What is the WTA filter, how was it tuned, and how does it differ from a conventional windowed sinc filter?

What the name and the filter are

In a 2016 Hifistatement interview Watts gave the name as "Watts Transient Aligned" and said "I'm trying to align the transients to what it was on the original recording" Hifistatement 2:24Hifistatement 2:28. Weekly ASCII (Japanese, machine-translated) gave the same expansion that year article, Mar 2016.

He describes the WTA as the algorithm, or recipe, that calculates the coefficients of an FIR interpolation filter; in his words "The WTA algorithm actually uses my own windowing function of the ideal sinc impulse response" (April 2016, DAVE thread) Head-Fi, Apr 2016Head-Fi, Feb 2017. His fullest written account (November 2020) says "The WTA algorithm is a unique windowing function designed to maximise the accuracy of recovering transient timing from the original analogue signal" Head-Fi, Nov 2020. He repeats the windowing-function description in talks and posts from 2017 through 2024 (the latest, the audio-2G interview, was published in 2025 and most likely recorded in May 2024) miceblue425 15:15c w 14:48Head-Fi, Apr 2019audio-2G 7:28. Once calculated, the coefficients are fixed in the FPGA Head-Fi, Apr 2016. In July 2022 he said he has never published the window, and that it looks visually different from one shown in the M Scaler thread Head-Fi, Jul 2022. He also said in 2022 that the window must be symmetric for the most accurate transient timing, and that he uses an asymmetric one only in the low-latency video mode, at a sound quality price Head-Fi, May 2022.

How it differs from a conventional windowed sinc

Not a half-band filter. In 2015 he explained that most conventional FIR filters are cascaded half-band filters that pass the original samples unchanged and are -6dB at FS/2, and that apodizing filters and "my filter the WTA filter" are not of this type, so the WTA modifies the original samples Head-Fi, Jun 2015. He accepted a questioner's description of it as a "full band" filter and said it converges on ideal sinc as tap length increases Head-Fi, Jun 2015.

More coefficients exactly sinc. From 2018 on he says about half or over half the WTA coefficients are identical to ideal sinc: in 2019 he wrote that the window value is exactly one for just over half the coefficients, and that in the 1M-tap filter the window starts to change the sinc value only where the coefficient is well below 16-bit level Head-Fi, Jul 2018Head-Fi, Apr 2019Head-Fi, Sep 2019Pursuit Perfect System 8:20AVstore 16:58. He gave figures, "assuming my spreadsheet is right," for the bit level at which WTA and sinc coefficients diverge: "12.2 bits Hugo 2, 13.9 bits Dave, and 16.6 bits M scaler" (March 2018) Head-Fi, Mar 2018Head-Fi, Jan 2018. He said in 2017 that an infinite-tap WTA "would become an ideal sinc function" Head-Fi, Feb 2017.

Fast roll-off. In December 2017, on the Audiophile Style forum, he wrote "all my WTA filters are extremely fast roll-off": the 49,152 tap filter (the WTA 1 tap count he gives for Hugo 2 and Qutest Head-Fi, Feb 2018) "is at -1dB within 75Hz of FS/2", and the 1M tap filter in the M Scaler "gets to 4Hz of FS/2" Audiophile Style, Dec 2017. These 2017 figures say how close to FS/2 the response holds (the 75Hz figure is a -1dB point; he gave no level for the 4Hz one). The 2020 figures below are transition bands, which he measured from corner to -100dB Head-Fi, Nov 2020, so the two sets are not directly comparable. In the same post he said the M Scaler's WTA coefficients are "identical to an ideal sinc function coefficients to a better than 16 bit accuracy" Audiophile Style, Dec 2017.

A compromise between sinc and taper. He describes a rectangular window (all coefficients sinc) as unusable because of the abrupt change from zero to sinc, and a Kaiser window as, "in my experience the best sounding of the conventional windows," but with very few coefficients close to sinc Head-Fi, Nov 2020Passion for Sound 23:37. In his November 2020 1M-tap comparison he wrote "Kaiser has the almost identical frequency domain performance as ideal sinc, but has a time domain performance that looks nothing like ideal sinc," and that the WTA is close to sinc in both domains Head-Fi, Nov 2020. In the same test he reported that the Kaiser measured better than the WTA on transition band (3 Hz against some 12 Hz), and the rectangular sinc far worse (some 6 kHz), and said the Kaiser still sounded worse because almost all its coefficients are not sinc Head-Fi, Nov 2020. In 2021 he framed the WTA as "in essence a compromise" between image attenuation and closeness to ideal sinc, with both affecting transient timing Head-Fi, Oct 2021. In 2025 he said of conventional filters "None of the coefficients are the ideal sinc value," and that the WTA avoids the sudden change to full sinc while maximizing the number of pure sinc coefficients Audiophile Junkie 9:19Audiophile Junkie 10:04.

Against other windowed sinc products. Discussing PGGB in 2021 he wrote "Windowed sinc is categorically not true Whittaker-Shannon interpolation," and "the key here is the number of taps that are Whittaker-Shannon - NOT the number of taps" Head-Fi, Jun 2021Head-Fi, May 2021.

The headline comparison. Across 2015 to 2019 he repeatedly reported that his first WTA at 256 taps sounded better than a conventional filter of 2,000 or 2,048 taps (described variously as a half-band Kaiser, a half-band conventional filter, or a Kaiser), which he calls at least a tenfold improvement from the algorithm alone Head-Fi, Jun 2015Head-Fi, Feb 2017MajorHiFi 10:05Passion for Sound 25:10. In 2020, at 1M taps, he wrote "the 1M Kaiser sounds similar to what I expect a 20k tap WTA filter would sound like" Head-Fi, Nov 2020.

How it was tuned

Origins (1990s). He says the idea came from wanting better timing accuracy without the roughly 1M taps a 16-bit-accurate sinc would need; his first version targeted aliasing, which meant changing the original samples, an idea he says he resisted at first (accounts from 2015 and 2017) Head-Fi, Jun 2015Head-Fi, Feb 2017. He dates the start of design and listening work variously to "the late 90's" and 1998, and names the DAC64 (1999, about 1,000 taps) as the first WTA product Head-Fi, Jun 2015Head-Fi, Jun 2021c w 14:48.

Evolving with tap length. In 2014 he described an overhaul during QBD76 development, from a 2x WTA plus 4x stage to a flat 8x structure, with "loads of listening tests to fine tune the recipe" Head-Fi, Apr 2014. In 2017 he said moving from 8FS to 16FS and reworking the algorithm mattered more than tap length Head-Fi, Jan 2017. In 2018 he said that "different assumptions in the past didn't quite work out with a longer tap length," and that early on he had focused on image aliasing but later found closeness to sinc mattered more, leaving a balancing act c w 13:58Head-Fi, Jul 2018.

The five-variable equation (2020 onward). His November 2020 post describes an equation with five variables: tap length, then four others, including two interdependent ones shaping the start and end of the window; "One of these variables has to be set to 10 parts per million accuracy" Head-Fi, Nov 2020. He wrote then that "The process of optimising the WTA equation took many weeks with probably thousands of listening tests" Head-Fi, Nov 2020. In his CanJam NYC seminar of late February 2022 (published March 2022) and an AVstore interview published August 2022 (recording date not stated) he said only that the windowing function was optimized to about 10 parts per million by listening MFG Audio Reviews 22:49AVstore 20:01. At CanJam SoCal in September 2022 he described listening to and modifying each of the five variables until “that sweet spot is set to within 10 parts per million”, then moving on to the next variable The Source AV 22:00The Source AV 22:30. In an Upscale Audio interview published October 2022 (recording date not stated) he said he tuned “each of those coefficients” to 10 parts per million, his word for the five variables there; his July 2023 post and his 2025 Malaysia talk say the same Upscale Audio 4:43Head-Fi, Jul 2023Audiophile Junkie 10:46. He repeated the five-variable, 10 ppm account in 2026, in a video published March 2026 (recording date not stated) and in a conversation recorded in autumn 2026 and published October 2026 Audio Force 10:45article, Oct 2026. Note the change: in 2020 he named one variable at 10 ppm, later statements apply it to each. He also describes the overall algorithm as optimized over 20 to 25 years miceblue425 15:15Dusty Audio 42:48.

The listening method. He scores specific attributes on dedicated tracks: instrument separation and focus, timbre variation, bass pitch, soundstage depth, and the starting and stopping of notes, then iterates across variables AVstore 18:29Head-Fi, Aug 2022Head-Fi, Sep 2023. In 2023 he said 100 ppm changes are easy to hear while at 10 ppm "sometimes you have to go back and forth 50 times to get a consensus" Head-Fi, Sep 2023. He says the optimum has been the same for every recording he tried, including modern ones made with half-band decimation filters Head-Fi, Dec 2017Head-Fi, Jul 2018Head-Fi, Sep 2021Head-Fi, May 2022.

Re-verification for Quartet (2024). He reported that a new unlimited-precision FFT showed the old coefficients were not measuring as expected; after recalculating, he repeated the listening tests and "the preferred settings were the same as before"; he added that "having the correct values did give a leap in overall SQ" Head-Fi, Oct 2024The Source AV 9:50.

By product

• WTA1 and WTA2 (Hugo 2, Qutest, TT2, DAVE; posts from 2018 to 2022). He describes WTA1 as the prime WTA filter to 16FS (49,152 taps in Hugo 2 and Qutest, 164,000 in DAVE) and WTA2 as a further WTA stage from 16FS to 256FS, followed by an IIR stage Head-Fi, Jul 2022Head-Fi, Feb 2018Head-Fi, May 2022. The orange and red Hugo 2 settings remove WTA2, which he says degrades transient timing Head-Fi, Mar 2022Head-Fi, Sep 2023. In 2018 he said all current DACs except Mojo used the same WTA2 Head-Fi, Oct 2018, and in 2022 that Mojo 2 has no WTA2 Head-Fi, Feb 2022Head-Fi, Jan 2022. He offers only one WTA1 version, on principle Head-Fi, May 2022.
• M Scaler (2018). 1M-tap WTA, identical to sinc to 16.6 bits by his figure Head-Fi, Jul 2018Head-Fi, Mar 2018.
• Quartet, Blackbird WTA (interviews published 2026). He says a research program he started in 2018, after the M Scaler, found two contrary effects degrading transient timing in all interpolation filters, "a fixed problem with a single-stage filter" article, Oct 2026Distorsiuni Interioare 17:04Distorsiuni Interioare 18:37. Blackbird WTA processes them separately; he says the sinc aspect gets ten times better and the other problem is eliminated, and "The story is the new algorithm, the new filter structure" article, Oct 2026article, Oct 2026Audio Force 14:13The Source AV 8:23. Asked in a video published March 2026 (recording date not stated) whether Blackbird WTA is only for Quartet, he answered "Only for quartet" Audio Force 14:55Audio Force 14:59. The name comes from the SR-71 aircraft article, Oct 2026.

What he has not said: he has not published the WTA window or its equation, the values of the five variables, or which two variables are the independent ones Head-Fi, Jul 2022Head-Fi, Nov 2020. In the passages collected here he does not name the two Blackbird effects beyond "the sinc function aspect" and "the other problem" article, Oct 2026.

Quotes

“The WTA algorithm is a unique windowing function designed to maximise the accuracy of recovering transient timing from the original analogue signal.”2020-11-11 Head-Fi, Nov 2020

“…today it is very similar to an ideal sinc function - over half the coefficients are identical to the ideal sinc function.”2018-07-26 Head-Fi, Jul 2018

“One of these variables has to be set to 10 parts per million accuracy - that means a 10 PPM shift is clearly audible.”2020-11-11 Head-Fi, Nov 2020

“…has been very carefully tweaked using thousands of listening tests. The adjustments are on 5 variables, and each one is fine tuned to 10 parts per million…”2023-07-01 Head-Fi, Jul 2023

Dated positions

Caveats and open points

• The 10 ppm figures in talks, and the 256 versus 2,000 or 2,048 tap comparison in talks, come; readers should check the recordings for exact figures. Head-Fi posts carry the same figures in writing (Head-Fi, Nov 2020, Head-Fi, Jul 2023 for 10 ppm; Head-Fi, Jun 2015, Head-Fi, Feb 2017 for 256 versus 2048).
• His stated start date for WTA work varies between 1998, the late 1990s and 1999; his theoretical interest in sinc interpolation he dates to the 1980s or 1981.
• In 2020 he said one variable is set to 10 ppm; later statements apply 10 ppm to each of the five variables.
• The Weekly ASCII article is Japanese with a machine translation and is cited only for the name expansion.
• The WTA window and equation are unpublished, so no statement here can be checked against the coefficients themselves.

Topics: WTA filter algorithm and coefficient design, Listening test methodology