Temperature Measurement

I am now certain that they don't know what they are doing.

They might perhaps have to use a function library function that only has a double precision interface I suppose. NAGLIB was one such.

Unless they are working on mass spectrometry, seismology, big science, precision timekeeping/lunar ranging they will not come close to having sensor data that exceeds the 24bit precision of a 32 bit real mantissa.

There are still only a handful of ADCs that can originate true 24 bit data these days but I can't see them ever being used in motor control. Most times Johnson noise dithers the lsb anyway.

What are the best ones?

The ones I know about (not too many, admittedly) top out at around 19 bits before noise takes over.

Cheers

Phil Hobbs

I heard it was quite good, but haven't seen it myself as yet. IMHO films or TV series based on books don't have to match the books to be good - in fact, generally they should /not/ match them too closely. Different media are good for different things. I'm sure DA would have loved to have Dirk Gently turned into a TV series.

Eh? The Dirk Gently books (there are two) were written about 20 years after the original Hitchhikers'. They were not "early work".

Wow, Winterpeg. Great place to focus single-mindedly on your work--lousy weather and no distractions. ;)

Yeah, fast sensors and actuators (TECs or very small heaters) help a lot--the extra loop bandwidth really helps with the forcing rejection at frequencies you actually care about. A fast proportional-only loop with a local sensor mounted on the heater/cooler can really improve loop performance. (Our former regular Tim Wescott used to talk about that sort of thing a fair amount.)

Not counting labour!

Nice.

Many years ago, when I worked for Microtel in Vancouver, I used to get 5 ppm crystals from C. R. Snelgrove in Toronto. It was a great outfit--they'd ship custom-ground crystals for any frequency you wanted, with +-2 ppm initial and +-5 ppm over temperature. IIRC they were mostly AT-cut, so basically limited to 1-25 MHz or thereabouts, but that was all I cared about. Cheers

Phil Hobbs

There are instrument ADCs that are way better than 19 bits, but they may be too slow for motor control. And too expensive.

Look at Texas Instruments ADS1262 and ADS1255, for example.

Joe Gwinn

I was always amazed by dot-matrix thermal printers that heated dots on sensitive paper in milliseconds. TI was big in that.

Here's my OCXO, lower right

It used a mosfet as the heater. The two bigger blocks are current controllers for NMR, using big Minco kapton heaters.

The one with the visible shunt was my first heated block; it regulated the currents to a bunch of electromagnets in a big particle accelerator.

This was the ocxo xtal, vacuum sealed with the turning point temp marked on each, from Lap-Tech in Canada.

Nowadays, I just buy OCXOs. They are a lot better and cheaper than they used to be.

oned

't

al

riving

th a 10

e that

rs

ure.

ot

I agree with all that! One of my last heaters was a TO220 FET with TO220 diode sensor bolted to the top. I set the temp of the FET and the apparatus had to follow. :^)

George H.

We make digital delay generators that go out to 2000 seconds, with 1 ps resolution. We store time as a 64-bit integer, lsb 1 picosecond.

I think you post makes many assumptions and ignores the realities of computations on a CPU. It all depends on the calculations and I don't know what they are. You keep talking about the ADC when that is not the issue.

'Nuff said. This is one of those conversations where you just keep repeating that you know what you are talking about even though you don't know what anyone else is talking about. So enjoy.

Depends on what you are measuring. Don't they use more accurate ADC in scales?

Ok, then he regressed and often used less interesting forms of the same humor as in the Hitchhiker's trilogy.

Interesting, thanks. I'm fairly shocked to see TI using the wrong definition for ENOB, though--they omit the factor of 1/sqrt(12).

Cheers

Phil Hobbs

I have data showing ~30 ps rise times in a bolometer. (It was about 100 nm wide.) Thermal diffusion gets faster quadratically as the device gets smaller.

I used to have a few of those (not sure of the vendor).

Progress is a wonderful thing, sometimes.

Cheers

Phil Hobbs

Well, I certainly think THGTTG series was better and funnier - so we can agree on that at least.

We should all be aware of the subtle importance of the number 42 - what engineer has not been asked to give an answer by someone who clearly does not understand the question they are asking?

And appreciating the meaning of an SEP cuts stress dramatically.

But Dirk Gently is not without its charms. We have all met problems like the sofa on his apartment stairs, and like his fridge.

TI certainly knows the difference. I'd guess that they have observed that the measured errors have a gaussian distribution, so just use that somehow. ADC errors are not just quantization errors, so this may be the easiest approach.

I'd hazard that there is a precise definition of what they mean in their app note on characterizing ADCs, which I've seen, but don't recall the number or title. Typically, the app note will be listed under relevant app notes on their web page for the device.

Joe Gwinn

There are lots of 1-Wire knockoffs available. {not recommending same...}