tmp36 Analog devices temperature sensor

"merco" wrote

The short answer is probably neither are "right" depending upon your definition of right. However, according to the manufacturers, both are probably within specs. What are you using to measure the output of the sensor? The sensor's stated accuracy is + or - 2C. That means that the true temperature could be anywhere from 14C to 18C and the sensor would be within stated specifications.

Further complications arise when measuring the output of the sensor. The device you are using to measure it is not exactly "right" either. It has an allowed amount of variation between devices called tolerance, meaning that the stated reading is not necessarily the "right" value.

You didn't say what kind of device the other thermometer is. We really need to know that information to fully answer your question.

Well the maths looks right.

V = 500 + 10*25 = 750 mV

Depends on what the temperature is of course. With two thermometers reading different values, you need to ascertain which is correct. Nobody on the web can tell you.

Make sure the one reading 19.2 deg C is not experiencing self-heating. A thermistor will if you are not careful.

Now, i'm using a digital multimeter to measure the output, but i'll use a PIC with analog ref to 2048mV...

So i must do an approximation ?

You need calibrated temperature equipment or PT100 temperature sensor to determine if the TMP64 provides the right value...you have to calculate worst case tolerance value rather than typical (not just the sensor but whole circuit).

I would suggest to use REF191 as reference rather than PIC internal reference, as it more accurate. Consider quick calculation:- you have

10bit resolution ADC. In worst case you have to drop 2 bit LSB due to non-linearity, etc, this leave 8 bit accuarcy or 4V/256 = 15.6mV. or 1.56degC (assuming 10mV = 1degC). The resolution is 4V/1024 = 3.9mV or 0.39degC.

You can improve accuracy by setting to 0degC or 20degC (next to PT100), measure the readout and then calculate the error between calibrated temp sensor and the actual PIC readout. use the software to subtract the difference and the accuarcy is much improved. This process is calibration and take place after board assembly.

use

Careful, the minimum Vref+ as stated by the datasheet is 2.5V for the etxtended temp (-40C to 85C) range and 2.2V for the industrial temp (0C to 85C) range. That being said, I have successfully used a Vref of

1.024V on a PIC 16F88 with full 10 bit resolution (1mV/count) at normal house interior temperatures. It works fine, but I wouldn't recommend doing that to a production project. I was able to easily acheive .1F resolution reading an LM34.

I don't understand. But since your looking at ADC readings that will mean .2C per ADC count this will be your maximum resolution. You could extrapolate that to .1C taking many samples and averaging the reading.

You may improve resolution using averaging technique which remove the noise. This is done by takeing many sample (say 4 sample) and then divide the result by 4. It depend if you willing to use 16 bits interger which may slow down the PIC since it 8 or 10 bits processor. However this will not resolve non-linearity of the ADC.

Some of the Analog Devices parts have a serial port with direct digital readout. This might make the measurement task easier for your microcontroller.

Chris

i don't understand the vref+ problem that said Anthony...

So, i can't use the 2.048V reference with 16f88?