How are Chord's filter stages split across 16FS, 256FS and 2048FS, and what does each stage do?

Short version

From 2017 on, for Hugo 2, Hugo TT 2, Qutest and DAVE, Watts describes three stages by the rate they work at. Stage 1 (WTA1) is a long WTA FIR filter that takes the signal up to 16FS (705.6 or 768 kHz). Stage 2 (WTA2) is a second WTA filter from 16FS to 256FS. The last stage is a three-stage filter from 256FS (or from 16FS where WTA2 is absent or switched out) to 2048FS, about 104 MHz, which feeds the pulse array noise shaper Head-Fi, Mar 2017Head-Fi, Feb 2018Head-Fi, Aug 2018Head-Fi, Oct 2018Head-Fi, Jul 2022HiFi Buys 1:15:13. The original Hugo and Mojo were built differently, and Mojo 2 has no WTA2. What follows is organized by period and product, then by what he says each stage is for.

2014 to 2016: Hugo, Mojo and the first DAVE

In April 2014 he wrote "Hugo has 5 filters - 2 FIR WTA stages": the first (26k taps) and a separate WTA stage to 16FS, then a linear interpolator from 16FS to 2048FS, then two IIR low-pass filters at 2048FS Head-Fi, Apr 2014. Two days earlier he had summed up the last part as "it filters with a 3 stage filter at 2048 FS" Head-Fi, Apr 2014. In January 2015 he said Hugo's volume control sits in the central WTA filter core, after which the signal goes to the cross-feed DSP and then "the 3 stage interpolation filters to take it to 2048FS" Head-Fi, Jan 2015. In March 2015 he gave the Hugo rates as 1FS to 8FS with the 26,368-tap WTA, then "8FS to 16FS WTA stage two which also does volume", then 16FS to 2048FS by a linear interpolator and two low-pass filters Head-Fi, Mar 2015.

At the Munich show in May 2015 (so check the recording) he said of DAVE "it's filtering up to 256 times on the FIR filter, and then from 256 times it's then further filtered up to 2048", with 164,000 taps HIFICLUBE.NET JVH 45. In November 2015 he wrote that "now I am WTA FIR filtering to 256 FS" Head-Fi, Nov 2015. In December 2015 he said DAVE's HF filter "is not something added, its just a different set of coefficients for the 16 FS to 256 FS WTA filter" Head-Fi, Dec 2015; for Hugo 2, TT2 and Qutest he later placed the HF filter in the 2048FS stage instead (see below).

In May 2016 he described the Hugo arrangement as a 16FS WTA filter followed by a linear interpolator and a two-stage IIR filter up to 2048FS, and said that changing it, for DAVE, to a 256FS WTA filter followed by "my usual 3 stage filtering" gave a massive change in sound quality Head-Fi, May 2016. In the Hifistatement interview (recorded about January 2016, published July 2016) he said Hugo was 8 times oversampled on the first stage with a second stage taking it to 16 times, that replacing this with a single-stage 16 times WTA filter "sounded a lot better", and that he then added a second stage from 16 times to 256 times WTA filtering Hifistatement 6:51Hifistatement 7:47.

In October 2016 he set the products side by side: "Hugo/Mojo is: WTA 16FS > Linear interpolation filter to 2048 FS > Low pass Filter > Low pass filter > noise shaper", while "Dave is: WTA 16FS > WTA to 256FS >Linear interpolation filter to 2048 FS > Low pass Filter > Low pass filter > noise shaper". He added that the low-pass filter is set to 260 kHz and is mathematically the same as a simple RC low-pass filter Head-Fi, Oct 2016.

2017 to 2019: Hugo 2, Hugo TT 2, Qutest

In January 2017 he said the decision to split the job over WTA1 (to 16FS), WTA2 (to 256FS) and a non-WTA third-order filter to 2048FS "was based upon listening tests to see which arrangement sounds the best" Head-Fi, Jan 2017. That month he also said Hugo 1 had two WTA filters (to 8FS, then 8FS to 16FS), whereas with Dave "it sounded better if I used a single WTA filter going from 1FS to 16FS", and that cascading filters does not give the same time-domain accuracy as a single longer filter Head-Fi, Jan 2017.

For Hugo 2 (March 2017 slides) the output of WTA 1 is passed to WTA 2, which goes from 16FS to 256FS, "so now the WTA is working to a resolution of 88 nS" Head-Fi, Mar 2017. In July 2017 he restated that Hugo's WTA 1 went 1FS to 8FS at 26,368 taps and WTA 2 went 8FS to 16FS, while Hugo 2's WTA 1 goes from 1FS to 16FS in a single filter of 49,152 taps and WTA 2 goes 16FS to 256FS Head-Fi, Jul 2017. The tap counts he quotes apply to WTA 1 only Head-Fi, Mar 2017. In July 2017 he said both Hugo 2 filter options filter up to 2048FS, with the 16FS option using only IIR filters and the 256FS option replacing "the 16>256 IIR filter with a WTA filter" Head-Fi, Jul 2017. In September 2017 he said the 2048FS stage is "actually a 3 stage filter running at 2048 FS", linear interpolation first and two IIR low-pass filters after it Head-Fi, Sep 2017.

In February 2018 he gave the Hugo 2 and Qutest path as 1FS to 8FS input, a 16FS WTA1 filter (49,152 taps), a 256FS WTA 2 filter, a 3rd order 2048 FS filter, then the pulse array noise shaper at 104 MHz; "When an M scaler is connected, the WTA1 filter is not used" and the signal passes to WTA 2 Head-Fi, Feb 2018. In August 2018 he gave Hugo TT 2's WTA 1 as 98,304 taps and said "The WTA 2 filter (16FS to 256 FS) is identical to Hugo 2 and Dave" Head-Fi, Aug 2018. Of the chain from 16FS onward he said "the code is identical for Qutest and up to Dave", the main differences being the noise shaper, the pulse array and the analog parts Head-Fi, Aug 2018. In late October 2018 (date estimated from the post ID) he said "All the current DAC's except mojo uses the identical WTA2 filter" and "WTA 3 does not exist as a WTA filter, but a three stage IIR filter", which filters to 9.6 nS ahead of the pulse array Head-Fi, Oct 2018.

In January 2019 he said that at 705/768 kHz input WTA1 is in pass-through mode; after WTA1 there are two choices, WTA2 (a linear-phase symmetrical WTA filter from 16FS to 256FS) or an IIR filter, and the 3-stage IIR filter either starts at 256FS (white or green) or takes over at 16FS (orange or red) Head-Fi, Jan 2019. That day he also said DAVE is the same but with "no option to lose WTA2" Head-Fi, Jan 2019.

2022 to 2026

In January 2022 he said the UHD DSP works at 705/768 kHz from WTA1's output or the DSD filter, and "There is no WTA2" Head-Fi, Jan 2022; in the Mojo 2 thread that day he wrote "remember no WTA 2" for Mojo 2 Head-Fi, Jan 2022. In another reply that day, comparing Mojo 2 with Hugo 2, he wrote "But the filtering from 705/768 (16FS) to 2048FS has been improved"; the archived post does not say what it is compared with Head-Fi, Jan 2022.

In June 2022 he wrote that on TT2, Hugo 2 and Qutest the HF filter is done in the filter after WTA2 (256FS to 2048FS), a multi-stage IIR filter whose time constants are adjusted, "So WTA2 always stays the same" Head-Fi, Jun 2022. In July 2022 he described DAVE as WTA1 (the "prime" WTA filter, to 16Fs), WTA2, which he called "the same as WTA1 but works from 16Fs to 256Fs", then a final three-stage IIR filter from 256Fs to 104.25 MHz, and added "each filter has very different requirements from the sound quality POV" Head-Fi, Jul 2022. In August 2022 he said orange and red "bypass the WTA2 filter" and the 256FS to 2048FS filters then run from 16FS, otherwise identical, so they are "definitely not bypassed" Head-Fi, Aug 2022. A January 2025 reply in the Qutest thread restates that red and orange switch WTA2 out and that white and green adjust the time constants of the 256FS to 2048FS filter Head-Fi, Jan 2025.

In May 2024 (in a Quartet Scaler explainer) he said "the sweet point is 16FS" and that DAVE "does another WTA filter to 256FS" and then has another filter stage to 2048FS Reactcore 1:00Reactcore 1:09Reactcore 1:13. In a podcast recorded in October 2025 and published in March 2026 he said DAVE's interpolation filter converts 48 kHz to 768 kHz, 16 times oversampled, and DAVE then upsamples that to 2048 times ecoustics 23:40. In an interview published October 2026 (recording date not stated, though the transcript places it before May 2026) he said that within all of the DACs "apart from Mojo 2" there is another WTA filter from 16FS to 256FS, then another set of filters from 12 MHz to 104 MHz, which are different filters doing different things HiFi Buys 1:15:13. Check the recordings for exact figures.

What he says each stage is for

Stage 1, to 16FS (WTA1). He says it is where the big tonal changes come from: "WTA1 we get big changes in bass, pitch, timbre and instrument separation - but WTA 2 only affects the perception of starting and stopping of notes" (July 2019) Head-Fi, Jul 2019. In a 2022 post (date estimated from the post ID) he said WTA1 "changes everything - timbre, pitch, soundstage, tempo and instrument separation and focus" Head-Fi, Mar 2022. On why 16FS, he wrote in January 2024 that "the 16FS is indeed the sweet spot for the interpolation stage", because lowering the output rate lets the filter see more input samples but leaves coarser output spacing (2.6 uS at 8FS for 48 kHz input), and "All the subjective evidence I have is that 16FS followed by appropriate extra filtering is the optimum approach" Head-Fi, Jan 2024.

Stage 2, 16FS to 256FS (WTA2). He says it improves timing: output spacing falls from 1.6 uS at 16FS to 88 nS at 256FS (March 2019) Head-Fi, Mar 2019. He was expecting little subjective change when he designed it, and was surprised that it changed only the perception of starting and stopping of notes: "I had expected that if WTA 2 sounded better, it would be like a mini WTA 1, but it doesn't sound like that at all" Head-Fi, Mar 2022. In 2022 he said "This is all about the transient timing accuracy" and that it has nothing to do with the in-band response Head-Fi, Mar 2022. He said ordinary measurements do not show it: "You won't measure any change from the 16FS to 256FS WTA filter" (December 2017, Hugo 2) Head-Fi, Dec 2017, and in 2020 that with sine waves there is "zero measured difference" between orange and white Head-Fi, May 2020. In June 2019 he said it "replaced a perfectly good third order IIR type filter with a WTA filter, and I was surprised how easy it was to hear the difference" Head-Fi, Jun 2019. In July 2018, for Hugo 2, TT2 and Qutest (DAVE always has it on), he said that with the filter off "it gets replaced with an IIR filter" and sounds warmer and softer with the timing imprecision Head-Fi, Jul 2018, and in 2023 that "switching out WTA2 degrades transient timing" Head-Fi, Sep 2023. At CanJam NYC 2022 (video published March 2022) he said WTA2 is much less sensitive to tap length than WTA1 MFG Audio Reviews 52:02. He said he added the orange and red options so that people could hear what WTA 2 does against the 3 stage IIR filters Head-Fi, Oct 2018Head-Fi, Sep 2023.

Stage 3, to 2048FS. He describes a three-stage filter (linear interpolation, then two IIR low-pass stages) that is mathematically the same as an analog low-pass filter Head-Fi, Sep 2017Head-Fi, Oct 2018. In August 2014, for Hugo, he gave a reason for digitally filtering up to 2048FS: "To make the analogue filtering easier, then the problem is transferred to the digital domain" Head-Fi, Aug 2014. In December 2017 he gave removing images as one of several reasons for filtering to 2048FS, "to remove all of these images" Head-Fi, Dec 2017. In June 2019 he said the HF filter works by changing this filter's time constant, and that "This filter attenuates the images above 12 MHz to 104 MHz and has a big impact on noise floor modulation"; he offered this as his likeliest of three explanations for why green sounds warmer with CD Head-Fi, Jun 2019. In the eCoustics podcast recorded in October 2025 (published March 2026) he gave the limit on running faster than about 100 MHz as hardware: "the sweet spot has been around about 100 megahertz" ecoustics 9:37.

Figures and descriptions that differ

He names the last stage in varying terms: a linear interpolator plus two low-pass or IIR filters Head-Fi, Apr 2014Head-Fi, Oct 2016Head-Fi, Sep 2017, "third order IIR filters" Head-Fi, Jan 2017, a "3rd order 2048 FS filter" Head-Fi, Feb 2018 and a "three stage IIR filter" Head-Fi, Oct 2018Head-Fi, Jul 2022. His stated corner for it varied: a 260 kHz low-pass for Hugo and Mojo in 2016 Head-Fi, Oct 2016, time constants from 0.7 uS (227 kHz) to 2.8 uS (57 kHz) in 2017 Head-Fi, Jul 2017, "about 240 kHz or 60 kHz" in January 2019 Head-Fi, Jan 2019, a -3 dB point of 150 kHz changing to 37.5 kHz in February 2019 Head-Fi, Feb 2019, and 160 kHz to 40 kHz in June 2019 Head-Fi, Jun 2019. He gave the 16FS output spacing as 1.4 uS in 2015 Head-Fi, Nov 2015, 1,417 nS in 2016 Head-Fi, May 2016, 1.6 uS in March 2019 Head-Fi, Mar 2019, and 1.3 uS in June 2019 and in 2024, the 2024 figure stated for 48 kHz input Head-Fi, Jun 2019Head-Fi, Jan 2024. Qutest's WTA1 was given as 49,152 taps in 2018 with Hugo 2 and as "48k taps" in 2022 Head-Fi, Feb 2018Head-Fi, May 2022; the January 2025 Qutest-thread reply gives WTA1 as 98,304 taps without naming the product in the archived text Head-Fi, Jan 2025. These posts do not explain the differences.

What he has not said

He gives no stage-by-stage description of the original Mojo beyond the 2016 line and the 2018 remark that it lacks the shared WTA2 Head-Fi, Oct 2016Head-Fi, Oct 2018. In January 2024 he said "what I talk about is simplified", partly to protect his IP Head-Fi, Jan 2024, and in March 2025 he called the whole reconstruction issue to 104 MHz "very much more complex than I have talked about so far", saying he would discuss it further in the future Head-Fi, Mar 2025.

Quotes

“WTA1 we get big changes in bass, pitch, timbre and instrument separation - but WTA 2 only affects the perception of starting and stopping of notes”2019-07-23 Head-Fi, Jul 2019

“it's actually a 3 stage filter running at 2048 FS. The first stage is linear interpolation, the second two stages are IIR low pass filters”2017-09-14 Head-Fi, Sep 2017

“WTA 2 then improves the time resolution to 256FS or 88 nS.”2018-10-31 Head-Fi, Oct 2018

“All the subjective evidence I have is that 16FS followed by appropriate extra filtering is the optimum approach.”2024-01-07 Head-Fi, Jan 2024

Dated positions

Caveats and open points

• Statements are Head-Fi replies and interview answers, often partial; the stage descriptions differ in detail and naming between posts and products, and the archive does not reconcile them.
• His figures for the 2048FS stage's corner frequency, the 16FS output spacing (1.3, 1.4, 1.417 or 1.6 uS) and Qutest's WTA1 tap count differ by post.
• Dates of Head-Fi, Oct 2018, Head-Fi, Mar 2022 and Head-Fi, Mar 2022 are estimated from post IDs.
• Head-Fi, Jan 2025 (January 2025, Qutest thread) gives WTA1 as 98,304 taps without naming the product in the archived text; it is cited for the filter settings and the tap figure with that hedge.
• His 2015 description of DAVE's HF filter (WTA2 coefficients) differs from his 2017 to 2025 descriptions for Hugo 2, TT2 and Qutest (2048FS-stage time constants); these are statements about different products.
• The reasons he gives for each stage are his listening-test conclusions and design rationale, reported here without assessment.
• Quartet and M scaler internals, and the Mojo 1 filter details, were not examined beyond the items cited.

Topics: Filter stage architecture and oversampling rates