Control engineers often default to double-precision floating point, but this can waste CPU cycles on embedded targets. The post breaks down three numeric representation types — floating point, integer, and fixed-point — and walks through their tradeoffs for control and DSP applications. Key concerns covered include quantization effects, overflow handling, and integrator precision in PID controllers. A concrete example using a 16-bit ADC shows how 32-bit float can miss small increments in the integrator state, causing drift or oscillation. Practical rules of thumb are provided for choosing data types based on processor capabilities, with a strong recommendation to benchmark early in the project.
