Glossary
Terms as Rob Watts explains them, each with its sources.
-301 dB and 350 dB performance · Aliasing · Bit perfect · Crossfeed · Davina · DC servo · Decimation filter · Depth · Desktop mode · Dither · DoP (DSD over PCM) · DPLL · Dual data and dual BNC · DX link · FPGA · Galvanic isolation · HF filter · Jitter · Minimum phase filter · Noise floor modulation (NFM) · Noise shaper · NOS (non-oversampling) DAC · Oversampling (16FS, 256FS, 2048FS) · Pass-through · Pre-ringing · Pulse array · RF noise · Sinc function and ideal reconstruction · Small-signal accuracy · Starting and stopping of notes · Taps (tap length) · Timbre variation · Transient timing error · Truncation and re-quantization · Upscaling (M scaling) · WTA filter
-301 dB and 350 dB performance
Also called: -301 dB test, minus 301 dB signal, 350 dB noise shaper, -301 dBFS test tone
The minus 301 dB signal is Watts's pass test for his digital modules: in Verilog simulation a 6 kHz minus 301 dB sine goes through the module, and its amplitude must stay within +/- 0.001 dB, and phase within +/- 0.001 degrees, of 0 dB Head-Fi, Jun 2021Head-Fi, Apr 2022. In 2016 he said passing it "ensures its much better than 50 bits accuracy" Head-Fi, Mar 2016.
His reason is depth: signals below a noise shaper's noise floor are attenuated and lost, and small echo cues tell the brain a sound is further away Head-Fi, Sep 2016. He has hedged on why: he does not know for sure whether the brain perceives minus 301 dB signals, or whether a system resolving them simply resolves minus 120 dB signals better Head-Fi, Sep 2016. In a podcast recorded in October 2025 (published March 2026) he said the tone is "never going to be audible", but "a minus 301 dB error is perceivable" ecoustics 49:13.
The 350 dB figure rates a noise shaper's THD and noise in the audio band Head-Fi, Dec 2019; he says better than 350 dB gives audibly deeper depth than 330 dB Head-Fi, May 2016. He allowed that the need may be "simply a proxy for something else" in the analog section, but heard the same benefit with purely digital truncation shapers Head-Fi, May 2016.
“pulse array is innately capable of resolving a -301 dB signal perfectly, which is now my standard test for depth capability”2016-05-17 Head-Fi, May 2016
“a -350 dB noise shaper has easily noticeable improved depth perception”2017-11-17 Head-Fi, Nov 2017
See: What does Rob mean by depth, and what in a DAC improves or…, Do measurements tell you how good a DAC sounds, and is Rob an…, Should I replace the supplied power supply with a linear or…, Which BNC cables, and what length, should I use between an M…, Do digital cables (USB, coax, glass or plastic optical) make a…, Will a USB purifier, JitterBug or USB isolator improve my DAVE,…
Aliasing
Also called: alias, ADC aliasing
Watts says sampling theory requires no energy at and above FS/2, and "If it is not bandwidth limited, aliasing occurs" Head-Fi, May 2022. The result is "signal components that are at the wrong frequency and around 20-24 kHz" Head-Fi, Jun 2019. He blames modern ADCs, whose cheap half band filters are "a paltry -6dB at FS/2" Head-Fi, Jul 2018Head-Fi, Jun 2019. Once aliasing is in the recording, no interpolation filter can repair the damage Head-Fi, Jan 2018. He says it is audible even though components near 20-24 kHz are not audible by themselves, because it changes the timing of transients Head-Fi, Jun 2019. He also reports hearing it in a DAC's DSD filter: in 2018 he said aliasing at -150dB in Hugo 1's filter was "surprisingly audible", and the better filter has aliasing better than -240dB Head-Fi, Jan 2018. For the decimation filters of his own ADC project his rejection figures range from 220 dB to 260 dB, one of them from a Verilog simulation Head-Fi, Jun 2019Head-Fi, Sep 2018Head-Fi, Oct 2017. In October 2026 he said Quartet's pulse array ADC eliminates the high frequency aliasing of standard pro ADCs, tying it to transient timing and to the pitch of female vocalists in his test recordings Distorsiuni Interioare 12:51Distorsiuni Interioare 13:35Part-Time Audiophile 58:34.
“modern ADC's are not effective at bandwidth limiting, being a paltry -6dB at FS/2. So we will get aliasing, and no filtering can eliminate that damage.”2018-07-19 Head-Fi, Jul 2018
“aliasing is a major issue - and current ADC's have huge levels built into the ADC performance”2018-01-03 Head-Fi, Jan 2018
See: Do Chord DACs decode MQA, and what does Rob think of it?, How do Chord DACs handle DSD, and does Rob consider DSD better or…, How can upscaling to 705.6/768 kHz improve sound if the recording…, Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…
Bit perfect
Also called: bit-perfect, bit perfect data
For Watts, bit perfect data is data the source has not upsampled or converted in any way, and it is all a DAC needs from a source. Head-Fi, Apr 2018Head-Fi, Nov 2021 He said even a single stage of gain scaling makes a big difference, as data not properly handled when adjusted loses depth. Head-Fi, Mar 2016 He used DoP DSD as a test: if it plays mute free, the path is bit perfect. Head-Fi, Mar 2017 He did not claim bit perfect sources all sound alike, only that a perfect DAC would not care which one it got. Head-Fi, Nov 2021
He distrusts the term as a goal. In 2019 he said he intensely objects to the term and to its use as a holy grail and prefers signal perfect, meaning faithful to the original continuous analog signal in the ADC. Head-Fi, Mar 2019 In a 2017 talk, he said a DAC is trying to reproduce that analog signal, not the data. miceblue425 11:08 A volume control can never be bit perfect unless set to unity, he said, so the useful question is whether it changes the sound. Head-Fi, Oct 2018 He said Hugo M Scaler pass-through and, in a Mojo 2 podcast, EQ are not bit perfect either. Head-Fi, Oct 2019Darko.Audio 19:43
“You need to ensure that the data is bit perfect, that means the source does not upsample nor convert the data in any way.”2018-04-25 Head-Fi, Apr 2018
“I much prefer the concept of signal perfect, that is exactly faithful to the original continuous analogue signal in the ADC…”2019-03-29 Head-Fi, Mar 2019
See: How do I make sure my player sends bit-perfect data at the native…, If the data reaching the DAC is bit perfect, can the source…, Is M Scaler pass-through bit perfect, and why is the scaled output…, Should I upsample in software (HQPlayer, PGGB or my player) before…
Crossfeed
Also called: cross-feed, cross feed
Watts describes crossfeed on headphones as an attempt to make them sound more like loudspeakers. Head-Fi, Apr 2021 It is meant to give an out of your head experience, with the sound coming from outside with depth. Head-Fi, Sep 2017 He said Dave's is a digital version of conventional 1980s analog crossfeed, mathematically identical to it. Head-Fi, Aug 2016 It includes EQ for the head's shadowing: a bass boost he gave in April 2017 Dave thread posts as 1.5 dB, then by setting as about 1, 1.5 and 1.9 dB for settings 1 to 3, and for Hugo 2 in October 2017 as 1.8 dB on full, with only 1 dB at red. Head-Fi, Apr 2017Head-Fi, Apr 2017Head-Fi, Oct 2017 On Dave, as on Hugo, settings run from 0 (off) to 3 (maximum). Head-Fi, Feb 2016
He said there is no right or wrong setting. Head-Fi, Nov 2017 He uses maximum on headphones and the minimum on small speakers that sit too wide apart. Head-Fi, Apr 2021 On Hugo 1 he found it useful mainly for hard-panned older recordings. Head-Fi, Sep 2017 With Hugo 2 he called it essential, because depth is much better. Head-Fi, Dec 2020 The original Mojo has no crossfeed, while Mojo 2 runs it in its DSP core. Head-Fi, Dec 2020Head-Fi, Feb 2022
“Crossfeed is supposed to give one that out of your head experience (with the sound coming from outside with depth)”2017-09-19 Head-Fi, Sep 2017
“Agreed - cross-feed has no right or wrong setting, that's why it's switchable.”2017-11-01 Head-Fi, Nov 2017
See: What does crossfeed do, which setting should I use, and why didn't…, What does Rob mean by depth, and what in a DAC improves or…
Davina
Also called: Davina project, pulse array ADC, Davina ADC
Davina is Watts's code word for his ADC project, notionally intended for Dave and the first product to use his pulse array ADC. Head-Fi, Apr 2016 In 2016 he said it targeted what conventional ADCs do badly: noise floor modulation, aliasing from half band decimation filters, and noise shaper resolution. Head-Fi, Feb 2016 He called it a technology proof of concept more than a commercial product. Head-Fi, Jun 2016 He also expected it to show how much sound quality the M scaler loses. Head-Fi, Mar 2017 He said in 2017 that it was an 11th order 104 MHz noise shaper. Head-Fi, Jun 2017
It was slow. In 2021 he said Davina had stalled when a key part became unavailable. Head-Fi, Sep 2021 In 2026 he said some of its components had been discontinued by the time he tested it, so the design was scrapped, the part being a flash converter. article, Oct 2026 In the same interview he said Davina was started in 2016, though in 2016 he had said the first prototype of his ADC work dated from 2001. article, Oct 2026Head-Fi, Feb 2016 He said Chord wanted an analog input on Quartet, which meant he had to finish the ADC, and a pulse array ADC was built into Quartet. article, Oct 2026article, Oct 2026Part-Time Audiophile 50:30
“Davina is the code word for the ADC project notionally intended for Dave. It will be the first product to use my Pulse Array ADC.”2016-04-24 Head-Fi, Apr 2016
“Davina is really a technology proof of concept, as the pro audio market for a simple 2 ch. ADC is limited.”2016-06-12 Head-Fi, Jun 2016
See: How does a pulse array DAC work, how does it differ from R2R and…, What does Rob mean by depth, and what in a DAC improves or…, How can upscaling to 705.6/768 kHz improve sound if the recording…, Does hi-res audio sound better than CD, and should I buy the 96k…, Is the filter algorithm more important than the tap count, and…, Why does tap length matter, and why a million taps?
DC servo
Also called: digital DC servo, servo
Watts's digital DC servo removes DC offset from the output without a coupling capacitor in the audio path. Head-Fi, Aug 2018 A conventional analog servo swaps the coupling capacitor for an integrator capacitor, which he said still adds distortion and noise. In his version the integrator output goes to an ADC, the FPGA filters out everything except a very small, slowly moving trim, and that trim is added at the noise shaper input. Head-Fi, Jun 2016Head-Fi, Aug 2018 He said this keeps offset under 0.1 mV and that switching the servo on and off made no change in sound quality. Head-Fi, Aug 2018Head-Fi, Feb 2022 The cost is time: about 16 seconds to settle, which he gave as the reason for the rainbow mode on Qutest, Hugo 2 and TT2 and for Dave's turn-on delay. Head-Fi, Aug 2018 In 2016 he put Dave's figure at 20 seconds. Head-Fi, Jun 2016 He called TT2 safe with a DC coupled power amp. Head-Fi, Aug 2021
The original Mojo is the exception: it keeps a single coupling capacitor, because he said a digital servo would be too expensive to implement there. Head-Fi, Jul 2017 Mojo 2 gained the servo, which he said in a 2022 talk makes the bass tighter and the sound more neutral and leaner. MFG Audio Reviews 27:42
“…the OP from the analogue DC servo, the DC integrator, is fed to an ADC, then the FPGA removes all of the distortion and noise by filtering it all out”2018-08-28 Head-Fi, Aug 2018
“Indeed, switching the digital DC servo on and off resulted in no change in SQ at all.”2018-08-28 Head-Fi, Aug 2018
See: How does a pulse array DAC work, how does it differ from R2R and…, What do the colored indicators mean (volume, sample rate,…
Decimation filter
Also called: decimation, decimating filter, ADC decimation
Decimation, Watts says, takes an ADC from its very fast internal rate down to the output rate: you filter, then throw away the samples you do not want (podcast published October 2026, recording date not stated) Part-Time Audiophile 56:23. The filter must remove everything at and above half the output rate (24 kHz for 48 kHz); then you can drop samples safely, 15 of every 16 going from 768 kHz to 48 kHz Head-Fi, May 2022. He says it does not matter how the filtering is done, so an IIR filter is acceptable if attenuation is infinite at 24 kHz and above Head-Fi, May 2022. Interpolation in a DAC creates samples rather than discarding them Head-Fi, May 2022. He complains that modern ADCs use half band decimation filters, only -6dB at FS/2, so they alias Head-Fi, Jul 2018Head-Fi, Jan 2018. For his own ADC's decimation filters he gave rejection figures of 220 to 300 dB between 2017 and 2019, one from a Verilog simulation Head-Fi, Jun 2019Head-Fi, Jan 2018Head-Fi, Sep 2018Head-Fi, Oct 2017Head-Fi, Aug 2017. In May 2021 he said his -240 dB filters had to be redone because work on the ADC sample rate converter indicated that aliasing at -350 dB is audible Head-Fi, May 2021, and in July 2024 that he was still determining the limits required for subjectively perfect decimation Head-Fi, Jul 2024. In 2022 he said they had only improved the sound, and that listening tests as a plain low-pass filter suggested their pre-ringing is inconsequential Head-Fi, May 2022. In 2026 he described an 18 kHz filter in Quartet as part of its ADC's decimation article, Oct 2026.
“It does not matter how the filtering is done, all that matters is that the filter removes everything at 24k and above.”2022-05-10 Head-Fi, May 2022
“in essence you filter it And then you throw away the samples you don't want.”2026-10-14 Part-Time Audiophile 56:23
See: How do Chord DACs handle DSD, and does Rob consider DSD better or…, Does hi-res audio sound better than CD, and should I buy the 96k…, How can upscaling to 705.6/768 kHz improve sound if the recording…, Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…
Depth
Also called: soundstage depth, depth perception
By depth Watts means simply how far away a sound seems. He tests it with an organ in a cathedral, listening only for whether one setting sounds farther away than another. Head-Fi, Jul 2020 His benchmark is live sound: an organ 100 feet away sounds that far live, but reproduced it is at best a few feet back. Head-Fi, Nov 2015 He said depth perception is mostly down to the ratio of reverberant to direct sound, with air absorbing high frequencies another cue, and that small signal non-linearity is the only factor he had correlated with depth. Head-Fi, Dec 2015 That is why he pushed Dave's noise shapers from 200 dB to 350 dB, hearing deeper depth at each step. Head-Fi, Nov 2015 Since the M scaler he has added timing: a phase shift that changes with amplitude also degrades depth. Head-Fi, Apr 2022Head-Fi, Mar 2023 He said a wider soundstage normally tells him depth is truncated. Head-Fi, Sep 2018
In June 2017 he hedged: the noise shaper tests were not double blind, so he could not rule out confirmation bias, though repeated tests convinced him, and he planned a blind test. Head-Fi, Jun 2017 Later he reported tests "(some blind)" in April 2022 Head-Fi, Apr 2022, hearing the changes in a blind test at CanJam NYC in March 2024 MFG Audio Reviews 45:45, and blind listeners matching his depth result in a podcast recorded October 2025 (published March 2026) and an interview published October 2026 (recording date not stated). ecoustics 39:09HiFi Buys 1:22:59 These are his own reports; this archive holds no published test data.
“When I talk about depth it's just simply how far away it sounds”2020-07-09 Head-Fi, Jul 2020
“small signal non-linearity is the only factor that I have correlated with depth perception”2015-12-01 Head-Fi, Dec 2015
See: What does Rob mean by depth, and what in a DAC improves or…, What does crossfeed do, which setting should I use, and why didn't…, What is a transient timing error, and why is it audible when a…
Desktop mode
Also called: desktop mode (magenta)
Desktop mode is a charging state of Hugo 2 that starts when a charger has been connected without interruption for more than 24 hours. Head-Fi, Mar 2017Head-Fi, Sep 2018 The charger then supplies the current Hugo 2 draws, so no net current flows into the battery, and the battery voltage is held at 8.1 to 8.2 V, which he said maximizes battery life. Head-Fi, Mar 2017 The battery only covers brief current surges, and auto shutdown is disabled. Head-Fi, Jun 2017Head-Fi, Jun 2018 He said the lower voltage means the battery only ever charges to 80 percent of capacity; turning Hugo 2 off charges it fully and then disconnects the charger from the battery. Head-Fi, Nov 2020 In March 2020 he first advised turning it off when idle, then revised that to keep it on if used 8 or more hours a day, while calling the whole question splitting hairs. Head-Fi, Mar 2020Head-Fi, Mar 2020
Mojo 2 differs: in desktop mode its battery is completely disconnected, extra regulation supplies the unit, and a trickle charge tops up the battery every 11 days. Head-Fi, Feb 2024 The original Mojo has no desktop mode. Head-Fi, Mar 2020
“Desktop mode means that the battery voltage is maintained at a reduced level, to maximise battery life.”2017-12-16 Head-Fi, Dec 2017
“Mojo 2 in desktop mode, the battery is completely disconnected.”2024-02-19 Head-Fi, Feb 2024
See: How does desktop mode work on Hugo 2 and Mojo 2, and will Mojo 2…, Can I use Hugo, Mojo or Mojo 2 while it is charging, and is it OK…, Is battery power better than mains, and can I power the M Scaler…, Why does my Hugo 2 switch itself off, and how does standby work on…
Dither
Also called: rectangular dither, triangular dither (TPDF), Gaussian dither, pseudo Gaussian dither
For Watts, dither is the usual way to cut a long word to a shorter one, and an imperfect one: "just truncating by discarding the unused bits creates distortion", but "dither isn't perfect" Head-Fi, Mar 2020. He describes dither schemes as relying on "randomization to encode small signals" Head-Fi, Oct 2016.
In 2024 he gave the technical ranking: rectangular dither is accurate to the first moment, triangular to the first and second, and Gaussian to infinite moments Head-Fi, Jun 2024. His listening put them in the same order: rectangular worst, then triangular, then pseudo Gaussian, with noise shaping best, the difference being depth perception Head-Fi, Jun 2024. Pseudo Gaussian, "almost pure random noise", was only half as effective subjectively as truncation noise shaping Head-Fi, Oct 2021. Its random sources have grown: 14 in 2017 with 56 planned Head-Fi, Jul 2017, around fifty in 2024 Head-Fi, Jun 2024.
The Hugo M Scaler uses Gaussian dither set high for 2x and 4x upsampling, and aggressive noise shaping at 16x Head-Fi, Jul 2022; no dither is added with the full 1M taps Head-Fi, Mar 2019. He also says "you can't dither a DSD system", unlike a 5 bit delta sigma Head-Fi, Dec 2017.
“Upon listening, rectangular dither was the worst, then triangular, then pseudo Gaussian, and the best was the noise shaping.”2024-06-01 Head-Fi, Jun 2024
“just truncating by discarding the unused bits creates distortion so don't do that - but dither isn't perfect”2020-03-09 Head-Fi, Mar 2020
See: How do I make sure my player sends bit-perfect data at the native…, What does Rob mean by depth, and what in a DAC improves or…, How do Chord DACs handle DSD, and does Rob consider DSD better or…, How does a pulse array DAC work, how does it differ from R2R and…
DoP (DSD over PCM)
Also called: DoP, DSD over PCM, DoP DSD
Watts described DoP as DSD sent as regular PCM with an 8 bit header that tells the DAC the bottom 16 bits are DSD data. Head-Fi, Sep 2019 The DAC decodes it, which he said is trivial in an FPGA, and receives identical data to native DSD transmission. Head-Fi, Sep 2019 The trade-off is efficiency: DoP carries the header overhead and so works only to DSD256, while native DSD, which needs ASIO, reaches DSD512. Head-Fi, Sep 2019Head-Fi, Feb 2022 DoP is driverless and works on Linux and macOS. Head-Fi, May 2018Head-Fi, Feb 2022 On TT2 he said it is accepted on all inputs, limited by each input's rate, so optical tops out at DSD64 and coax at DSD128 if the source can handle it. Head-Fi, Mar 2025 He put DoP dropouts down to a source processor that cannot pack DoP in real time. Head-Fi, Sep 2017 He also used it as a test of a bit-perfect path: if it plays mute free, it is bit perfect. Head-Fi, Mar 2017 The original Hugo takes DSD via DoP only. Head-Fi, May 2016
“With DoP you transmit it via regular PCM, with a 8 bit header that tells the DAC that the bottom 16 bits is regular DSD data.”2019-09-23 Head-Fi, Sep 2019
“But to test it use DoP DSD - if its mute free, it's bit perfect.”2017-03-28 Head-Fi, Mar 2017
See: Should I send DSD as DoP or native?, How do Chord DACs handle DSD, and does Rob consider DSD better or…, How do I make sure my player sends bit-perfect data at the native…, My DAC drops out, clicks or loses lock. What should I check before…
DPLL
Also called: digital phase locked loop, digital phase lock loop, DFLL (digital frequency locked loop)
Watts says "You can think of the DPLL as a frequency locked micro buffer, with the DAC using the local jitter free clock" Head-Fi, Mar 2022. A new word clock on the DAC's own 104.25 MHz domain reads out the buffer, frequency locked to the source with a time constant of about ten seconds Head-Fi, May 2018Head-Fi, Jun 2018.
He says the name is slightly off: phase is used only at start-up, and once locked (within 10 mS) source jitter "has no bearing" Head-Fi, Jul 2019. In 2022 he wrote that "DFLL is more appropriate term" Head-Fi, Mar 2022.
He says it eliminates source jitter, where a conventional PLL only attenuates it: with 2 uS of jitter added to SPDIF, none showed down to -180 dB Head-Fi, Jul 2019. He began it in 2000 because the earlier large buffer was useless with video Head-Fi, Jun 2016 and to replace the analog PLL, which he says creates huge problems Head-Fi, Mar 2022; he puts the work at 6 or 7 years Head-Fi, Jun 2016.
His account changed: in 2016 you could "barely hear the difference" from a local clock and buffer Head-Fi, Jun 2016; by 2022 it sounded "identical to the RAM buffer" Head-Fi, Mar 2022. In 2016 he preferred Dave's USB, where "the clock comes from Dave" Head-Fi, Jul 2016, calling that wrong in 2018 Head-Fi, Jan 2018; in 2019 he called USB "better from the clock POV" yet heard optical as identical, proof to him that the DPLL removes source jitter Head-Fi, Feb 2019.
“You can think of the DPLL as a frequency locked micro buffer, with the DAC using the local jitter free clock.”2022-03-02 Head-Fi, Mar 2022
“Once it is locked (within 10mS) it only relies upon frequency not phase.”2019-07-25 Head-Fi, Jul 2019
See: Does source jitter matter with a Chord DAC, and would a reclocker,…, Which input sounds best on a Chord DAC (optical, coax, USB or…, What is a transient timing error, and why is it audible when a…, How does a pulse array DAC work, how does it differ from R2R and…
Dual data and dual BNC
Also called: dual data mode, DDATA, dual BNC, dual AES, 768 kHz dual data
Dual data is an AES3/SPDIF mode in which each cable carries one channel as mono data, so left and right travel on two cables. Head-Fi, Jan 2018 He explained that regular SPDIF sends two samples (left, then right) in a 384 kHz time period, while dual data sends two mono samples in that period, which makes a 768 kHz sample rate. Head-Fi, Jan 2019 It is how the Hugo M Scaler delivers 705.6/768 kHz over two BNC cables, each a mono SPDIF at 352.8/384 kHz. Head-Fi, Jul 2018Head-Fi, Sep 2018 He said 768 kHz works reliably over dual BNC and called that impossible on AES/EBU. Head-Fi, Nov 2015 On the Hugo M Scaler only the white output setting uses it. Head-Fi, Sep 2018
Products differ. Hugo 2 and Qutest read the AES3 dual data flag to decide which channel is left and right, while Dave predates that flag, uses a sync detector, and needs each cable connected to the right input by hand. Head-Fi, Jan 2018 So matched cable length matters for Dave, while he said the other DACs do not need exact sync. Head-Fi, Feb 2020Head-Fi, Aug 2020 Mojo 2 and Hugo 2 both accept it. Head-Fi, Jul 2023 The original Mojo does not, because two SPDIF decoders in the FPGA cost too much power on a battery product. Head-Fi, Jan 2020
“Dual data is again two samples in a 384 kHz time period, but mono data; so it's 768 kHz sample rate.”2019-01-22 Head-Fi, Jan 2019
“Dave requires sync, so length is important, but all the other DACs use the AES3 dual data flag so don't require exact sync.”2020-02-25 Head-Fi, Feb 2020
See: How do I connect the M Scaler to DAVE, and why does dual BNC only…, Which BNC cables, and what length, should I use between an M…, Which Chord DACs accept 705.6/768 kHz from an M Scaler, and how do…, What do the colored indicators mean (volume, sample rate,…
DX link
Also called: DX, DX output, DX input, DX amp, DX protocol, DX mode
DX is Watts's name for a digital link: regular SPDIF in AES dual data format at 705.6 or 768 kHz, with volume control data carried in the user data. Head-Fi, Feb 2017Head-Fi, May 2016 A header in the user data identifies the stream as DX. Head-Fi, Feb 2017 The idea is to send full-level data and let a DX power amp apply the volume at the best point in its own path, so Dave becomes what he called a DX pre-amp. Head-Fi, May 2016Head-Fi, Jan 2018 The only difference from the Hugo M Scaler's ordinary 768 kHz output is that volume and header data. Head-Fi, Jul 2019
Dave and TT2 have DX outputs from their volume controls, and the Hugo M Scaler can send DX volume when its DX option is on. Head-Fi, Jul 2019 Qutest accepts a DX stream but ignores the volume data, being a fixed-level DAC. Head-Fi, Jan 2018 He described the DX outputs as meant for future DACs powerful enough to drive big loudspeakers directly (a 2023 talk). Dusty Audio 1:01:31 In 2022 he said the DX amp project had stalled on two technical issues from the first prototype, one of them since solved. Head-Fi, Mar 2022 In 2024 he said he still had the large power DX DAC concept in mind. Head-Fi, Oct 2024
“DX is a regular SPDIF in AES dual data format - but with a special user data that has the required volume control data.”2017-02-20 Head-Fi, Feb 2017
“Qutest will accept the DX data, but will ignore the volume data, as Qutest does not have a volume control - it is a fixed OP DAC.”2018-01-11 Head-Fi, Jan 2018
See: Should I use a pre-amp (tube or solid state), or run the Chord DAC…, Which Chord inputs are galvanically isolated, and which setups…
FPGA
Also called: field programmable gate array, Xilinx FPGA
Watts describes an FPGA as "a field programmable gate array": you can "design any digital circuit, as long as it will fit inside the device", and do things that would be vastly too expensive in a chip The Source AV 4:47. In 2018 he called it "basically a blank digital circuit" Pursuit Perfect System 14:00. He writes the design as RTL in Verilog, and says the RTL "completely defines the digital performance" of a properly executed design Head-Fi, Aug 2014.
His case for it is parallelism: a PC does the steps of a filter tap in sequence, whereas "With an FPGA you can do all of these things in parallel at once, so a single FIR tap can be accomplished within a single clock cycle" Head-Fi, Aug 2016. Dave's FPGA has 132 DSP slices plus 34 cores built from fabric, 166 in all Head-Fi, Jan 2016.
Size limits what he can attempt: he says he was stuck at 200 dB noise shapers until Dave's much larger FPGA ecoustics 35:41.
He agrees an FPGA DAC is "at least an order of magnitude more difficult" than using an audio chip Head-Fi, Aug 2014. In pulse array the FPGA does the digital work and discrete components do the conversion to analog, keeping digital noise out of the analog signal Pursuit Perfect System 14:00Pursuit Perfect System 14:47.
“With an FPGA you can do all of these things in parallel at once, so a single FIR tap can be accomplished within a single clock cycle”2016-08-09 Head-Fi, Aug 2016
“you can design any digital circuit, as long as it will fit inside the device”2023-09-30 The Source AV 4:47
See: Why does Chord use FPGAs and discrete conversion instead of DAC…, What is the WTA filter, how was it tuned, and how does it differ…, Why does tap length matter, and why a million taps?, Is the filter algorithm more important than the tap count, and…, How does a pulse array DAC work, how does it differ from R2R and…, Should I leave my Qutest, TT2, DAVE or M Scaler switched on all…
Galvanic isolation
Also called: isolation, RF isolation, USB isolation
Watts uses the term for keeping a source's ground currents out of the DAC: completely separate grounds, isolation from low frequency noise, and effective isolation at RF Head-Fi, Mar 2019. He agrees the term "hide a multitude of sins", citing a supposedly isolated BNC adaptor that was simply two capacitors Head-Fi, Mar 2019.
In 2016 he gave two reasons: RF noise from the source creates intermodulation products, which are noise floor modulation and sound bright, and tiny correlated currents in the ground plane add to small signals and upset depth perception; isolation keeps both out Head-Fi, Jan 2016.
Qutest uses high speed RF digital isolators, not transformers, with 2pF of total coupling capacitance, plus chip ferrites because 2pF matters at GHz frequencies Head-Fi, Mar 2019. Of Dave's 2pF he said it is 79 M ohms at 1 kHz but 79 ohms at 1 GHz Head-Fi, Jun 2016. Optical he calls "the perfect galvanic isolation" Head-Fi, Jul 2018.
He qualifies it: the isolation can be short-circuited externally through ground/mains loops Head-Fi, Mar 2019, and in January 2018 he wrote that his Hugo 2 headphone tests (reported February 2017) found it unnecessary, USB sounding the same as optical; mains-powered amplifiers need it Head-Fi, Jan 2018Head-Fi, Feb 2017; in March 2022 he added that not all USB sources match optical, as the isolation is imperfect Head-Fi, Mar 2022. The DACs' BNC inputs are not isolated, though the M Scaler's BNC outputs are Head-Fi, Aug 2020.
“This means that RF noise from the source can't get into Dave, and small correlated currents can't get in too.”2016-01-04 Head-Fi, Jan 2016
“the term itself hide a multitude of sins, and can vary considerably”2019-03-12 Head-Fi, Mar 2019
See: Which Chord inputs are galvanically isolated, and which setups…, Why do I hear a buzz or feel a tingle when the charger is…, Which input sounds best on a Chord DAC (optical, coax, USB or…, Will a USB purifier, JitterBug or USB isolator improve my DAVE,…, Does source jitter matter with a Chord DAC, and would a reclocker,…, If the data reaching the DAC is bit perfect, can the source…
HF filter
Also called: HF fil, green or red filter
On Hugo 2 and TT2, Watts says the HF filter is on in the green and red settings and off in white and orange Head-Fi, Mar 2022Head-Fi, Mar 2022. In a Qutest-thread post he said it is not an extra stage: the 2048FS filter's time constants change, moving the -3dB point from 150 kHz (off) to 37.5 kHz (on) Head-Fi, Feb 2019. His figures vary: second order, -3dB at 40 kHz (2017), and 160 kHz to 40 kHz (2019) Head-Fi, Mar 2017Head-Fi, Jun 2019. He described DAVE's differently: in 2015 he called it a 60 kHz filter, and in 2017 said that with the M Scaler it is best off Head-Fi, Nov 2015Head-Fi, Jan 2017. He says it reduces the out of band noise-shaper noise ADCs put into high-rate recordings, which causes noise floor modulation in the analog stages Head-Fi, Dec 2017Head-Fi, Jan 2018. On, he says, it sounds warmer and smoother, with better separation and focus Head-Fi, Mar 2022. The change is small at 44.1 kHz and large at 96 and 192 kHz Head-Fi, Mar 2017. In 2019 he called noise floor modulation the likeliest of three explanations for the change with CD Head-Fi, Jun 2019; in 2022 he said it is not its -1dB at 20kHz, since a Mojo 2 filter flat at 20kHz sounded the same Head-Fi, Mar 2022.
“what actually happens is the time constants of the 2048FS filter is changed, not a signal path being switched”2019-02-09 Head-Fi, Feb 2019
“So the HF filter reduces this, and it sounds noticeably warmer with HD recordings - much more so than with RBCD.”2017-12-05 Head-Fi, Dec 2017
See: Which filter setting should I use on Hugo 2, TT2 or Qutest, and…, Does hi-res audio sound better than CD, and should I buy the 96k…, What is noise floor modulation, how does Rob measure it, and why…
Jitter
Also called: clock jitter, master clock jitter, source jitter, low frequency jitter, cycle to cycle jitter
Watts defines jitter simply: "jitter, this is An uncertainty in clock timing" (2024) MFG Audio Reviews 2:33. He separates two kinds of master clock jitter: low frequency, where the frequency shifts over the millisecond to second range, and cycle to cycle, which is mostly high frequency Head-Fi, Jun 2024.
He says the low frequency kind is "much more important subjectively than cycle to cycle jitter", and that the skirts it causes on fundamentals sound like noise floor modulation, brighter Head-Fi, Jan 2018; "if you can measure a skirt it is audible" Head-Fi, Sep 2016. That is why he avoids femto and atomic clocks, which have low cycle to cycle jitter but, because they use PLLs, lots of low frequency jitter Head-Fi, Sep 2016. Quoted cycle to cycle figures, he says, tell only half the story The Headphone Show 3:16.
Source jitter, he says, is eliminated by the DPLL Head-Fi, Jan 2018, and in 2024 he said it permanently fixed the low frequency problem; what remains, cycle to cycle jitter, depends on DAC architecture, and enough pulse array elements make even 1nS unmeasurable Head-Fi, Jun 2024. On his pulse array DACs, switching the clock off improves the noise by only 0.5 dB Head-Fi, Jan 2020.
He has changed his mind on details. In August 2016 he wrote that USB "has slightly better timing as the source clock is Dave's local clock" Head-Fi, Aug 2016; in January 2018, "I have changed my mind on this now", as the DPLL eliminates the source jitter Head-Fi, Jan 2018; in February 2019 he still wrote "Technically USB is better from the clock POV" Head-Fi, Feb 2019.
“jitter, this is An uncertainty in clock timing”2024-03-20 MFG Audio Reviews 2:33
“With low frequency master clock jitter no DAC architecture can fix it - it's like wow and flutter from tape”2024-06-10 Head-Fi, Jun 2024
See: Does source jitter matter with a Chord DAC, and would a reclocker,…, Which input sounds best on a Chord DAC (optical, coax, USB or…, How does a pulse array DAC work, how does it differ from R2R and…, Do digital cables (USB, coax, glass or plastic optical) make a…, Will a USB purifier, JitterBug or USB isolator improve my DAVE,…, If the data reaching the DAC is bit perfect, can the source…, What is a transient timing error, and why is it audible when a…
Minimum phase filter
Also called: min phase, minimum phase type filter, no pre-ringing filter
Watts describes minimum phase filters as ones that look only into the past, with post-ringing but no pre-ringing Head-Fi, Oct 2017Passion for Sound 18:00. In a 2017 talk a minimum phase or IIR filter, "any filter that's got no pre-ringing", loses the start of a transient, about 100 microseconds of timing error in his example (so check the figure) miceblue425 12:06. He says they use fractions of a millisecond of data and sound "flat, soft and ill defined" with poor timbre variation and pitch Head-Fi, Mar 2018. He treats MQA's interpolation filter as this type and says it sounds "very much worse than the WTA" Head-Fi, Aug 2018. He says he does not offer a minimum phase option: "If there's something that you know sounds worse I don't give that as an option to people" Passion for Sound 16:34. Qutest's orange and red settings are separate: a third-order IIR stage replaces WTA2 from 16FS, after WTA1 Head-Fi, Mar 2019Head-Fi, Jan 2019. He groups IIR with minimum phase when criticizing other makers Head-Fi, Oct 2017, but does not call this stage minimum phase. In the M Scaler's video mode he uses an asymmetric filter that he says sounds better than short pre and post ringing but gives lower latency "at a sound quality price" Head-Fi, Nov 2023Head-Fi, May 2022. For EQ inside the audio band he does recommend causal or IIR filters Head-Fi, Mar 2025.
“I do not rate min phase filters at all, and compared to a WTA filter minimum phase sounds flat, soft and ill defined”2018-03-06 Head-Fi, Mar 2018
“take the approach of minimum phase, with only post ringing using asymmetric FIR filters (or IIR filters)”2017-10-17 Head-Fi, Oct 2017
See: Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…, Do Chord DACs decode MQA, and what does Rob think of it?, What is a transient timing error, and why is it audible when a…
Noise floor modulation (NFM)
Also called: NFM, RF noise floor modulation, slew rate dependent noise floor modulation
Watts explains that every DAC makes noise, and a fixed amount is unimportant, like an air conditioning system in a concert hall; the problem is that "the noise changes with signal" (2018) Pursuit Perfect System 9:36. In 2024 he put it as "as the signal rises your noise pumps up and down with the signal" Reactcore 0.
He says the brain cannot separate the instrument from the modulated noise, as with white noise pumping up and down with a rich, dark saxophone MFG Audio Reviews 5:54. It makes warm instruments sound artificially bright, which people can mistake for transparency Reactcore 0.
To measure it he overlays 0 dBFS with a -999 dBFS signal, all APx555 settings identical, which he calls the only way Head-Fi, Nov 2021. A design is signed off only with no measurable noise floor modulation Head-Fi, Feb 2018, and he says pulse array is capable of zero Head-Fi, Mar 2016.
In 2018 he wrote that even levels well below measurability are audible Head-Fi, Feb 2018. In 2021 he concluded "the audible limit is zero" Head-Fi, Apr 2021, and in 2024 that his threshold is much smaller than minus 300 dB, a number he would not give HIFICLUB 48:04.
“noise floor modulation is when as the signal rises your noise pumps up and down with the signal”2024-05-10 Reactcore 0
“any trace of noise floor modulation, no matter how small, seems audible with a very careful AB listening test”2021-04-03 Head-Fi, Apr 2021
See: What is noise floor modulation, how does Rob measure it, and why…, How does a pulse array DAC work, how does it differ from R2R and…, Do measurements tell you how good a DAC sounds, and is Rob an…, What does Rob mean by depth, and what in a DAC improves or…, Do digital cables (USB, coax, glass or plastic optical) make a…, Should I replace the supplied power supply with a linear or…
Noise shaper
Also called: noise shaping, delta sigma modulator, truncation noise shaper, pulse array noise shaper, analogue noise shaper (output stage)
In a CanJam NYC talk (late February 2022; published March 2022) Watts described a noise shaper as "a feedback mechanism that recovers the error and corrects for it": it remembers the error and feeds it back so it is reduced on the next sample MFG Audio Reviews 36:44. The downside, he says, is that it must run at a higher oversampling rate, 705 kHz rather than 44.1 in that talk (a figure) MFG Audio Reviews 36:44. In a delta-sigma DAC it converts a 32-bit signal down to around 5 bits, because the DAC has only a few switches, 20 in Dave ecoustics 34:09.
He says its performance depends on how fast it runs and on the number of integrators inside it ecoustics 34:57. Signals below its floor are lost: a 200 dB shaper "will not resolve" them, and most work at 140 dB MFG Audio Reviews 44:05.
He separates truncation noise shapers from pulse array output noise shapers, which he says are "very different" Head-Fi, Jan 2023. In 2015 he gave Dave's as 5 bits at 104 MHz, 17th order, with 350 dB digital-domain THD and noise Head-Fi, Jul 2015. In March 2016 he gave its output as 6 bits, including overload bits (0 to 20 for its 20 elements) Head-Fi, Mar 2016. In a 2016 interview he added that it has 46 integrators Hifistatement 3:54.
“a noise shaper is a feedback mechanism that recovers the error and corrects for it”2022-02 MFG Audio Reviews 36:44
“if your signal is below Minus 200 dB, the noise shaper will not resolve it”2024-03 MFG Audio Reviews 44:05
See: How does a pulse array DAC work, how does it differ from R2R and…, What does Rob mean by depth, and what in a DAC improves or…, What is a transient timing error, and why is it audible when a…, Do measurements tell you how good a DAC sounds, and is Rob an…, Does hi-res audio sound better than CD, and should I buy the 96k…, How do Chord DACs handle DSD, and does Rob consider DSD better or…, Which filter setting should I use on Hugo 2, TT2 or Qutest, and…
NOS (non-oversampling) DAC
Also called: NOS, non-oversampling, no oversampling, zero order hold, first order hold, sample and hold
Watts says the NOS label is wrong: every DAC oversamples, and a NOS DAC uses the simplest digital interpolation filter, a sample and hold Soundnews 14:48Head-Fi, Jul 2019. He has named that filter inconsistently, zero order hold in 2019 and first order hold in 2023 and 2026 Head-Fi, Jul 2019Head-Fi, Nov 2023HiFi Buys 42:01. He calls it "the crudest digital filter possible" that "creates huge levels of timing and aliasing distortions" Head-Fi, Jul 2019. As an illustration he gives a one-tap filter with a worst-case timing error of "100uS or so" Head-Fi, Jan 2017. He says he has heard NOS and it sounds "soft, flat and out of focus" Head-Fi, Dec 2017. He also denies that NOS DACs are bit perfect, since sampling theory treats samples as zero width and NOS fills the gaps with a sample and hold Head-Fi, Mar 2019. In 2026 he said its timing errors leave the brain unable to perceive transients, so things sound soft and warm, and paired with a DAC making a lot of high frequency distortion the two balance one another out, one distortion compensating another HiFi Buys 42:01. He says converting his DACs to NOS would take zero effort but would destroy musicality Head-Fi, Nov 2023.
“the term NOS for a DAC is a non sequitur or a delusion”2023-11-13 Head-Fi, Nov 2023
“NOS DAC's - if your intent is to be as transparent as possible - are an unmitigated disaster.”2017-12-18 Head-Fi, Dec 2017
See: Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…, What is a transient timing error, and why is it audible when a…, Why does tap length matter, and why a million taps?
Oversampling (16FS, 256FS, 2048FS)
Also called: upsampling, interpolation, FS, 16FS, 256FS, 2048FS, oversampling rate
Watts writes oversampling rates as multiples of FS: Hugo's first up-sampling is 8 times, "1FS to 8FS" Head-Fi, Mar 2015. He gives 16FS as 768 kHz and 256FS as 88 nS resolution Head-Fi, Jan 2017. Every DAC upsamples, he says, because turning sampled data into a continuous waveform is upsampling; the question is how well it is done Head-Fi, Dec 2019. The ideal sinc is infinitely oversampled, so a finite filter faces "a balancing act" between a high rate with less data and a lower rate with more data Head-Fi, Jan 2019. Doubling the rate with the same number of samples improves the sound, with a massive step from 8FS to 16FS Head-Fi, Jan 2019. Going from 16FS to 256FS sounded dramatically better, particularly in the starting and stopping of notes Head-Fi, Jan 2017. The 16FS to 256FS WTA filter came in with the DAVE program, and he says splitting the job into WTA 1 to 16FS, WTA 2 to 256FS and a non-WTA third-order filter to 2048FS was based on listening tests Head-Fi, Jan 2017Head-Fi, Jan 2017. Above 16FS he heard more limited gains: clarity, soundstage and the starting and stopping of notes Head-Fi, Sep 2021. Taps are a separate matter from the rate Head-Fi, Mar 2015.
“either running at a high oversampling rate, and working with less data, or running at a lower oversample rate but the filter processing more data”2019-01-29 Head-Fi, Jan 2019
“in the case of 256FS we can say that to 88 nS resolution, we have close to ideal sinc recovery of the original timing”2017-01-09 Head-Fi, Jan 2017
See: Which filter setting should I use on Hugo 2, TT2 or Qutest, and…, Why does tap length matter, and why a million taps?, What is the WTA filter, how was it tuned, and how does it differ…, Should I upsample in software (HQPlayer, PGGB or my player) before…, Is the filter algorithm more important than the tap count, and…
Pass-through
Also called: pass thru, pass-thru, pass through mode, OP SR red
Pass-through is a Hugo M Scaler output setting (OP SR button red) that sends the input on at its own sample rate, which he added so it is easy to hear the difference the scaler makes. Head-Fi, Sep 2018Head-Fi, Nov 2018Head-Fi, Jul 2018 Inside, the WTA filter becomes a 1 tap filter, and the delay stays the same as in normal operation. Head-Fi, Dec 2018Head-Fi, May 2019 It is not bit perfect: the level is adjusted to match the scaler's insertion loss, which he gave as 2.76 dB and said is essential to avoid clipping from inter-sample overloads and Gibbs phenomena. Head-Fi, Dec 2018Head-Fi, Oct 2019 That keeps the two modes volume matched, but against a DAC fed directly he advised turning the direct feed down by 3 dB, and said even that is not exactly matched. Head-Fi, Oct 2019 Only one BNC is active in pass-through, so the right cable must be used. Head-Fi, Aug 2019 In 2021 he said 705/768 kHz inputs are treated like pass-through: scaled in volume, then noise shaped to 24 bits for the BNC outputs. Head-Fi, Dec 2021
“Please remember that pass-thru is not bit perfect - it's intent is to allow one to easily hear the effect of the M scaler.”2019-10-18 Head-Fi, Oct 2019
“In pass-thru the WTA filter is actually a 1 tap filter, so 1 million multiplications becomes just one!”2018-12-07 Head-Fi, Dec 2018
See: Is M Scaler pass-through bit perfect, and why is the scaled output…, How do I connect the M Scaler to DAVE, and why does dual BNC only…, What do the colored indicators mean (volume, sample rate,…
Pre-ringing
Also called: ringing, pre and post ringing
Pre-ringing is a filter response before the impulse; symmetric filters that use future samples produce it, causal filters do not Head-Fi, Mar 2025. In 2014 he said the high end's insistence on no pre-ringing filters was "completely the opposite to my thinking" Head-Fi, Apr 2014. For a DAC's interpolation filter he says it is essential: the filter needs to know the future to place a transient correctly, and this creates the pre-ringing Head-Fi, Jul 2022. The paradox that an ideal sinc rings forever yet reconstructs perfectly is resolved, he says, because an impulse test signal is not bandwidth limited Head-Fi, Jul 2022. With a bandwidth limited impulse a sinc filter does not ring at all, while minimum phase or MQA filters ring more Passion for Sound 15:30. Any ringing would sit at 22 kHz for 44.1k material, which he calls inaudible Head-Fi, Mar 2025. Where he says it does matter is filtering inside the audio band, such as EQ, where he recommends causal or IIR filters Head-Fi, Mar 2025. For the decimation filters of his ADC, his 2022 listening tests as a plain low-pass filter suggested pre-ringing is "subjectively completely inconsequential" Head-Fi, May 2022.
“in order to correctly reconstruct the timing of transients the interpolation filter needs to know the future - and this creates the pre-ringing.”2022-07-15 Head-Fi, Jul 2022
“pre-ringing for an interpolation filter isn't an issue at all, in spite of what the rest of the audio industry says”2025-03-26 Head-Fi, Mar 2025
See: Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…, What is a transient timing error, and why is it audible when a…, Do Chord DACs decode MQA, and what does Rob think of it?
Pulse array
Also called: pulse array DAC, 4e, 10e, 20e (element counts), Power Pulse Array (his planned amplifier technology)
Pulse array is Watts's own DAC technology, which he calls "a type of Delta Sigma" Passion for Sound 3:22. In 2023 he described it as "an array of discrete flip-flops and resistors" carrying PWM signals, with a rising edge balanced by a falling edge The Headphone Show 45:40. A noise shaper feeds the array, and each flip-flop and resistor is one element, written 1e Head-Fi, Mar 2016.
In his 2016 blog post he explains the balancing: with 16 PWM time states across 16 elements with different timing, a rising edge always meets a falling edge, so switching errors become constant and signal independent, with no noise floor modulation or distortion from switching Head-Fi, May 2016.
Element counts differ by product: Dave has 20 Head-Fi, May 2016, TT2 is "a 10 element pulse array" Head-Fi, Oct 2018, and Mojo's 4e array is noisier than Hugo 2's 10e Head-Fi, Nov 2020.
His jitter wording varies. In 2016 he called the array "fundamentally jitter immune" Head-Fi, Mar 2016, while noting that jitter can still play a part when the noise shaper changes its output, at 1/16 of the DSD or PWM error Head-Fi, May 2016. In 2020 he said switching the clock off improves the noise by "a negligible 0.5dB" as pulse array is "innately jitter tolerant" Head-Fi, Jan 2020.
“with 16 elements and 16 PWM time states we can completely balance switching activity - this in a nutshell is how pulse array works”2016-05-17 Head-Fi, May 2016
See: How does a pulse array DAC work, how does it differ from R2R and…, What is noise floor modulation, how does Rob measure it, and why…, Does source jitter matter with a Chord DAC, and would a reclocker,…, Why does Chord use FPGAs and discrete conversion instead of DAC…, How do Chord DACs handle DSD, and does Rob consider DSD better or…, What does Rob mean by depth, and what in a DAC improves or…
RF noise
Also called: random RF noise, signal correlated RF noise, correlated RF noise, correlated noise
Watts describes two forms of RF noise with different effects. Random RF noise, from sources such as fluorescent tubes and even LEDs, is the one he worries about most Eric's Hi-Fi 14:59. It gets into the analog electronics and creates intermodulation products that are themselves random, so "the random noise pumps up and down with the signal level" (in his CanJam RMAF talk, October 2017; video published December 2017), which is noise floor modulation and makes rich, dark instruments sound brighter RMAF 11:24Head-Fi, Nov 2017. Fixed-frequency switching noise matters far less: its products fall around the carrier, outside the audio band, and it is easy to filter MFG Audio Reviews 9:00Head-Fi, Oct 2021.
Signal correlated RF noise comes from digital processing: bursts in the gigahertz region that follow the signal, worst for very small signals, which the analog circuitry demodulates, "corrupting your small signals" and degrading depth MFG Audio Reviews 20:22MFG Audio Reviews 21:06. In 2017 he traced it to ground bounce from logic transitions in the FPGA, centered around 2 GHz, getting into the DAC's ground plane Head-Fi, Nov 2017.
He says removing random RF noise is very hard: in a SPICE study from 100kHz to 10GHz, switching noise was easy to filter but random noise was "a complete nightmare" Head-Fi, Oct 2021.
“The distortion products is random noise. And that means the random noise pumps up and down with the signal level.”2017-10 RMAF 11:24
“Depth changes are due to signal correlated RF noise which gets into Dave's ground plane, then is demodulated by the analogue electronics.”2017-11-29 Head-Fi, Nov 2017
See: What is noise floor modulation, how does Rob measure it, and why…, Do digital cables (USB, coax, glass or plastic optical) make a…, Which Chord inputs are galvanically isolated, and which setups…, Should I replace the supplied power supply with a linear or…, Will a USB purifier, JitterBug or USB isolator improve my DAVE,…, Which input sounds best on a Chord DAC (optical, coax, USB or…, If the data reaching the DAC is bit perfect, can the source…
Sinc function and ideal reconstruction
Also called: sinc function, ideal sinc, sinc function filter, Whittaker-Shannon interpolation filter, ideal reconstruction filter
Watts says sampling theory proves that a Whittaker-Shannon interpolation filter, "a filter that has a sinc function and processes from the infinite past to the infinite future", will perfectly reconstruct a bandwidth limited signal Hifistatement 0. What it recovers is the analog signal that existed in the ADC before sampling, not the digital data itself Head-Fi, Aug 2018Head-Fi, Oct 2017. The signal has to be bandwidth limited, with no energy at and above half the sample rate Head-Fi, May 2022. The catch is that the ideal filter rings before and after an impulse for an infinite period, so it needs infinite processing and an infinite wait c w 12:28. In the real world, he says, the coefficients must always be windowed, and perfect reconstruction is impossible Head-Fi, Aug 2023. His stated objective is a filter as close as possible to sinc in both the time and the frequency domain Head-Fi, Nov 2020, and he says the M Scaler is identical to ideal sinc to better than 16-bit accuracy Head-Fi, Aug 2018Head-Fi, Mar 2019.
“The ideal reconstruction filter (sinc function) will perfectly reconstruct the bandwidth limited analogue signal that was in the ADC, before it was sampled.”2017-10-18 Head-Fi, Oct 2017
“In the real world you must always window the coefficients, as sinc function converges on zero at +/- infinity.”2023-08-22 Head-Fi, Aug 2023
See: What is the WTA filter, how was it tuned, and how does it differ…, What is a transient timing error, and why is it audible when a…, How can upscaling to 705.6/768 kHz improve sound if the recording…, Why does tap length matter, and why a million taps?, Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…
Small-signal accuracy
Also called: small signal linearity, small signal amplitude linearity, small signal resolution, small signal non-linearity (the error), fundamental linearity
Watts's example at CanJam NYC 2022 (late February, show dates unverified; video published March 2022): a signal at maximum output, 0 dB, reduced to minus 120 dB should come out at minus 120 dB, but measured it may be minus 121 or minus 119, a small signal amplitude error MFG Audio Reviews 9:00. In 2017 he called it fundamental linearity, tested by setting the system to minus 60 dB and measuring the amplitude of a minus 100 dB signal RMAF 17:37.
He says the error matters because it "confuses the brain" and makes depth hard to perceive, so more accurate small signals give better depth MFG Audio Reviews 9:00.
In digital processing he traces the loss to noise shapers: signals smaller than a shaper's resolution are lost and those approaching it are progressively attenuated, which he calls mathematically identical to the nonlinearity of metal/metal interfaces Head-Fi, Mar 2016. With Dave he put a number on it, at least 350 dB noise shapers, saying small signal linearity "has to be completely perfect" Head-Fi, Mar 2016. He admits "great difficulty" accepting that the brain is this sensitive, but reports what he hears Head-Fi, Mar 2016.
In 2022 he listed it first of three issues, with noise floor modulation and transient timing, that set his DACs apart Darko.Audio 6:56Darko.Audio 8:26.
“This error is very important subjectively because it confuses the brain and it makes it difficult for you to perceive depth.”2022-02 MFG Audio Reviews 9:00
“the small signal linearity has to be completely perfect, otherwise depth gets degraded”2016-03-01 Head-Fi, Mar 2016
See: What does Rob mean by depth, and what in a DAC improves or…, Do measurements tell you how good a DAC sounds, and is Rob an…, What is noise floor modulation, how does Rob measure it, and why…, How do Chord DACs handle DSD, and does Rob consider DSD better or…, What is a transient timing error, and why is it audible when a…, Why does Chord use FPGAs and discrete conversion instead of DAC…
Starting and stopping of notes
Also called: starting and stopping of a note, perception of timing
By the starting and stopping of notes Watts means hearing where a note begins and ends: the initial pluck or crack at the start, and at the end the stop itself. Hifistatement 2:57Head-Fi, Oct 2016Head-Fi, Apr 2017 He first noticed it as an unexpected quality of the original Hugo. Head-Fi, Nov 2015Head-Fi, Oct 2016 He said it was not the WTA filter but the much better 2048 FS interpolation filters, so what mattered was time domain accuracy. Head-Fi, Nov 2015 On Dave a 256 FS WTA filter improved it further, so things sounded sharper, faster and tighter. Head-Fi, Oct 2016 He compared the effect to focusing a lens: edges become more vivid and immediate. Head-Fi, Nov 2015 In talks he said we perceive it from transients, and uncertainty in their timing blurs the sound. c w 7:49Hifistatement 2:57 He heard the stopping most easily on electronic music, where a note can stop immediately. Head-Fi, Apr 2017 Of Hugo 2's filter options he noted that some listeners hear big changes in the starting and stopping of notes from the 16FS to 256FS WTA 2 filter, while some cannot perceive it at all. Head-Fi, Oct 2018
“Now when it becomes easier to hear the starting and stopping of notes, things become much brighter and more immediate.”2015-11-22 Head-Fi, Nov 2015
“for perception of the stopping of notes you hear this most easily with electronica”2017-04-18 Head-Fi, Apr 2017
See: What is a transient timing error, and why is it audible when a…, Why does tap length matter, and why a million taps?, What is the WTA filter, how was it tuned, and how does it differ…, Which filter setting should I use on Hugo 2, TT2 or Qutest, and…, Do Chord DACs need burn-in or warm-up?, How can upscaling to 705.6/768 kHz improve sound if the recording…
Taps (tap length)
Also called: tap length, number of taps, FIR filter order plus 1
A tap, in his explanation, is a single unit of processing in an FIR filter: a stored sample is multiplied by a filter coefficient and the result is accumulated Head-Fi, Mar 2019. The name comes from early analog FIR filters, where you would tap into an analog delay line Head-Fi, Mar 2019. In 2014, Hugo's 26,368 taps meant 26,368 multiplications and additions per CD sample Head-Fi, May 2014. He says perfect reconstruction needs an infinite number of taps, and as taps are removed the precision of the timing of a transient falls off Head-Fi, Apr 2014. He ties the gain to how fast the sinc function decays: doubling the tap length halves the sinc values Head-Fi, Mar 2019, and with his WTA filter it doubles the reconstruction accuracy, which he counts as one more bit Head-Fi, Mar 2019Head-Fi, Jul 2018. The gain depends on the algorithm, since "it's pointless to increase the tap length using existing algorithms" Head-Fi, Mar 2019. He says the only way of guaranteeing better than 16-bit reconstruction is a sinc function with a million taps Head-Fi, Mar 2019. In a 2023 interview he said that target "wasn't based upon listening tests" but on reconstructing 16-bit data to 16-bit accuracy, needing "more than a half a million taps", so "you're looking at a million taps" Eric's Hi-Fi 2:11.
“A tap is a single unit of processing for an FIR filter.”2019-03-21 Head-Fi, Mar 2019
“Doubling the tap length with my WTA filter will double the reconstruction accuracy - that's why tap length is crucial for performance.”2019-03-21 Head-Fi, Mar 2019
See: Why does tap length matter, and why a million taps?, Is the filter algorithm more important than the tap count, and…, What is the WTA filter, how was it tuned, and how does it differ…, Should I upsample in software (HQPlayer, PGGB or my player) before…, Which Chord DACs accept 705.6/768 kHz from an M Scaler, and how do…
Timbre variation
Also called: variation in timbre, variations in timbre, timbre
Timbre variation, for Watts, is the reproduction of the wide differences in tone between instruments that live music has (a 2017 talk): a hard, bright instrument next to a saxophone that sounds rich, smooth and warm. RMAF 6:51 He lists it among the variations a system should play back, and in 2022 said that more variation of this kind marks a fundamental improvement. RMAF 6:51MFG Audio Reviews 48:56 On Dave he tied it to two things: no noise floor modulation, and accurate reconstruction of transient timing. Head-Fi, Nov 2016 Timbre perception comes from transients, so when transient timing is distorted one cannot perceive timbre correctly. Head-Fi, Nov 2016 And noise floor modulation, even in small amounts, adds brightness and suppresses timbre variation, so warm instruments sound unnaturally bright. Head-Fi, Mar 2018 His example, from a 2018 talk: a saxophone should sound rich, dark and smooth, so if detail resolution seems better but timbre variation is worse, that is a red flag. Pursuit Perfect System 25:55 In a 2017 interview clip he said every increase in WTA tap length made timbre variation more accurate. Hifistatement 0
“in small amounts it adds brightness and suppresses timbre variation - in that warm instruments sound unnaturally bright too”2018-03-11 Head-Fi, Mar 2018
“Timbre perception comes from transients, and when transient timing is distorted, one can't perceive timbre correctly.”2016-11-10 Head-Fi, Nov 2016
See: What is the WTA filter, how was it tuned, and how does it differ…, Why does tap length matter, and why a million taps?, What is a transient timing error, and why is it audible when a…, How can upscaling to 705.6/768 kHz improve sound if the recording…
Transient timing error
Also called: timing of transients, transient timing, timing uncertainty
By a transient timing error Watts means a transient edge that comes out too early or too late compared to the rest of the signal Head-Fi, Jan 2017. His worked example (2016): a signal sampled at 44.1 kHz crosses zero 20,155 nS after a sample; an ideal sinc filter reproduces that crossing exactly, but a finite non-sinc filter might put it "maybe at 19,000 nS or 21,000 nS" Head-Fi, May 2016. He defines a transient as everything that is not steady state Head-Fi, Sep 2018. He is not talking about ringing but about "the non-linear timing uncertainty that the interpolation filter introduces" Head-Fi, Feb 2017. The error is not constant: it changes "dependent upon the sampling point and the past and future samples" Head-Fi, Aug 2018. A fixed timing error is, in his account, inaudible, while one that is constantly changing is audible Head-Fi, Jun 2019. He has said errors in conventional filters can reach 100uS Head-Fi, May 2016, but in 2018 he said he did not have numbers to show timing error, and his working assumption was that any constantly varying error is likely to be audible, no matter how small Head-Fi, Sep 2018. In 2026 he described two contrary effects degrading transient timing in all interpolation filters article, Oct 2026.
“A fixed timing error is not important - so a shift of 1us at 5 kHz say is inaudible; but a 1us shift that is constantly changing is audible”2019-06-14 Head-Fi, Jun 2019
“And it is these differences in the timing of transients, are what I am talking about.”2016-05-17 Head-Fi, May 2016
See: What is a transient timing error, and why is it audible when a…, Why does tap length matter, and why a million taps?, What is the WTA filter, how was it tuned, and how does it differ…, Isn't pre-ringing bad? Why doesn't Chord offer minimum-phase,…, Is the filter algorithm more important than the tap count, and…
Truncation and re-quantization
Also called: re-quantization, requantisation, truncator, truncation noise shaper, word length reduction
Watts uses truncation for reducing the number of bits, and distinguishes it from decimation, which removes samples to lower the sample rate Head-Fi, Jan 2015. He also calls it re-quantization: DSP output has a bigger bit depth, and "It's the truncation where the difficulties lie" when it must be cut back for the next stage Head-Fi, Jul 2017.
His examples are Hugo's 51-bit internal volume result, truncated at 16FS "via noise shaping and dithering" with a fixed noise at -180dB (2015) Head-Fi, Jan 2015, and the M Scaler's 56-bit outputs, converted to 24 bits by truncation noise shapers (2019) Head-Fi, Apr 2019.
He says simply discarding bits introduces severe small signal distortions Head-Fi, Jul 2021, and that any volume change involving truncation degrades depth and detail Head-Fi, Feb 2020. Dither only partly solves it Head-Fi, Jul 2017; the only transparent method he has found is aggressive 11th order noise shaping at 705/768 kHz Head-Fi, Feb 2020.
His wording varies with what he means by the word: he calls his stage a "truncation noise shaper" Head-Fi, Dec 2019, yet in 2021 said the volume control output "is simply not truncated" but noise shaped Head-Fi, Jul 2021, and in 2022 that his DACs "never truncate", meaning "no bits are thrown away" Head-Fi, Apr 2022.
“The term you are looking for is truncation, where bits are reduced.”2015-01-23 Head-Fi, Jan 2015
“When you use DSP, the output has a bigger bit depth; but to get to the next stage, you need to truncate the bit depth.”2017-07-24 Head-Fi, Jul 2017
See: How do I make sure my player sends bit-perfect data at the native…, What does Rob mean by depth, and what in a DAC improves or…, Is the filter algorithm more important than the tap count, and…, How does a pulse array DAC work, how does it differ from R2R and…, How do I connect the M Scaler to DAVE, and why does dual BNC only…
Upscaling (M scaling)
Also called: upsampling, up-scaling, M scaling, interpolation
Watts says upscaling to 705.6 or 768 kHz adds no information: "converting the 44.1 kHz to 705.6 kHz is not adding any more information", and "It's not making things up", it reproduces the analog signal that was in the ADC Head-Fi, Aug 2018. In his words, upsampling "is not about creating more information, but about eliminating information loss or errors" Head-Fi, Sep 2018. The M Scaler's purpose is to convert a CD "into a format that was identical to an original 705 kHz recording", to better than 16 bits Head-Fi, Jan 2019. It does not sample rate convert, so 44.1 becomes 705.6 and 48 becomes 768 Head-Fi, Aug 2018. He puts the audible gain in transient timing, not content above 20 kHz: "it's about how accurate the timing of transients actually are, not being able to hear 768k" Head-Fi, Jan 2019. He adds that every DAC upsamples, so the question is how accurately Head-Fi, Dec 2019. His long-standing advice against software upsamplers rested on tap length and the WTA algorithm; by 2020 he noted other solutions claim long tap lengths too, so the advice "is losing out on one benefit" Head-Fi, Nov 2020.
“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.”2018-09-01 Head-Fi, Sep 2018
“every single DAC on the planet upsamples - as a DAC converts sampled data into a continuous waveform, and to do this we are upsampling”2019-12-14 Head-Fi, Dec 2019
See: How can upscaling to 705.6/768 kHz improve sound if the recording…, Should I upsample in software (HQPlayer, PGGB or my player) before…, Which Chord DACs accept 705.6/768 kHz from an M Scaler, and how do…, How do I make sure my player sends bit-perfect data at the native…
WTA filter
Also called: WTA, WTA algorithm, Watts Transient Aligned, WTA1, WTA2, Blackbird WTA
He describes the WTA algorithm as the recipe to create the coefficients of his interpolation filters Audiophile Junkie 10:46, and gave its name as "Watts Transient Aligned" Hifistatement 2:24. It is "a unique windowing function designed to maximise the accuracy of recovering transient timing from the original analogue signal" Head-Fi, Nov 2020. A windowing function is what tailors the infinite sinc function into a finite one of usable size Head-Fi, Jul 2018. He calls the WTA a compromise between keeping as many coefficients as possible at the sinc value and avoiding a rapid change from zero up to full sinc Head-Fi, Nov 2020. He says over half the coefficients are identical to the ideal sinc function Head-Fi, Jul 2018, whereas in a conventional filter none are Audiophile Junkie 9:19. Its equation has five variables, the first being tap length, and optimizing it took "probably thousands of listening tests" Head-Fi, Nov 2020. In 2020 he said one variable must be set to 10 parts per million accuracy; in 2025 he said the variables were fine tuned by ear to that level Head-Fi, Nov 2020Audiophile Junkie 10:46. He has never published the window Head-Fi, Jul 2022. In 2026 he described Quartet's Blackbird WTA as a new filter structure that handles separately two contrary problems a single-stage filter can only balance article, Oct 2026article, Oct 2026Audio Force 14:13.
“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
See: What is the WTA filter, how was it tuned, and how does it differ…, Is the filter algorithm more important than the tap count, and…, Why does tap length matter, and why a million taps?, Should I upsample in software (HQPlayer, PGGB or my player) before…, Why does Chord use FPGAs and discrete conversion instead of DAC…