Questions: Dithering Techniques and Quantization Noise Shaping

Question 1

An audio engineer quantizes a quiet musical passage with a 16-bit ADC. Without dithering, the passage sounds unpleasantly harsh. With dithering, it sounds much cleaner despite the spectrum analyzer showing a slightly higher noise floor. What explains the improvement?

• A. Dithering increases the effective bit depth of the ADC, lowering quantization error overall
• B. Without dithering, the small signal only traverses a few quantization levels, creating correlated harmonic distortion; dithering converts this into random white noise, which is perceptually less objectionable
• C. Dithering reduces the power of the quantization error, which is why the noise floor appears lower on the spectrum analyzer
• D. The ADC's SQNR improves with dithering because the random noise cancels some of the quantization error

Answer: (B)

When a small signal moves through only a few quantization levels, the quantization error is a deterministic, periodic function of the input — producing harmonic distortion: spurious tones at harmonics of the signal frequency. These are highly audible even at low amplitudes because the ear is sensitive to tonal artifacts. Dithering adds random noise before quantization, ensuring the quantizer input fluctuates through decision boundaries randomly. The quantization error becomes white noise rather than harmonic distortion. White noise is far less perceptually objectionable than tonal artifacts, even if the total noise power is slightly higher. Option C is wrong: dithering does not reduce the noise floor — it raises it. Option D is incorrect: SQNR does not improve; the trade is distortion for noise.

Question 2

What does dithering actually do to the quantization error, and what is the fundamental trade-off?

• A. It eliminates quantization error entirely by correcting rounding errors before they accumulate
• B. It converts correlated, signal-dependent distortion into approximately white (uncorrelated) noise, at the cost of a slightly increased noise floor
• C. It reduces the step size of the quantizer, effectively increasing resolution by one or two bits
• D. It pushes quantization error to high frequencies using a feedback filter, reducing noise in the signal band

Answer: (B)

Dithering does not eliminate or reduce quantization error — it changes its character. The random noise added before quantization 'confuses' the quantizer so that rounding errors are no longer deterministically tied to the input signal. The result is white noise instead of harmonic distortion. Total noise power increases slightly (you added noise), but the character of the error improves from structured distortion to unstructured noise. This is the fundamental trade: distortion for noise. Option D describes noise shaping, which is a related but distinct technique. Option C describes oversampling. Option A is incorrect — no technique eliminates quantization error without infinite resolution.

Question 3

Noise shaping can achieve higher effective resolution in a target frequency band by redistributing quantization noise to out-of-band frequencies without reducing the total quantization noise power.

• T. True
• F. False

Answer: True

Noise shaping feeds back the quantization error through a filter and subtracts the filtered error from the input of the next quantization step. This creates a feedback loop that spectrally shapes the noise — pushing quantization noise energy toward high frequencies while suppressing it in the band of interest. Total noise power is conserved (energy cannot be destroyed), but it is redistributed. Delta-sigma modulation exploits this aggressively: a 1-bit quantizer running at very high sample rates with heavy noise shaping can achieve 20–24 bits of effective resolution in the audio band because almost all the quantization noise has been pushed above the decimation filter cutoff.

Question 4

Dithering reduces the total quantization noise power, which is the primary reason it improves signal quality in audio applications.

• T. True
• F. False

Answer: False

This is a common misconception. Dithering does not reduce total noise power — it slightly increases it, because you are adding a random noise signal to the input before quantization. The improvement in signal quality comes from changing the character of the quantization error from correlated harmonic distortion (which the ear finds very objectionable as spurious tones) to uncorrelated white noise (which sounds like gentle hiss). Total SNR may actually worsen slightly with dithering; the perceptual quality improves. This distinction between objective SNR and perceptual quality is central to understanding why dithering works.

Question 5

Why does dithering improve perceived audio quality for small-amplitude signals even though it increases the measured noise floor?

Model answer: Without dithering, a small signal that traverses only a few quantization levels produces quantization error that is a periodic, deterministic function of the signal. This appears in the frequency domain as harmonic distortion — spurious tones at harmonics of the signal frequency — which the auditory system readily detects as a harsh, unpleasant artifact. Dithering adds random noise before quantization, randomizing the rounding decisions. The quantization error no longer tracks the signal; it becomes white noise. White noise is perceived as gentle hiss, which is far less objectionable than harmonic distortion even at higher total power. The ear tolerates broadband noise much better than tonal artifacts at the same power level, so the trade-off (slightly higher noise floor, elimination of distortion) is a net perceptual improvement.

The key distinction is between objective noise power (dithering worsens it slightly) and perceptual quality (dithering improves it by changing noise character). The explanation should connect the mechanism — correlated vs. uncorrelated quantization error — to the auditory system's differential sensitivity to tonal artifacts versus broadband noise.