Sfdr

Spurious-free dynamic range is the answer to a question signal-to-noise ratio alone cannot answer: not how noisy a converter's output is on average, but how tall the single worst unwanted tone in its spectrum rises above the noise floor. A converter can have excellent average noise performance and still be useless for a specific application if one spurious tone — a single wrong frequency the converter itself manufactures — sits close enough to a wanted signal to be mistaken for it.

The distinction from signal-to-noise ratio is exactly what makes SFDR the metric that matters for certain applications and irrelevant for others. SNR averages all the unwanted power in the spectrum — noise plus every spur — into a single number, so a converter with very low average noise but one large, narrow spur can report an excellent SNR while still being unusable for a task that needs to detect a faint signal sitting right next to that spur. SFDR instead isolates specifically the gap, in decibels, between the fundamental tone and the single largest spurious component anywhere in the spectrum, regardless of whether that spur came from harmonic distortion, a timing nonlinearity, or some other mechanism entirely. A radar or communications receiver trying to detect a weak target signal in the presence of a strong nearby interferer cares almost exclusively about this one number, because the interferer's own harmonic spurs are exactly what could be mistaken for the weak signal of interest.

Spurs have real physical origins, and most trace back to a single root cause: some nonlinearity in the signal path that was supposed to be linear. An amplifier or converter with perfectly linear transfer characteristics, driven by a single input tone, would output only that same tone — any additional frequency content in the output spectrum is direct evidence of a nonlinear term somewhere in the chain. A quadratic nonlinearity generates a second-harmonic spur at twice the input frequency; a cubic term generates a third-harmonic spur at three times the frequency, and also generates intermodulation spurs at sum-and-difference frequencies when two input tones are present simultaneously. Differential nonlinearity in a converter's own transfer curve, code-dependent settling errors, and the same aperture jitter and kickback-noise mechanisms covered elsewhere in this series can all contribute their own characteristic spurs, which is why SFDR is treated as a system-level specification rather than attributable to any one circuit block in isolation.

Intermodulation spurs are frequently the worse problem precisely because they cannot simply be filtered away. A single-tone test reveals harmonic distortion, which lands at clean integer multiples of the input frequency and, for a narrowband signal, often falls outside the band of interest where a filter can remove it. A two-tone test, standard practice for characterizing receivers and wideband converters, reveals intermodulation products at sum and difference frequencies — most critically the third-order terms at 2f1−f2 and 2f2−f1 — which land close to the original two tones rather than at a clean multiple, often directly inside the signal band where no filter can separate them from the wanted signal. This is why two-tone SFDR testing, not single-tone testing, is the standard characterization method for any converter or receiver meant to handle more than one simultaneous signal.

SFDR's dependence on input amplitude, unlike SNR's relative constancy, is what makes it a curve rather than a single number in a real datasheet. Because spurs originate from nonlinear terms whose order determines how they scale with input amplitude — a third-order spur grows three decibels for every one decibel of input increase, faster than the fundamental's own one-to-one growth — SFDR actually degrades as input level rises, even though the fundamental signal itself is getting stronger. This produces a distinctive curve when plotted against input amplitude, rather than the roughly flat relationship SNR shows, and is why application designers care about SFDR specifically at the input level their own system actually expects to see, not just at full scale.

SFDR is the metric that exists because an average is sometimes the wrong question. A converter's typical noise floor can be excellent while a single, specific spur quietly sets the actual limit on what a real system can detect — which is exactly why this number, not SNR alone, governs whether a converter is fit for receivers, radar, and any other application where one strong unwanted tone can hide the one weak signal that actually matters.