Sample & Hold, did they go out with the buggy whip?

ISTM that dedicated S&Hs went away when jellybean muxes got really good. 1 pC used to be really amazing, and it's now fairly run-of-the-mill.

Cheers

Phil Hobbs

I wouldn't think it practical to make a s/h with anything like 22 bit accuracy. Well, maybe 18 bits with expensive parts and really slow acquisition time.

Can you just power up the sensor+ADC, measure, and shut down?

People don't make s/h chips much any more. Actually, never did.

You can't leave it on long enough for the number crunch? No S/H needed. (Is it a _really_slow_ ADC?) ...Jim Thompson

True but it means that nowadays one must use a mux plus a cap plus a buffer amp while in the olden days a single IC was enough. I guess it's a pill I have to swallow, meaning some super tiny SON package to squeeze it all in. What I dread most is making the CAD models for those because there isn't much of a standard.

I really only remember the LF398, HA5351, and AD684. I don't think there ever were very many. Between improved MUXes and sampling ADCs, they sort of got squeezed out in the middle.

Cheers

Phil Hobbs

??? Making a CAD model for a S/H should be downright trivial... at least for me... seems I have a building block for every occasion >:-} ...Jim Thompson

Den torsdag den 13. august 2015 kl. 23.37.21 UTC+2 skrev Joerg:

how long do you intend to power the sensor?

-Lasse

It's the layout footprint that is sometimes a lot of work. Nothing can be copied from anywhere, all from scratch, everything must be down to a tenth of a millimeter, radiuses, and whatnot.

That's what I asked. If the ADC is fast enough, who needs a S/H? ...Jim Thompson

No, because low cost ADCs >20 bits are generally slow delta sigma converters that need the input stable the whole time and spew out an incredible amount of spikes at the input.

Oh, there are some good ones. But all in the twodigit Dollar range and they need half a nuclear power station to run.

Can't. It tops out at 60 samples per second and we need almost that many.

Yup, it's slow. Because of power consumption requirements.

As short as possible. It is a 1k strain gauge and ideally I'd like to push that to the low single-digit milliseconds. If I can achieve a 5% duty cycle that would cut the power consumption down by almost 95%.

Realistically we need around 20 bits of precision from a 22-bis delta sigma converter running at 60 samples/sec.

Is the sensor anything like 20 bits accurate? Actually, some load cells are.

Run it at lower voltage! Save v-squared power.

It is a strain gauge and WRT dynamic loads it is more than 20 bits accurate but we can't squeeze more than 20 bits ENOB out of the converter. DC accuracy is the pits but we don't need much there.

Only if it's a switcher driving it. Could be done but we've already tried it with less than 1V and then there is a lot of in-band noise creeping up on us.

Next thing that you will be claiming is that there is no free lunch.

My experience with load cells is that they have enough vibration noise that you *have* to average them for a while. So a s/h may not work.

I can envision a power saving scheme, but it's awful.

We now use these pan balances (from china) that do 1 mg out of

George H.

I assume by "need half a nuclear" plant you mean they draw as much power as your sensor?