Time-Interleaved ADC Mismatch

A fixed DC voltage difference between the baseline levels of parallel sub-ADCs. This error is independent of the input signal amplitude and creates fixed-frequency spurs in the output spectrum.

• Cause: Transistor threshold voltage variations in comparators or residue amplifiers.
• Spectral Signature: Tones appear at multiples of fs / M, where fs is the total sample rate and M is the number of interleaved channels.
• Calibration: Corrected by digitally subtracting the measured offset value for each sub-ADC. A common technique involves random chopping to measure and nullify the offset in the background.

A difference in the full-scale voltage range or slope of the transfer function between sub-ADCs. This error scales with the input signal amplitude.

• Cause: Mismatches in reference voltages or capacitor ratios in the sampling network.
• Spectral Signature: Creates spurs at ±fin + k * fs / M, where fin is the input frequency. These spurs are amplitude-modulated copies of the input signal.
• Calibration: Mitigated by multiplying each sub-ADC's output by a correction coefficient. Foreground calibration uses a known test tone, while background methods correlate the outputs of adjacent channels.

A deviation in the exact sampling instant of each sub-ADC from its ideal periodic position. This is the most performance-limiting mismatch at high input frequencies.

• Cause: Mismatched clock path lengths, buffer delays, and aperture uncertainty in the track-and-hold circuits.
• Spectral Signature: Produces spurs at the same locations as gain mismatch (±fin + k * fs / M), but the spur magnitude increases with input frequency.
• Calibration: The most complex to correct. Digital techniques estimate the skew by correlating the derivative of the signal with the output error, then apply a fractional delay filter (e.g., a Farrow structure) to resample the data to the correct time instant.

In a practical TI-ADC, all three mismatches occur simultaneously, creating a complex spurious spectrum that degrades the Spurious-Free Dynamic Range (SFDR).

• Offset spurs are fixed in frequency and independent of the input signal.
• Gain and timing spurs move with the input frequency, making them harder to distinguish and cancel.
• Timing skew dominates at high Nyquist-zone inputs, often requiring foreground factory calibration or sophisticated background adaptive filters to achieve >70 dB SFDR in 4+ channel arrays.

Corrects amplitude mismatches between sub-ADCs by estimating and normalizing the gain of each interleaved channel. A reference signal or blind estimation technique measures the relative output amplitude of each slice. A digital multiplier then scales the output of each channel to match a common reference, eliminating the gain spurs that appear at frequencies of f_s/M ± f_in in the output spectrum. This is often the simplest mismatch to correct.

Neutralizes DC offsets that vary between interleaved sub-ADCs. Each channel's offset is measured by grounding the input or averaging the output in the absence of a signal. The measured offset is then subtracted digitally from each sample. Uncorrected, these mismatches create a fixed offset spur at multiples of f_s/M in the spectrum, independent of the input signal. This spur is often the most visually obvious artifact in a wideband spectrogram.

The most challenging mismatch to correct, timing skew arises from sub-picosecond differences in the sampling clock edges between channels. Correction requires a fractional delay filter, often implemented as a polyphase FIR filter or a Farrow structure, to resample the misaligned channel's data to the correct temporal position. Blind estimation techniques often use the correlation between adjacent channel outputs to drive a feedback loop that continuously minimizes the skew error.

Addresses differences in the analog input bandwidth of each sub-ADC track-and-hold circuit. This frequency-dependent error cannot be fixed by a simple scalar. Correction requires an adaptive FIR filter per channel that models the inverse of the bandwidth mismatch transfer function. The filter coefficients are often derived from a foreground calibration routine that injects a known wideband chirp signal into the ADC array.

Two fundamental operational modes for mismatch correction. Foreground calibration interrupts normal operation to inject a known training signal, achieving high accuracy but causing downtime. Background calibration runs continuously during normal operation by exploiting statistical properties of the input signal (blind estimation) or injecting a low-level pseudo-random noise sequence. Background methods are essential for mission-critical systems that cannot tolerate interruption.

The primary metric degraded by interleaving mismatches. SFDR is the ratio of the RMS signal amplitude to the largest spurious spectral component. Mismatch spurs directly reduce SFDR, limiting the ability to detect weak signals near strong interferers. A well-calibrated 4-channel TI-ADC can achieve an SFDR exceeding 90 dBFS, whereas uncorrected mismatches may limit it to 50-60 dBFS, rendering it useless for sensitive SIGINT applications.