Time-Interleaved ADC

Variations in the full-scale amplitude range between sub-ADCs create amplitude-modulated artifacts in the digitized output. A gain error in one channel scales its samples relative to others, producing spurs at frequencies of f_in ± k * (f_s / M), where M is the number of interleaved channels and k is an integer.

• Signature: Spurs appear as amplitude modulation sidebands around the input signal.
• Exploitability: Gain mismatch is static over short timeframes, making it a stable, high-confidence identifying feature.
• Measurement: Quantified as the percentage deviation from the mean gain across all channels.

Each sub-ADC possesses a unique DC offset voltage. When channels are cycled, this creates a fixed, periodic pattern of DC shifts at the interleaving rate. This produces a distinct set of spurs in the frequency domain that are completely independent of the input signal.

• Signature: Fixed tones at multiples of the channel switching rate (k * f_s / M).
• Key Feature: These tones persist even with no input signal present, acting as a constant, always-on beacon for device identification.
• Origin: Caused by random transistor threshold voltage mismatches in each sub-ADC's front-end comparator.

The most critical mismatch is the deterministic clock phase error between channels. Instead of sampling at exact intervals of T_s, one channel consistently samples slightly early or late. This non-uniform sampling phase-modulates the input signal.

• Signature: Spurs appear at the same locations as gain mismatch but with a 90-degree phase shift relative to the signal, making them orthogonal.
• Impact: Timing skew errors grow linearly with input frequency, dominating the error budget in wideband systems.
• Fingerprint Value: Extremely sensitive to physical layout parasitics, making it a highly unique, process-dependent signature.

Differences in the input sampling network bandwidth of each sub-ADC channel cause frequency-dependent gain and phase errors. The track-and-hold circuit in each channel has a slightly different RC time constant due to on-resistance and capacitance variations.

• Signature: The effective gain and phase error for a channel become a function of the input frequency, not just a static value.
• Complexity: This introduces a dynamic, frequency-dependent fingerprint that is significantly harder to clone or calibrate out.
• Modeling: Requires a per-channel transfer function H_i(jω) rather than a simple scalar gain error.

The composite output of a TI-ADC is a cyclostationary process. Its statistical properties—specifically the variance of the error—are not constant but vary periodically with the channel switching sequence. This is a fundamental departure from a standard ADC's stationary noise behavior.

• Analysis: This periodicity is directly observable in the autocorrelation function of the output error signal.
• Fingerprint Extraction: Cyclostationary analysis techniques can isolate the periodic error power from the random thermal noise floor, enhancing the signal-to-noise ratio of the fingerprint.
• Distinction: A perfectly matched TI-ADC would produce a stationary error; the degree of non-stationarity is a direct measure of the mismatch signature strength.

Mismatch errors cause aliasing of out-of-band noise and distortion back into the Nyquist band. The periodic switching of offset and gain errors effectively multiplies the input signal by a periodic sequence, convolving the signal spectrum with the mismatch spectrum.

• Mechanism: Broadband noise from each channel is folded back into the first Nyquist zone at multiples of f_s/M.
• Result: The in-band noise floor is not flat but exhibits a channel-dependent noise pedestal with periodic peaks.
• Exploitation: The specific shape of this folded noise spectrum is a direct map of the per-channel noise power spectral density, creating a rich, multi-dimensional fingerprint.

The static gain, offset, and timing skew errors between parallel sub-converters in a time-interleaved ADC. These mismatches produce deterministic, repetitive spurs in the output spectrum at frequencies of f_s/M and its multiples, where M is the number of interleaved channels. This is the dominant and most exploitable hardware signature in TI-ADCs.

• Gain mismatch: Amplitude modulation of the sampled signal by channel-dependent scaling factors
• Offset mismatch: A fixed DC pattern that repeats every M samples, creating tones at multiples of f_s/M independent of the input
• Timing skew: Channel-dependent sampling clock phase errors that produce input-frequency-dependent spurs

A class of digital and analog techniques used to detect and correct sampling clock phase errors between interleaved channels. Without calibration, picosecond-level skews create spurs that severely degrade SFDR. Foreground methods inject a known test tone, while background techniques estimate skew from the digitized signal itself using statistical correlation.

• Foreground calibration: High accuracy but requires interrupting normal operation
• Background calibration: Runs continuously by exploiting signal statistics or injecting low-level pilot tones
• Derivative-based estimation: Uses the product of adjacent channel outputs to estimate relative timing error

The deviation of the actual slope of a data converter's transfer function from the ideal slope. In a time-interleaved ADC, channel-to-channel gain mismatch scales each sub-ADC's output differently, creating amplitude modulation sidebands around the mismatch spur frequencies. This static linear imperfection contributes a systematic, identifiable bias.

• Expressed as a percentage of full-scale range or in LSBs
• Interacts with offset error to create a composite DC and low-frequency signature
• Can be calibrated digitally by multiplying each channel's output by a correction coefficient

A constant, static voltage difference between the ideal and actual transfer function of a data converter. In a TI-ADC, each sub-converter has its own DC offset, creating a fixed pattern that repeats every M samples. This pattern produces tones at multiples of f_s/M that are completely independent of the input signal.

• Creates a deterministic, input-independent fingerprint
• Offset tones are the easiest mismatch spurs to identify and exploit
• Typically calibrated by measuring and subtracting the mean output of each channel in the digital domain

The ratio of the fundamental signal's RMS amplitude to the highest spurious component in the output spectrum. In time-interleaved ADCs, interleaving mismatch spurs are often the SFDR-limiting factor, making this metric a direct indicator of the strength of a device's exploitable fingerprint.

• Measured in dBc (relative to carrier) or dBFS (relative to full scale)
• A TI-ADC with poor calibration may exhibit SFDR limited to 40-50 dB by mismatch spurs
• Post-calibration SFDR improvements of 20-30 dB are typical in high-performance designs

The short-term, non-cumulative deviation of a clock edge from its ideal position in time. In a TI-ADC, the shared sampling clock's aperture jitter directly translates to sampling uncertainty, while channel-specific clock skew creates the interleaving timing mismatch. Both contribute phase noise and broadband noise pedestals.

• Random jitter: Gaussian-distributed timing errors that raise the noise floor
• Deterministic jitter: Periodic timing variations from coupling or power supply, creating identifiable spurs
• Jitter requirements scale with input frequency: a 1 ps RMS jitter limits SNR to ~50 dB at 1 GHz input