Re: newbie question about bit depth



On 15 ene, 21:57, "deluded.s...@xxxxxxxxx" <deluded.s...@xxxxxxxxx>
wrote:
Hi everyone,

This is a very newbie question from an equally newbie electronic
guy...

I have a requirement where I need to measure something to accuracy of
0.5 temperature units. How many bits would be needed for digitization
of data? What information do I need to be able to calculate the bit
depth of the acquisition card?

Cheers,
D

Hello,

Looking into your perfil, you are a math/software person. Before you
can decide on the A/D converter or data acquisition card, you need to
know the characteristics of your sensor and electronics.

For this you need to know the lowest and higher voltage that your
circuit / sensor combination will generate and the lowest dV/dT if the
sensor / electronics combination isn't linear (as is the case with,
for example, NTC type sensors).

For example: your sensor outputs minimum 1V and maximum 4V, and dV/dT
is minimum 20mV/degree. This means you must be able to detect a 10mV
step. Assuming a 0 to 5V input ADC, the amount of codewords required
must be > 5/0.005 = 1000. That requires at least 10 bits. Maybe you
want more resolution (for example 0.1 degree readout), this also
requires more bits (13 bits).

But there is more. Your sensor has inaccuracy; your ADC will introduce
errors (like offset, gain error, non-linear error, etc). Also you
conversion formula from resistance/voltage/current to temperature will
introduce some error. As a result, a 10 bit ADC from the data***
will not be sufficient.

Probably you would like to calibrate the sensor / ADC combination.
During calibration, you measure also with the system to be calibrated;
this introduces an error also. Your standard (controlled temperature
bath) will also have some error.

So guaranteeing a (worst case) accuracy of 0.5 degree requires a full
analysis of all components. You can be lucky. Probably the speed is no
issue. There is a reasonable choice in AD conversion products with 20
bits or more, with low offset, gain error en extremely good
linearity. You will probably have more problems with the analog
electronics.

You might consider buying a complete temperature-measuring instrument
with digital output.

Best regards,

Wim
PA3DJS
www.tetech.nl
.