Figure 2-2 shows the sensor output vs ambient temperature data for the same data as in Figure 2-1. The goal of regression is to obtain the simplest acceptable relationship between the desired variable (ambient temperature), and the known variable (sensor output). Most data analysis software provides curve fitting/trend line tools that can be used to quickly obtain a linear or polynomial relationship.

Figure 2-2 Ambient Temperature as a Function of Sensor Output Temperature

The R² value can be used here to evaluate the quality of the regression. A higher R² value will result in less system temperature error contributed by the curve fit. Using a linear fit will result in a faster computation time, but in a typically lower R² value when compared to a polynomial regression. In this example, a linear regression line of T_amb = -4.641 +1.03T_c gives an R² value of greater than 0.999. This quality of fit is sufficient for most system requirements. In this case, the regression shows that the error can be corrected with just a simple offset and gain correction. The I²C host should be used to apply this equation to compensate the sensor temperature readings.