How can upscaling to 705.6/768 kHz improve sound if the recording has nothing above 22 kHz?

His core answer: no new information, only better reconstruction (2014 to 2022)

In July 2014, writing about the original Hugo, Watts said "the interpolation is not trying to re-create information that is not already present at the point of sampling", and that an infinite tap length filter would reconstruct the original bandwidth limited signal perfectly, as if no sampling had taken place, while finite filters create timing errors in the reconstruction that he called significant Head-Fi, Jul 2014. He added that such a filter "absolutely will not recreate the signal before the brick wall input filter", that his listening tests had shown linear phase brick-wall filters to be completely transparent if done correctly, and that the sound quality problem was the interpolation reconstruction filter, not the initial brick wall filter Head-Fi, Jul 2014.

In May 2016 he illustrated what he means by transient timing: a bandwidth limited signal crossing zero between two 44.1 kHz samples will be reconstructed crossing at exactly the right moment by an ideal sinc filter, but at a slightly wrong moment by a finite, non-sinc filter, and "it is these differences in the timing of transients, are what I am talking about" Head-Fi, May 2016. He said in the same post that ringing is not what he means by timing accuracy Head-Fi, May 2016.

While developing the M Scaler he applied the same argument to 705.6/768 kHz upscaling. In April 2017 he wrote that "with audio we do not need to reproduce HF content above 20 kHz simply because we can't hear it", but that the moment a signal changes state must be accurately reproduced, "and this is where the interpolation filter fails, as it creates timing uncertainty" Head-Fi, Apr 2017. In August 2018, on the Hugo M Scaler thread, he wrote that "converting the 44.1 kHz to 705.6 kHz is not adding any more information" and "It's not making things up, but reproducing the original analogue signal that was in the ADC before it was sampled." Head-Fi, Aug 2018. He also said there that the M Scaler does not sample rate convert, so 44.1 becomes 705.6 and 48 becomes 768 Head-Fi, Aug 2018.

In September 2018 he put it as "So upsampling, or interpolation, is not about creating more information, but about eliminating information loss or errors, or preserving the information contained in the samples." Head-Fi, Sep 2018. In July 2019 he conceded that this is hard to see: "It's hard to see that, as we are filling in the missing bits, but it's because the signal is bandwidth limited - so there is no extra information above FS/2." Head-Fi, Jul 2019. In December 2019 he said every DAC upsamples, so the real question is how accurately it reconstructs the original analog signal Head-Fi, Dec 2019; in the Passion for Sound interview (published in parts December 2019 to January 2020; recording date not stated) he said "you're not creating new information", adding that this assumes bandwidth limiting on its own makes no difference to sound quality Passion for Sound 21:30. In May 2022 he wrote that decimating a perfectly bandwidth limited signal introduces no change to its information content, and that "interpolation back to the original with an ideal sinc function will not create new information but will perfectly preserve the information content" Head-Fi, May 2022.

Why he says the sinc reconstruction is audible

He has tied the audible benefit to transient timing, not to frequency content. In August 2018 he said conventional filters can make a transient too early or too late, and that this matters because transients are used by the brain to perceive pitch, timbre, soundstage and the starting and stopping of notes Head-Fi, Aug 2018. In October 2018 he said the timing uncertainty depends on the program material, so "the actual music effectively randomly modulates the timing of the transients" Head-Fi, Oct 2018. In June 2022, on the DAVE thread, he wrote that "nobody can hear above 20kHz, but everybody can hear transients constantly shifting backwards and forwards in time" Head-Fi, Jun 2022. In part 4 of the same interview series (published January 2020; recording date not stated), asked why higher sampling rates matter if we hear only to 20 kHz, he said "it's not about perceiving the sine waves. It's about the timing." Passion for Sound 1:03.

He has also said that a steady tone does not show the difference. In September 2018 he wrote that a simple sine wave would be recreated perfectly irrespective of the type of filter, so long as the filter had infinite stop band, and that errors appear only with transients or non-steady-state signals Head-Fi, Sep 2018. In April 2019 he said a regular 1 kHz square wave would show the same effect through the M Scaler as through DAVE, because "a square wave is not a transient", and "You need transients, which will have energy up to 22.05k to be able to see the change." Head-Fi, Apr 2019.

Why 705.6/768 kHz in particular (2017 to 2024)

He has framed higher output rates as reducing the innate timing error. In December 2017 he wrote that doubling the sample rate only improves the area of the timing error by a factor of four (2 bits), while the M Scaler against a conventional filter is about a 16 bit improvement Head-Fi, Dec 2017. He repeated the factor of four per doubling in June 2019 Head-Fi, Jun 2019, in a podcast published in June 2020 (recording date not stated) Part-Time Audiophile 39:38, in September 2021, where he said a 768 kHz signal has 256 times less transient error by area than a 48 kHz one Head-Fi, Sep 2021, and at a Mojo 2 seminar at CanJam NYC in late February 2022 (video published March 2022), where he called 16FS or 705 kHz "the sweet spot" MFG Audio Reviews 52:02.

In January 2019 he said the "whole purpose of the M scaler is to convert your CD into a format that was identical to an original 705 kHz recording", to better than 16 bits, and "the only difference to a true 705kHz is the bandwidth limiting, which I do not think makes any real difference to SQ" Head-Fi, Jan 2019. Two days later, answering a poster's question, he wrote "it's about how accurate the timing of transients actually are, not being able to hear 768k" Head-Fi, Jan 2019. In July 2019 he said the 768k output gives a sample every 1.3 uS, "which of course is much easier for the brain to deal with", and that his 768k recordings sound very different to 384k ones, suggesting "there is something special about hitting 768k" Head-Fi, Jul 2019. In June 2019 he said you can clearly hear the M Scaler changing from 384 to 768 Head-Fi, Jun 2019. In January 2024 he wrote that "the 16FS is indeed the sweet spot for the interpolation stage", weighing 16FS's 1.3uS timing resolution against the filter seeing fewer input samples Head-Fi, Jan 2024.

His view of how fine the timing must be has shifted, by his own account. In May 2014 he cited the 4 microsecond interaural figure Head-Fi, May 2014, and in July 2014 he said he had no idea what timing errors are significant, "whether its 4uS or 4nS" Head-Fi, Jul 2014; in May 2016 he wrote "now I know that instead of worrying about uS we need to worry about getting it correct to nS's" Head-Fi, May 2016; in January 2019 he wrote "I now know for certain that 4uS resolution is not good enough." Head-Fi, Jan 2019.

Content above 20 kHz: what he says does and does not matter

His statements on ultrasonic musical content point the same way in every item reviewed from 2017 to 2022. From 2017 to 2021 he qualified each one. In February 2017 he said his answer on ultrasonics and realism was conjecture, but that he did not subscribe to the idea that ultrasonics are important to realism, adding that he could be wrong Head-Fi, Feb 2017. In January 2019: "Now I do not think any harmonics above 20k are audible at all; the benefit of HD recordings from the SR POV is the reduction in timing uncertainty" Head-Fi, Jan 2019. In June 2019: "There is no real evidence that anything above 20 kHz is audible", while noting that ultrasonic components can create audible intermodulation in air that a microphone would capture anyway Head-Fi, Jun 2019. In June 2021: "at this stage there is zero evidence that we need signals above 20 kHz - but what we do need are transients that have the same timing as the original, with zero uncertainty", and "HD recordings help that process, but are not innately essential to do this." Head-Fi, Jun 2021. The June 2022 "nobody can hear above 20kHz" line quoted above carries no such qualifier Head-Fi, Jun 2022.

He separates this from out-of-band artifacts, which he says can become audible indirectly. In June 2019 he said an out-of-band artifact can adjust the timing of transients, and out-of-band noise can create noise floor modulation in the DAC's analog stages Head-Fi, Jun 2019. In March 2022, on the TT2 thread, he wrote that out of band signals "are not in themselves audible, but can become audible because they can degrade transient timing accuracy, or because they create noise floor modulation in the analogue stages" Head-Fi, Mar 2022. He made the same distinction in February 2022 Head-Fi, Feb 2022.

The hedge: is bandwidth limiting itself harmless?

From 2017 to 2021 he treated the harmlessness of proper bandwidth limiting as unproven. In April 2017 he called CD perfect "at least that's my assumption today", and said his Davina 768 kHz ADC project would let him test it by filtering a 768 kHz recording to 24 kHz without decimating; "Of course, I may be wrong about my assumptions here, but until you do a carefully controlled listening test you know nothing for certain" Head-Fi, Apr 2017. In February 2017 he raised the possibility that the badly non-linear ear might produce audible products from ultrasonics, hence his interest in recording flat to 50 kHz Head-Fi, Feb 2017. In February 2018 he said he suspected bandwidth limiting would not create sound quality issues, "but nobody knows for sure" Head-Fi, Feb 2018. In April 2019 he said different bandwidth limiting filters that all meet sampling theory's requirement "may end up sounding very different of course", hence his planned listening tests Head-Fi, Apr 2019. In June 2019 he wrote "my view is that properly bandwidth limiting to 20kHz would be perfect SQ wise. But what do I know?" and said only rigorous tests would settle it Head-Fi, Jun 2019. In June 2021 he said initial listening tests of his ADC bandwidth limiting filters showed no evidence of any loss of transparency, while calling his conclusions tentative Head-Fi, Jun 2021.

Later posts report listening tests in which bandwidth limiting filters improved the sound, which he attributes to removing high frequency noise and distortion. In May 2022 he said his FIR decimation filters, auditioned as plain low pass filters, "have only improved sound quality" Head-Fi, May 2022. In June 2024 he wrote that "the SQ benefits of bandwidth limiting is due entirely from the elimination of HF components" Head-Fi, Jun 2024. In March 2025 he wrote that "for the best sound quality, you need to very aggressively remove HF content from the source (particularly with HD recordings)" Head-Fi, Mar 2025. In a podcast and a Soundnews conversation published in October 2026, he described a turntable digitized through Quartet's built-in ADC as sounding better, which surprised him, and credited the 18 kHz filter used in that ADC's decimation, which he said improves transient timing by removing out-of-band noise Distorsiuni Interioare 1:12:51Distorsiuni Interioare 1:14:52article, Oct 2026. These reports concern removing ultrasonic noise from ADC and high resolution sources; they are not the 768 kHz against band-limited-and-reconstructed comparison he proposed in 2017.

That comparison was still pending in the latest items. Davina itself was scrapped when components were discontinued, and the work continued as Quartet's pulse array ADC article, Oct 2026. In a Soundnews interview recorded in autumn 2026 and published in October 2026 he said "We've now got the ability to take those 768 kHz recordings, decimate them down to 48 kHz, bring them back up to 768 kHz again, and then compare and see what the losses are in that process." article, Oct 2026. In a podcast published in October 2026 (recording date not stated) he still described that comparison as something "we'll be able to" do Part-Time Audiophile 1:03:53.

Limits he puts on the benefit

He says the reconstruction assumes the recording was properly bandwidth limited before sampling. In October 2018 he wrote that the M Scaler's approach "requires that the signal be bandwidth limited before sampling", and that with modern half band ADC decimation filters "once aliasing happens it is permanently damaged" Head-Fi, Oct 2018. He said in the same post that all recordings benefit, including wholly digital electronica Head-Fi, Oct 2018. In September 2021 he wrote that the reconstruction problem applies to perfectly sampled recordings as well as aliased ones, and that he never found a different WTA sweet spot with different recordings Head-Fi, Sep 2021.

Quartet M Scaler and later interviews (2024 to 2026)

Interviews from Quartet's May 2024 prototype showings onward repeat the same account. In May 2024, after an interviewer said "bandwidth", he said "No, it's not about bandwidth. No, no, it's about timing accuracy." Reactcore 24. In another May 2024 interview he said "we're not creating new information" and "We're maintaining the information content that was originally there." The Speaker Shack 2:51The Speaker Shack 2:56. In a third he said that with limited processing transients are "constantly moving backwards and forwards", which upsets the brain ConvinceMeAudio 6:51. An interview at the High End show in Munich (May 2024), published by Hifistatement in June 2024 in German as the publisher's translation of his answers, carries the same point that interpolation concerns the missing waveform between samples rather than missing information article, Jun 2024.

In a podcast recorded in October 2025 and published in March 2026, asked whether this was like upscaling a TV picture, he answered "In a simple sense, yes, but what's actually happening is you've got an information content on your 16 bit C D", then said a regular interpolation filter distorts the information content of a CD, while his aim is that "you completely preserve the information content", adding that "the information content is the thing that's important is your timing of your transients" and that "that timing information is embedded in the sampled data" ecoustics 15:16. He did not name a product in that answer.

What he has not said: He does not claim upscaling restores or invents content above 22 kHz; from 2014 to 2024 he said the opposite Head-Fi, Jul 2014Head-Fi, Jul 2019Head-Fi, May 2022The Speaker Shack 2:51. No result of the 768 kHz decimate-and-reconstruct comparison appears, through October 2026 article, Oct 2026Part-Time Audiophile 1:03:53.

Quotes

“converting the 44.1 kHz to 705.6 kHz is not adding any more information”2018-08-02 Head-Fi, Aug 2018

“the whole purpose of the M scaler is to convert your CD into a format that was identical to an original 705 kHz recording”2019-01-27 Head-Fi, Jan 2019

“nobody can hear above 20kHz, but everybody can hear transients constantly shifting backwards and forwards in time”2022-06-22 Head-Fi, Jun 2022

“No, it's not about bandwidth. No, no, it's about timing accuracy.”2024-05-10 Reactcore 24

“interpolation back to the original with an ideal sinc function will not create new information but will perfectly preserve the information content”2022-05-10 Head-Fi, May 2022

Dated positions

Caveats and open points

• Most statements concern the Hugo M Scaler and the WTA filters in his DACs; the 2014 and 2016 statements were made about Hugo and DAVE, and the Quartet statements come from 2024 to 2026 interviews.
• The June 2024 Hifistatement interview is the publisher's German translation of his English answers, shown here via machine translation, so it is paraphrased rather than quoted.
• The Soundnews piece is an edited transcript of a conversation from autumn 2026, published October 2026; the author's April factory visit, described in its introduction, was a listening session.
• His 2022 to 2026 reports that bandwidth limiting improves sound concern removing out-of-band noise from ADC and high resolution sources; they do not settle the question he posed in 2017 of whether a perfectly band-limited, reconstructed CD-rate signal matches a native 768 kHz recording, for which no result appears.
• His claims that transient timing errors are audible rest on his own listening tests, as he states; he also says aliasing already in a recording cannot be repaired by the M Scaler.
• Some Head-Fi posts carry [...] scraper gaps; no quotation here spans one.

Topics: Why upscale, and why a separate scaler, Transient timing error from reconstruction, Sample rates, hi-res and bit depth