DS18x20 hi accuracy

I might have dreamt this , but thought I’d seen a hi precision version of this sensor , which had a better than the +- 0.5C accuracy . I have googled this without success . As a last try has anyone else heard of this ?

You can do that well or better with any of the DS18xxx series, by calibrating each sensor individually, in which case its accuracy is limited by the inherent resolution, internal noise and nonlinearity. You just need an accurate standard against which to calibrate.

The manufacturer's specifications are given for "out of the box" sensors and are statistical in nature.

Yes , I realise that, thought there were some , say +- 0.25 C out of the box . Was looking use one as a reference itself , haven’t anything to calibrate against .

If you have several of those sensors, simply make sure that they are all at the same temperature (tape them together or something) and average the readings.

As with any measurement, that average is more likely to be closer to the "true" temperature than any of the individual readings.

The accuracy goes as the square root of the number of measurements. So if each sensor is specified to be within +/- 0.5 degrees of the correct value, the average of four sensors will be within +/- 0.25 degrees of the correct value.

Unless there is an inherent bias on the sensors - ie they all read towards the + 0.5C say , or they all drift in the same direction, then the average is not better

Sounds like you have talked yourself out of this project.

But for what it is worth, the manufacturer has done the "sensor averaging experiment" for you and was kind enough to present the results in the data sheet.

I think that means that 99.7% of all sensors will fit beteeen the +3s and -3s lines.

It doesn't say that the error on any individual sensor will even follow that same curve. But it would be reasonable to think that an individual sensor will likely be pretty close to the center line.

You can also calibrate against some reference like melting ice or boiling water. Find the calculations to determine these temperatures for your altitude.

I have no doubt that the "mean error" line is the error in an average taken over some number of randomly selected sensors, and is well within +/- 0.2 degrees from the reference values, over the range of 0-70 degrees.

And yes, the +/- 3 sigma line indicates that that a buyer would be rather unlucky to obtain a sensor that was even off by 0.5 degrees, approximately 3 times out of 1000. These sensors work amazingly well.

Yes I think I agree with your argument - I’ve a few I can experiment with , the ice and boiling point test is worth is punt too .
I was hoping for a “mil” spec 1820, but as there doesn’t seem to be one I need to look into trying a few together and see how it looks as per your thoughts - some basic science lol.


In case anyone interested:

Did a quick experiment this afternoon with 3 DS18B20, two from one recent batch , a third I've had for a while.
I compared these against a digital thermometer (new) stated as +/-0.3C and a Dow Corning lab type Mercury in glass thermometer (-5 to +50C).

All were mounted in drilled holes in the same aluminium block, allowed to stablise, DS resolutions set to 11bit.

Batch A first sensor 20.25C
Batch A second sensor 20.25/20.37C

Batch B sensor 20.12C

The resolution setting, I think, means all three are within +/- 1 step/increment of each other.

Average of A,A&B = 20.22C

Mercury in glass : 20.5C. (error not known)
Digital thermometer : 20.5C. +/-0.3C

The characteristic curve of the DS1820 suggests they tend to read low around 20C ; so within errors they are agreeing well.

I'm waiting for some distilled water to do a (pressure corrected) boiling pt and freezing point test ;which should be and accurate check for some/all of them. Might (ought to) repeat set at 12bit resolution too, if I get the time.

I have noticed a temp difference if you read the sensors in different resolutions.
Better always stick to the default 12-bit.

There is ofcourse also the problem of averaging, since the resolution of these sensors in 12-bit mode is 1/8 degree C and humans think in 0.1 or 0.01C.