Description
Background: The STM32F429ZI MCU includes an analog sensor whose output voltage changes linearly with the temperature of the
chip and an internal analog-to-digital (A/D) converter that can be used to convert that voltage into a digital value. That value differs
from one chip to another due to variations in the manufacturing process that affect: (1) an offset in the sensor output and (2) the value
of an internal reference voltage (Vref) that determines the gain of the A/D converter. (Note: A/D gain refers to the change in the digital output for a given change in input voltage.)
To improve the accuracy of the temperature sensor (especially for absolute temperature measurement), calibration values are individually measured for each part during production test and stored in the system memory area. First, the reference voltage Vref is set to 3.3
volts, measured by the A/D converter and recorded. Then with Vref held at 3.3 volts, temperature measurements are taken at 30o and
110o Celcius, converted by the A/D converter and recorded.
When we run our program, however, Vref will not be 3.3 volts. However, we can use the A/D converter to measure the actual value.
The ratio of the actual measurement to the calibrated measurement is a multiplier that we can use to scale the two calibrated temperature measurements into values that correspond to the actual reference voltage.
With scaled A/D measurements of 30o
and 110o C, we can calculate the slope (𝑚) and offset (𝑏) of the linear relationship (𝑦 = 𝑚𝑥 +
𝑏) between temperature and A/D measurements.
Assignment: Create a single ARM Cortex-M4 assembly source code file containing the linear interpolation function given below. This
function is called several places in the main program (download from here) that displays the calibration values, the scaled calibration
values, and the chip temperature as both raw A/D output and Celcius values. The program also displays a moving plot of temperature
as shown.
int32_t MxPlusB(int32_t x, int32_t mtop, int32_t mbtm, int32_t b) ;
This function should return the value of the familiar linear function, 𝑦 = 𝑚𝑥 +𝑏. Except for the displayed value of the scale factor, all
of the variables in the program are integers because they are derived from integer values returned by the A/D converter. This requires
that the value of the slope (m) be represented as the ratio of two integers, mtop divided by mbtm, and thus your function must include
an integer division. (Be careful with the order of evaluation or the result will be incorrect.)
Since integer division discards (truncates) the fractional part of the quotient, you will
get better accuracy if the quotient is rounded. A common approach (for positive operands) is to add 0.5 to the quotient produced by real division and then copy the result
into an integer (discarding the fractional part of the sum), as in:
int q = (int) (15.0/4.0 + 0.5) ; // 15/4 = 3.75 rounds to 4
However, your function will be using integer division, so we must rearrange the operands so that the addition occurs before the division, by adding one-half of the divisor
to the dividend as in:
int quotient = (15 + 4/2)/4 ; // (15 + 4/2)/4 → 17/4 → 4
Unfortunately, this code rounds in the wrong direction when the quotient is negative!
Use the following solution that always rounds the integer quotient a/b correctly:
rounding = (((( (dvnd*dvsr) >> 31) * dvsr) << 1) + dvsr) / 2 ;
quotient = (dvnd + rounding) / dvsr ;
The “>>” shift operation must be implemented with an arithmetic shift right (ASR) instruction; this causes the value of the expression (dvnd*dvsr) >> 31 to be 0 when
the signs of dvnd and dvsr are identical, and -1 when they differ. The magnitude of
rounding will be |dvsr/2|, but its sign will be the same as that of dvnd. Adding it
to dvnd thus increases the magnitude of dvnd before the integer division by dvsr
truncates the quotient.
x,y=(0,48)
