fixed-point number

A whole number with an agreed scale, so that it can stand for a fraction without floating-point hardware.

A fixed-point number stores a value v as the integer q = v × 2^n for a fixed number of fraction bits n, so addition is integer addition and a product must be shifted right by n to return to the same scale.

Examples

21.5 °C stored as the integer 5504 with an agreed scale of 256 lets the ESP32-C6, which has no floating-point unit, filter it with integer instructions.
Two Q15 numbers 0.5 × 0.5 multiply as (16384 * 16384 + 2^14) >> 15 = 8192, which is 0.25.