Multiplexer sample and hold

Or there's those floating-gate voltage references, which are sampled at manufacture time and that's it. ;-)

Tim

Delta-sigmas don't need a s/h, and most SAR ADCs these days are capacitive ladder designs, which have inherent sample/hold.

Some communications ADCs have s/h bandwiths far above their sample rate. That can be useful.

Capacitive ladder? You mean C...C/2 instead of R...2R?

My knowledge is 30 years out of date. I never heard of a capacitive ladder back then.

So they interleave multiple s/h and ADC channels fed from one signal?

I'd be interested in what you finally decide, and why.

If you can use the capacitors barefoot rather than having to buffer them, it's often a win. (I'm thinking about nulling offset voltages on CMOS or JFET parts, for instance.)

Some recent-vintage low-voltage muxes have really low charge injection, like 1 pC, so that you could use a 4-cent 33-nF NP0 cap up to like 16 bits.

Cheers

Phil Hobbs

Also, you do not need 8 opamps, just one fast opamp, and switching of the feedback nodes also

Cheers

Klaus

It's a tad discomfiting, though, that a mux limits the output voltage range you can achieve. I'd really like the full common mode range of a +/- 12V op amp available, if that's possible. The op amp buffer can apply gain and offset where the range is important.

Like an LF398? ;)

Cheers

Phil Hobbs

I see that they crow about +/- 3%..there are other makers that does

Well, it's too expensive, and it hasn't the ability to add an offset and gain, and... it makes the octal DAC look good. My bad, I shouldn't ask for a magic bullet without stringent specs for ballistic accuracy.

A tube of 4051's I can buy and put on the shelf until I need one. Ditto TL061 or TL062. Not so for the LF398 (nor any other track/hold); they're not really affordable.

The LF398 never really made it to jellybean status, and today it's far inferior to things you can cobble together for pennies.

Agreed. OTOH the DG2042 or ISL8454x are a bit on the pricy side too ($2).

CHeers

Phil Hobbs

I was scoping the usual supplier sites, and octal 8 bit SPI DACs are pretty cheap, around $2 something in quantity (Maxim tho, and JL says friends don't let friends use Maxim.) I think from a board space and parts perspective it's going to be hard to beat just using two of those vs. some multiplexer ararrangement.

I really would like 10 bits, and when you start going up in bit depth the cost starts jumping up quite a bit to the point it makes you wonder again if you should multiplex.

But, I don't really need high accuracy, just higher resolution to limit "stepping" effects when values are rapidly changed between significantly different levels.

And well well look at this

But are you sure that guy knows what he's talking about?

If you do that, make sure that you have the processor bandwidth to support it, and that you can sample the DAC updates out at an even sampling rate, and that the bandwidth of the stage following the DAC is low enough that you get the average out, instead of an output that's madly jumping around.

If the size of the step is acceptable but the suddenness isn't, and if the following stages are high impedance, then just put an RC filter after each sensitive DAC output. That'll prevent pops, while still being a low- component count solution.

Mostly I prefer analog switch chips with internal circuitry more like the 4016 than the 4066 style, because driving the back-gates (wells) under the transistors *using charge stolen from the signal pins* just to get lower ron seems like a bad deal in most of my applications. This is especially true if the manufacturer didn't correctly implement break-before-make on the back-gate switching so that the back-gate is disconnected from the signal pin *before* it is connected to the supply rail, and vice versa.

Of course if the control input of the switch is just supplied with a constant logic level rather than being repeatedly switched, then it doesn't matter, but that is rarely the case for me. Most people seem to just look at the chip with the lowest on-resistance and think "ooh-shiny", even though their circuit may not care about on resistance as much as signal-dependent charge injection (e.g. S/H, or many mixers).

I'm not sure whether you can get the 4016-style (with back-gates just tied to the supply rails) in a MUX configuration like 4051 etc.

Chris

Not sure how much that sort of consideration means these days, because even apparently simple parts have a bazillion transistors inside. Studying the shape of the charge injection vs voltage curve is a good idea, for sure, but since the ON resistance of a modern switch is usually a good two orders of magnitude better than a CD4016, with 1-2 pC of charge injection, you can usually just jack up the hold capacitor value a bit and have a big win.

Cheers

Phil Hobbs

I know this is a bit old post, but I am researching this topic as well, and would like to add a little idea which is also doable and relatively cheap. Its about the R-2R topology dacs. For multichannel application dtype octal transparrent latches could be used with the benefit of having parallel inp ut (fast speed), where the application microcontroller IO count permits. Th e latches can be paralleled to any amount necesary, as well as the bit widt h. Otherwise serial to parallel shift registers could do the same, driven b y fast SPI.

Another idea which I saw somewhere was implementing multichannel PWM with s erial to parallel shift registers. Driven fast enough, with proper pwm chan nel data encoding this also could work.

Just my 2 cents. Greetings to all!