Degrees in two bytes

What the chip sends

The LM75B keeps four registers. Register 0 is the temperature, and at power-up it is the one selected, so a read with no request first gets it. The sketch asks for it by name anyway: one byte, 0x00, then a request for two bytes back.

The first byte is whole degrees, the second is the fraction. Bit 8 is one degree, bit 7 half a degree, bit 5 an eighth. Below that the chip sends zeros.

One line of arithmetic

Put the two bytes side by side and the result is 256 times the temperature. So the whole conversion is:

uint8_t hi = Wire.read();                 // whole degrees
uint8_t lo = Wire.read();                 // the fraction
int16_t raw = (int16_t)((hi << 8) | lo);  // signed: 256 x the temperature
float c = raw / 256.0;

Read the two bytes into two variables first. Written as one expression, (Wire.read() << 8) | Wire.read(), C++ is free to call the second read first.

21.5 °C arrives as 0x15 0x80: 5504, and 5504 over 256 is 21.5. The same line works at any resolution, because both makers send the unused low bits as 0.

Below zero

The top bit is the sign. At −4 °C the bytes are 0xFC 0x00, which is −1024 as a signed 16-bit number: −4 exactly.

The int16_t is what makes it signed. The code on the page this book replaces put the bytes in an int, shifted them right by five and multiplied by 0.125. On an Uno, whose int is 16 bits, that works. On an ESP32 or a Pico an int is 32 bits, the top bit of the temperature is just bit 15 of a positive number, and −4 °C prints as 252 °C. Press the second button in the figure to see it.

A room never goes below zero, so the mistake survives until the sensor goes into a freezer or outside in winter.