Temperature measurement

Signal average some. That's what computers are good at.

John

That is what always gets neglected in failed tasks.

The requirements analysis is required to give a proper quote as well.

This is true for database development, circuit development, programming, etc., any time a person's thought process is what is being paid for, the list of final items produced must be fully specified, so the you can get paid for any additional work required when they say "well you had to know that it needed to do that, it's just common sense..."

Yeah right.

Late at night, by candle light, vince penned this immortal opus:

Look up the ICL7106. I think there's a schematic in the datasheet or some of the app notes.

- YD.

un

Integrating A/D converters do it better and cheaper. Successive aprroximation A/D converters can get seriously messed up by a signal that changes appreciably during the conversion process, and averaging the kind of output that you get after such an event doesn't let you accumulate a signal that is accurately representative of the of the average input to the A/D converter.

-- Bill Sloman, Nijmegen

Geez, you're about 20 years behind the times. You can buy a cheap uP that had a good 10 or 12 bit ADC on board, with inherent sample-and-hold. Averaging does improve resolution and reduce noise. We do this all the time, and getting 0.2% accuracy isn't hard.

Integrating ADCs used to be popular because they were relatively cheap. Now the cheapest ADC is an SAR, a tiny spec of silicon on the corner of a uP chip, no external opamps/capacitors/references/switches/counters/logic required.

John

vince:

Take a 1000V battery, a DC/DC step-down converter, a laptop, a USB thermometer, and write a Visual Basic program that uses a table to get to the required accuracy.

But it isn't a ratio sensor, so that requires an accurate voltage reference (not an insignificant cost addition, either in pennies or milliamps).

d Run

or

The sample and hold is the key, but if you are using the rail as the reference for the A/D converter, you need a sample and hold on that as well. I'm actually more than twenty years behind the times on this - I got read the lesson on successive approximation A/D converters a little more than 40 years ago and I have kept it in mind ever since, along with a lot of other potential gotchas.

It helps if you've got enough noise on the signal to dither the A/D converter output across the least signficant bit - the RMS value of the noise needs to be at least one third of the LSB, otherwise averaging won't necessarily improve resolution.

But you do need the sample and hold (which should come on an adjacent tiny spec of silicon on the same chip), and almost certainly some kind of external filter. The nice 24-bit (actually 20-bit) sigma-delta A/D converters do integrate (or a least have a built-in digital filter) but even they do need a bit of filtering to keep out noise at around the sampling frequency. Happily this is a lot less bulky and expensive than the filters to block the 50/60Hz noise which can be a nuisance with faster A/D converters.

-- Bill Sloman, Nijmegen

Why? The conversion is over in a few microseconds, and the rail won't change much in that time. If it does, so much the better: it with dither out the quantization. For a temperature measurenment, you can afford to average 10,000 samples maybe.

I'm actually more than twenty years behind the times on this - I

Most ADCs on uP chips have a couple of bits of noise. If not, add some! Waggle a port pin pseudo-randomly; couple into the thermistor node with a big, many megs, resistor; bypass with a cap. I've used the program itself as a lookup table of pseudo-random noise, but a real PN shift register is easy in software, too.

Capacitive SAR ADCs have inherent sample-and-hold.

John

Mmm... be sure you do this at a rate that's not at all synchronous with your ADC sample clock, though?

As far as I'm aware, the whole idea of twiddling the LSB is that it forces transitions from some low-level input signal and you thereby essentially end up with a PWM-type output on the lowest bits which (overall) average to what the analog input was... so long as the "noise average" was zero (and presumably also had some "reasonable" distribution -- white, Guassian, triangular, etc.). But if your noise is strongly correlated to, e.g., the ADC sample clock itself, that noise average is probably not zero anymore and you no longer recognize the benefit...

---Joel

In the IR thermometry industry, we used a 'chopper' wheel to physically switch between ambient (instrument ambient) and the target.

This is similar to a chopper effect.

The microcontroller ADCs normally have separate reference inputs, so you can connect them, and your thermistor, to a low pass filtered version of the supply. That way you can make the variation in reference voltage during the ADC cycle arbitrarily small withot needing a dedicated "reference".

Of course a separate external delta-sigma ADC will give even better results - a few ppm - but is probably overkill for this application.

Spehros suggestion of a RC oscillator calibrated with a switched-in precision resistor is very interesting too. I worry about non-linearity caused by the "switching time". I wonder is there a particularly good circuit for this?

of a

rees

.and Run

istor

rive

ood

NTC

s.

for

c.

of

If it does - and you do get switching spikes on power rails - you screw up the successive approxinmation process in exactly the same way as changing the signal does during the conversion process. If the comparator puts the signal on the wrong side of a major transistion early in the successive approximation process, the result is going to be wrong, whether the signal or the reference is the source f the error.

Not if your errors aren't normally distributed. You don't seem to know enough about dither.

Did you know that there was an optimal - triangular - amplitude versus probability distribution for dither signals?

But they still need input filters.

-- Bill Sloman, Nijmegen

es

end

what

ly

Not really. The chopper wheel lets you look at two different sources with the same detector, and the output signal that you get from that detector contains its useful information as an alternating voltage at the frequency and phase set by your chopper wheel.You extract that information with a phase sensitive detector.

Dither applied to the input of an A/D converter adds noise to the output, but you can reduce that noise by averaging successive outputs

- effectively throwing away the dither that you added. One variation adds a carefully determined analog dither signal to the input to the A/ D converter and subtracts the digital value of the added signal from the A/D output after conversion, which is actively throwing away the dither signal.

-- Bill Sloman, Nijmegen

It's a cheap technique, until you figure in the cost of calibrating the sensors. Physicists are notoriously bad at this - they figure that everything needs to be calibrated, so they don't see it as a cost that can be evaded by buying close tolerance parts.

-- Bill Sloman, Nijmegen

-- Bill Sloman, Nijmegen

"John Devereux" kirjoitti viestissä: snipped-for-privacy@devereux.me.uk...

Somehow I missed Spehro's suggestion. Even though I backed out all the references, I can't find it.

Can you post a message-ID? Thanks! ...Jim Thompson

It does't generally have to be very good with a thermistor.. 0.5% of resistance is around 0.1°C resolution, which is enough for most practical purposes (granted the OP's accuracy spec is not very clear, but if we assume it's resistance it makes some sense). Measuring human skin temperature to 0.1 degree using a typical thermistor is not going happen anyway, unless make a fairly deep incision first.

Armpit?

Where is your "RC oscillator" post? I can't find it. ...Jim Thompson