kvantisointivirheestä

Quantization error, also called quantization noise, is the difference between the actual continuous signal value and its quantized value after analogue-to-digital conversion or digital representation. The error arises because continuous amplitude levels are mapped onto a finite set of discrete levels. The maximum possible error in a uniform quantizer is one half of the quantization step size, while the average error is typically one third of the step size.

In digital audio and image processing, quantization error contributes to distortion and lowers signal‑to‑noise ratio. The human perceptual system may tolerate a certain amount of error; therefore, psychoacoustic models are used to choose quantization step sizes that minimize audible distortion while reducing data size. In other domains such as scientific measurement and control systems, quantization error can affect accuracy and stability; techniques such as dithering or oversampling are employed to mitigate its impact.

The level of quantization error depends on the bit depth of the representation: higher bit depths provide finer resolution and smaller step sizes, reducing the error. For example, 16‑bit audio has a step size of about 1/65536 of the full scale, yielding a typical quantization noise floor of 96 dB SPL for full‑scale sine waves. 24‑bit systems reduce the noise floor to roughly 144 dB SPL.

Mitigation methods include non‑uniform quantization, which uses variable step sizes optimized for a given signal statistics; sigma‑delta modulation, which spreads quantization noise over a wide bandwidth; and error‑feedback techniques that incorporate previous quantization errors into future samples.

Quantization error is a fundamental limitation of finite‑resolution systems and is accounted for in the design of digital signal processing algorithms, compression schemes, and measurement instrumentation. It is a key ingredient in the trade‑offs between dynamic range, data size, and computational complexity.