Quantization Noise Floor

Unlike thermal noise, the quantization noise floor is deterministic and correlated with the input signal. In Nyquist-rate converters, quantization error approximates white noise only when the input is sufficiently complex and the converter resolution is high. For low-amplitude or periodic inputs, the error becomes highly structured, creating harmonic spurs rather than a flat noise pedestal. Sigma-delta converters exploit this by intentionally shaping the noise power out of the band of interest through a feedback loop, creating a distinctive high-pass noise transfer function that is a direct artifact of the modulator order and architecture.

Dithering is the intentional injection of a small, uncorrelated noise signal prior to quantization to break the deterministic relationship between the input and the quantization error. This technique linearizes the converter's average transfer function, eliminates idle tones, and whitens the noise floor. Common dither types include:

• Subtractive dither: Noise is added before quantization and digitally subtracted after, preserving SNR while decorrelating error.
• Non-subtractive dither: Noise is added without subtraction, trading a slight SNR degradation for complete harmonic suppression. The specific dithering strategy employed leaves a measurable signature in the residual noise floor statistics.

Real-world converters deviate from the ideal quantization model, introducing additional noise and distortion that elevate and color the noise floor:

• Differential Non-Linearity (DNL): Large DNL errors create localized deviations in step size, generating input-dependent noise modulation and missing codes that appear as spectral gaps or spurs.
• Integral Non-Linearity (INL): Smooth, low-order INL curvature produces harmonic distortion that folds back into the band of interest, raising the effective noise floor.
• Aperture Jitter: Timing uncertainty in the sampling instant phase-modulates the input signal, producing a broadband noise pedestal proportional to the input frequency and slew rate.
• Thermal and kT/C Noise: These fundamental analog noise sources add to the quantization error, setting the ultimate physical limit on the achievable noise floor.

In time-interleaved ADCs, multiple sub-converters sample in a round-robin sequence to achieve higher aggregate sample rates. However, mismatches in gain, offset, and timing skew between the interleaved channels produce deterministic, periodic spurs that appear above the quantization noise floor. These spurs are located at predictable frequency offsets from the input signal and are a dominant, exploitable hardware signature. The pattern of these interleaving spurs—their amplitude, frequency location, and phase—is unique to each physical device and highly stable over time, making it a prime candidate for RF fingerprinting.

The quantization noise floor, when combined with analog front-end imperfections, forms a composite noise signature that is unique to each ADC or DAC. Key fingerprinting features include:

• Noise spectral shape: The frequency-dependent power distribution, including sigma-delta noise shaping profiles and flicker noise (1/f) corners.
• Idle tone patterns: Fixed-frequency spurs generated by limit cycles in sigma-delta modulators or DNL-induced missing codes.
• Amplitude-dependent noise modulation: Variation in the noise floor power as a function of input signal amplitude, revealing the converter's large-signal non-linearity.
• Temperature drift signature: The rate and direction of noise floor elevation as the die temperature changes, reflecting the thermal behavior of the analog front-end.

The quantization noise floor must be distinguished from other noise sources in the signal chain:

• Thermal noise floor: A truly random, Gaussian, and spectrally flat noise source set by the physical temperature and resistance of the analog front-end. It is uncorrelated with the input signal and cannot be shaped or eliminated by dithering.
• Phase noise: Originates from oscillator instabilities and appears as a spectral skirt around the carrier, with a characteristic 1/f² or 1/f³ roll-off. Unlike quantization noise, phase noise is multiplicative, scaling with signal power.
• Quantization noise floor: Signal-dependent, spectrally shapeable, and deterministic. In high-resolution converters with proper dithering, it can be pushed below the thermal noise floor, making thermal noise the dominant residual.

The inherent difference between an analog input value and its discrete digital representation. This fundamental, signal-dependent noise source has statistical properties shaped by the converter's architecture and non-idealities.

• Ideal Model: Uniform distribution within ±0.5 LSB
• Non-Ideal Behavior: Correlated with input signal when resolution is low
• Fingerprint Utility: Deviation from ideal white noise reveals device-specific artifacts

A dynamic performance metric expressing the true resolution of a data converter after accounting for noise and distortion. ENOB directly quantifies the usable dynamic range above the quantization noise floor.

• Calculation: Derived from SINAD: ENOB = (SINAD - 1.76) / 6.02
• Typical Values: A 16-bit ADC may achieve only 13.5 ENOB in practice
• Fingerprint Relevance: ENOB variations between units reflect aggregate hardware imperfections

The ratio of total signal power to the sum of all noise and harmonic distortion components. SINAD provides a single figure of merit capturing the aggregate analog imperfections exploitable for device identification.

• Components: Includes quantization noise, thermal noise, and harmonic distortion
• Measurement: Typically expressed in dB, measured with a pure sine wave input
• Fingerprint Context: Unit-to-unit SINAD variation reveals unique impairment signatures

The intentional injection of a small amount of noise into an analog signal prior to quantization. This technique decorrelates quantization error from the input, linearizing the converter but also modifying the intrinsic fingerprint.

• Types: Triangular PDF dither, Gaussian dither, subtractive dither
• Benefit: Eliminates harmonic distortion from low-amplitude signals
• Fingerprint Impact: Alters the noise floor structure, potentially masking or modifying device-specific signatures

The ratio of the fundamental signal's RMS amplitude to the highest spurious component in the output spectrum. SFDR identifies device-specific non-linear artifacts rising above the quantization noise floor.

• Typical Range: 80-100 dB for high-performance converters
• Spur Sources: Interleaving mismatch, INL patterns, clock feedthrough
• Fingerprint Utility: Spurs are deterministic, repeatable, and highly device-specific

The broadband, unavoidable noise generated by random thermal agitation of charge carriers in resistive components. This sets the fundamental detection limit and contributes a Gaussian, device-specific noise pedestal.

• Formula: P = kTB, where k is Boltzmann's constant
• Interaction: Sums with quantization noise to form the total noise floor
• Fingerprint Role: Component tolerances create unit-specific thermal noise contributions