Numerical Discrepancy
Computational noise arises from the mapping of continuous signal values into a constrained set of discrete levels within digital hardware. This fixed point quantization error occurs because bit depth limitations prevent the exact representation of fractional coefficients or analog voltage levels. A processor rounding a 24 bit sensor input to a 16 bit register introduces a deterministic variance that behaves as a noise floor.
Rounding Mechanism
Logic circuits apply truncation or convergent rounding to manage word length growth during multiplication. The resulting fixed point quantization error accumulates across successive arithmetic operations, especially in digital filters where feedback loops amplify the mismatch. A single multiplication of two N bit numbers produces a 2 N bit result, forcing the hardware to discard the least significant bits.
Signal Fidelity
Signal to noise ratios degrade as the available bit depth decreases. Designers quantify fixed point quantization error through the calculation of the least significant bit value (LSB) and the associated probability density function of the noise. If the input signal is sufficiently complex, the error approximates a white noise distribution with a variance of LSB squared divided by twelve.
Implementation Boundary
Saturation logic provides a hard limit when signal excursions exceed the available dynamic range of the fixed point format. This overflow represents a non linear distortion distinct from the standard fixed point quantization error seen during normal scaling. Designers select the integer and fractional bit counts to keep these artifacts below the required measurement tolerance of the instrument.