Right.... I'll try to be clearer... lets say you need to read a temerature 0° to 100°C if you look at the Pic's ADC the total resolution is 0 to 1023... So in a perfect world you would get precision to 0.1°C... Now put that into practice, the sensor may only output 1v to 2.5v where 1v = 0°C and 2.5v = 100°C (lets assume a linear output)
1.5v span. SO using a 5 volt referenced ADC you get 1.5v * 1023 / 5v = 307...... Using a resolution of 306 we only get a temperature precision of 0.3°C which may not be enough... Also the reading will be VERY unstable moving from 0.0°C to 0.6°C erratically..... To steady the reading we over-sample... ( Interpolation is the term for sampling a frequency enough times to re-create or process a moving signal..ie... sound waves ) With a mainly steady input voltage we need to read many times to obtain a more true / average sample.
As explained in the above, the voltage produced by sensors are noisy to some extent.. If you over sample by a factor of 4 you achieve a "sort of" 12bit result.... 1.5v * 4092 / 5v = 1227. Now we have 1227 bits to represent the temperature... We can do 0.1°C quite comfortably and as there are more readings the value is a little more stable.
If you go even further... You can watch the noise on a scope... Calculate the average high's and low's of the signal and work out the best sampling times..
The other option is to reduce the ADC span by using the Vref+ / - ... Moving the ADC's span to fit the signals voltage window... But! its still worth over-sampling to smooth the value ready for displaying.
1.5v span. SO using a 5 volt referenced ADC you get 1.5v * 1023 / 5v = 307...... Using a resolution of 306 we only get a temperature precision of 0.3°C which may not be enough... Also the reading will be VERY unstable moving from 0.0°C to 0.6°C erratically..... To steady the reading we over-sample... ( Interpolation is the term for sampling a frequency enough times to re-create or process a moving signal..ie... sound waves ) With a mainly steady input voltage we need to read many times to obtain a more true / average sample.
As explained in the above, the voltage produced by sensors are noisy to some extent.. If you over sample by a factor of 4 you achieve a "sort of" 12bit result.... 1.5v * 4092 / 5v = 1227. Now we have 1227 bits to represent the temperature... We can do 0.1°C quite comfortably and as there are more readings the value is a little more stable.
If you go even further... You can watch the noise on a scope... Calculate the average high's and low's of the signal and work out the best sampling times..
The other option is to reduce the ADC span by using the Vref+ / - ... Moving the ADC's span to fit the signals voltage window... But! its still worth over-sampling to smooth the value ready for displaying.