GTS Learn · Digital · Advanced

ADAA: Anti-Derivative Anti-Aliasing

6 min read

ADAA, anti-derivative anti-aliasing, is a newer way to make digital distortion clean. Instead of reading the distortion curve at each sample, it works out the curve’s average over the gap between two samples. That small change suppresses much of the folding that causes aliasing, at a fraction of the cost of heavy oversampling.

Measure the methods

Computed live in your browser; textbook curves, not a plugin's tuning

The Idea

A digital curve only sees the signal at the sample points, so a sharp corner between them is either caught or missed, and that hit-or-miss is what folds back as aliasing. ADAA asks a different question: what did the curve do, on average, while the signal travelled from the last sample to this one? Averaging smooths exactly the sharp edges that alias.

ADAA and Oversampling Together

On its own at the session rate, ADAA helps but can’t do everything: the demo shows a hard-clipped 5 kHz tone improving only a few dB at 1×. Combined with modest oversampling it shines: ADAA at 4× reaches about −81 dB, where plain 4× sits near −55. That pairing is how a plugin gets 16×-class cleanliness for a quarter of the work.

The Trade-Offs

  • Half a sample of delay for first-order ADAA (one for second-order), and a very gentle low-pass on the top octave.
  • Careful maths: the formula divides by the difference between samples, which is unstable when they are nearly equal, so implementations need well-designed fallbacks.
  • The curve must have a closed-form antiderivative, which shapes which curves are practical.

In GTS Plugins

Drive Strip uses first-order ADAA on its output clip (inside 16× oversampling) and on the GRIT voicing, and second-order ADAA on RECTIFY’s corner, all inside its oversampling. The harmonic signature is unchanged to within 0.1 dB; the cost is a top-end change of under half a dB at 19 kHz, matched on the dry path so Delta and Mix stay exact.

Try It

Try this in the demo: set the note to 8 kHz and the drive to 12 dB, then step through the four methods. Watch how much of the floor between the spikes each one removes, and how much processing each one costs.

Common Mistakes

  • Expecting ADAA alone to replace oversampling. It is strongest as a partner to modest oversampling.
  • Forgetting its delay. If a plugin mixes dry and processed signal, the dry path needs the same half-sample delay or the blend combs.
Go deeper: the maths Advanced · equations

First-order ADAA

With F the antiderivative of the curve f:

y[n] = ( F(x[n]) − F(x[n−1]) ) / ( x[n] − x[n−1] )
when x[n] ≈ x[n−1]:   y[n] = f( (x[n] + x[n−1]) / 2 )

For a hard clip, F(x) = x²/2 inside the ceiling and |x| − ½ outside it. Second-order ADAA applies the same idea to the second antiderivative and suppresses more, at the cost of a full sample of delay and harder numerics.

Reading

Parker, Zavalishin and Le Bihan, "Reducing the aliasing of nonlinear waveshaping using continuous-time convolution" (DAFx-16, 2016), and Bilbao, Esqueda, Parker and Välimäki, "Antiderivative antialiasing for memoryless nonlinearities" (IEEE Signal Processing Letters, 2017), are where ADAA comes from.

From the GTS Lab

Measured on Drive Strip 1.1.2 with a −18 dBFS sine through the shipped oversampling, worst audible fold below 16 kHz, over eleven probe tones from 1 to 14 kHz: aliasing stays below −80 dBc across the whole Drive range on Silk, Hearth, Gold and Rectify, and up to drive 13.7 on Tape, 10 on Grit and 8.9 on Warm. A hotter input reaches that edge sooner. At the default drive of 4, Grit measures −106 dBc and Rectify −105.6 dBc.

The output clip runs inside 16× oversampling with first-order ADAA: in its default hard mode it stays below −80 dBc up to 18 dB over the ceiling, and in soft mode below −95 dBc up to 12 dB over.

Keep Going

Hear measured, level-matched saturation on your own tracks: Drive Lite is free, and Drive Strip has a 14-day trial of the whole plugin.