There are many experts who can do the work.

No choice. It only comes in chips. However, there are many local suppliers who can do the work.

That would be insane.

They could have just moved C2 a bit to the right.

I can't see how you reconcile

with

Maybe you could post a sketch of what you have in mind.

Correct. Charge injection.

The two fets are cascaded. The lower one sets the current. The upper one is the RF input and is biased at +4V. The output signal is at VCC and is AC coupled to the load.

You could bias the lower one at -4V and the upper one at GND. The output is now at VCC - 4V and would have to be AC coupled. So it would still require DC restoration to recover the DC value at the input.