Tap length and diminishing returns

Reconstruction filters and timing

Answers on this topic

• How many taps and pulse array elements does each Chord DAC have?
• Is the filter algorithm more important than the tap count, and would more than a million taps keep improving the sound?
• Should I upsample in software (HQPlayer, PGGB or my player) before a Chord DAC, and can it match an M Scaler?
• What does the M Scaler actually do, and is it worth adding to my DAC?
• Why does tap length matter, and why a million taps?

Everything he said on this topic (357)

One-line summaries in our words, oldest first. Each link opens the original.

• 2014-04-15Watts says Hugo's DAC is a pulse array with a 104 MHz noise shaper, the DPLL eliminates source jitter, and more taps keep helping, guessing 1M taps for 16-bit accuracy. Head-Fi, Apr 2014
• 2014-04-16Watts compares QBD76's 16-element pulse array with Hugo's 4, says a replacement is in work, and recounts how QBD76 development reshaped the WTA filter to 18k taps. Head-Fi, Apr 2014
• 2014-04-17Watts explains that bandlimited samples are perfectly reconstructed by an infinite-tap filter, including transients starting mid-sample, and timing precision falls off as taps are reduced. Head-Fi, Apr 2014
• 2014-04-17Watts says FPGA progress may allow 10M taps in ten years, stresses that he has argued for tap length and timing since DAC64, against a high-end focus on pre-ringing. Head-Fi, Apr 2014
• 2014-04-17Watts says the tap length is set by the FPGA, reached through many iterations to fit and meet timing while leaving resources for other functions. Head-Fi, Apr 2014
• 2014-05-16Watts credits Hugo's 26,368 WTA taps for its timing, says more taps should help further, and suggests hi-res can underperform red-book because ADC out-of-band noise causes noise floor modulation. Head-Fi, May 2014
• 2014-05-22Watts explains FIR taps and WTA coefficients, says perfect reconstruction needs infinite taps, and that Hugo's 26,368 taps run on 16 parallel DSP cores at 208 MHz. Head-Fi, May 2014
• 2014-06-10Watts corrects that taps belong to the FIR oversampling filter, not the DAC section, and says Hugo's tap length is two orders of magnitude above conventional silicon DACs. Head-Fi, Jun 2014
• 2014-07-03Watts explains that an infinite-tap filter perfectly reconstructs bandlimited signals while finite taps cause transient timing errors, hundreds of uS for one tap, so he lengthened taps by listening. Head-Fi, Jul 2014
• 2014-07-03Watts says a sin(x)/x impulse response reconstructs missing points perfectly, but an 8 times oversampling filter needs nearly a million taps for 16-bit accuracy. Head-Fi, Jul 2014
• 2014-08-30Watts says standard upsamplers have too few taps, so let Hugo do it; the exception is Red Reference's 18K WTA 4x upsampler, equivalent to 36,864 Hugo taps. Head-Fi, Aug 2014
• 2015-02-14Watts says ADC anti-aliasing filters need about 2K taps for -144 dB, ADC chip aliasing near 21 kHz is audible, and his pulse array ADC project has run five years. Head-Fi, Feb 2015
• 2015-03-01Watts says short filters cause transient timing errors of tens of microseconds; Hugo uses 26,000 taps, 164,000 still beats 100,000, and about a million taps would give 16-bit accuracy. Head-Fi, Mar 2015
• 2015-03-02Watts says the 164,000-tap filter runs on an LX75 FPGA after four redesigns for timing closure, using 160 DSP cores at half speed for the first WTA stage. Head-Fi, Mar 2015
• 2015-03-04Watts says FIR filters have no simple dB/octave figure, more taps narrow the gap to hi-res, but poor SRC from 96 kHz masters cannot be undone, so recordings vary. Head-Fi, Mar 2015
• 2015-04-06Watts says 24/192 still sounds better than 44.1/16 on Hugo, but its 26,000-tap WTA filter makes the gap much smaller than on other DACs, and AAC can sound surprisingly good. Head-Fi, Apr 2015
• 2015-05-20Watts says DAVE's FPGA is 10 times more complex than Hugo's, giving 164,000 taps, an enhanced WTA algorithm, FIR filtering to 256 times, then further filtering to 2048. HIFICLUBE.NET JVH 45
• 2015-06-18Watts explains the pulse array's analog-like distortion, why half-band filters keep original samples, and how a 256-tap WTA beat a 2048-tap half-band Kaiser filter in the late 90s. Head-Fi, Jun 2015
• 2015-06-18Watts says half-band filters time transients poorly (-6dB at FS/2); 16-bit sinc accuracy needs about 1M taps; Hugo is about 12 bits, Dave about 14. Head-Fi, Jun 2015
• 2015-06-27Answering what '12 bits accurate against the ideal' means, Watts says the WTA impulse response matches the sinc to 12 bits, and he has not yet hit the audible limit. Head-Fi, Jun 2015
• 2015-10-06Watts says the DSD+ filter could not be routed alongside the 164,000-tap PCM+ filter despite under 70% capacity, but the 82,000-tap low-latency filter is smaller. Head-Fi, Oct 2015
• 2015-11-24Watts says his new 64-core DSD filters could not be routed with the 164,000-tap WTA filter, hence two programs: PCM+ with Hugo DSD, and DSD+ with 82,000 taps. Head-Fi, Nov 2015
• 2015-12-11Watts rejects 'more than one way to skin a cat': transparency to live music requires very long tap-length sinc filters, and other designers fail to address timing and NFM. Head-Fi, Dec 2015
• 2016-01-29Watts says a single-stage 16FS WTA first stage sounded much better, by more than 26,000 versus 164,000 taps, and 256FS second-stage filtering improved timing accuracy from microseconds to nanoseconds. article, Jan 2016
• 2016-03-02Watts says 44.1/16 could in principle be near perfect, going from -330 to -350 dB noise shaping changed little, and 1M taps gives 16-bit sinc accuracy but needs costly FPGAs. Head-Fi, Mar 2016
• 2016-03-07Watts is quoted that simple filters in DAC ICs give timing errors around 100 microseconds on a transient burst; he says more FIR taps and processing accuracy are needed. article, Mar 2016
• 2016-03-07Watts says ordinary DAC ICs use 100 taps at 8x or 16x; the article gives DAVE 164,000 taps at 256fs on 166 DSPs, and DAC64 1,000, QBD76 18,000, Hugo 26,000 taps. article, Mar 2016
• 2016-03-30Watts says perfect reconstruction needs an infinite-tap, infinitely oversampling sinc filter; he does not know how many taps suffice, and Davina will let him test 768k to 48k and back. Head-Fi, Mar 2016
• 2016-07-24Watts says Hugo's quality came from perceiving the starting and stopping of transients, not the WTA filter, since earlier 32,000-tap filters lacked it (Hugo had 26,000); DAVE gets bigger FPGAs. Hifistatement 3:31
• 2016-07-24Watts says an infinite-tap FIR filter would perfectly recover the original; coefficients at 16-bit accuracy need a million taps, while conventional DACs have about a hundred. Hifistatement 1:26
• 2016-07-24Watts says studying sampling theory in the early 1980s showed him long tap lengths were needed to recover timing, which he knew mattered for perception, though he then hated digital. Hifistatement 0
• 2016-07-24Watts says longer taps with normal filters on DAC64 changed the sound greatly, but refining the filter algorithm improved timing accuracy, leading to WTA and a tenfold improvement in sound. Hifistatement 1:39
• 2016-07-24Watts says WTA aims to align transients to the original recording, was fine-tuned through thousands of listening tests, and a 256-tap WTA beat a 2000-tap conventional filter. Hifistatement 2:28
• 2016-07-24Watts says Hugo has about 26,000 taps, Mojo similar, and DAVE 164,000 taps on 166 DSP cores filtering to 2048 times; the 100,000-line WTA code repeatedly failed timing closure. Hifistatement 3:15
• 2016-07-24Watts says this mattered more than going from 26 thousand to 164 thousand taps, 16x filtering is accurate only to microseconds, and he added a 16x-to-256x WTA stage. Hifistatement 7:47
• 2016-10-16Watts argues musicality comes first and explains why Mojo and DAVE excel on transient timing (NOS errors, tap length, 1M taps), small-signal accuracy and noise floor modulation. Head-Fi, Oct 2016
• 2017-01-07Watts presents the Blu Mk2 slides: why ringing worries are mistaken, the 16-bit case for about 1M taps, delaying launch to reach 1M, and why the scaler is separate. Head-Fi, Jan 2017
• 2017-01-09Watts says in the early 1980s he believed digital needed 16-bit sinc accuracy and a million taps, thought that impossible, and the M scaler shows his intuition was right. Head-Fi, Jan 2017
• 2017-01-09Watts says listeners find the M scaler a bigger change than Dave, believes on another DAC it would be an ordinary tap-length gain, and wonders why sixfold taps transform sound. Head-Fi, Jan 2017
• 2017-01-10Watts contrasts a one-tap NOS filter's roughly 100 uS worst-case timing error with an ideal infinite filter, plans Davina round-trip files, and links 1M-tap 16FS filters to Fourier uncertainty. Head-Fi, Jan 2017
• 2017-01-23Watts says Dave's 26,000 to 164,000 tap step disappointed him, algorithm and 16FS mattered more, 0.5M taps was transformational, and explains reducing errors until inaudible with a biscuit-barrel analogy. Head-Fi, Jan 2017
• 2017-01-23Watts says 16-bit coefficient accuracy has no bearing on recording bit depth: timing errors arise in interpolation, and an ideal sinc filter with 16-bit values needs over 1M taps. Head-Fi, Jan 2017
• 2017-01-28Watts explains peak error, why 16-bit accuracy needs about 1M taps, why doubling sample rate only halves conventional filters' error, and that taps are not the whole story. Head-Fi, Jan 2017
• 2017-02-10Watts says both tap count and algorithm matter; in the 1990s a non-aliasing filter became the first WTA, and a 256-tap WTA beat a 2048-tap conventional filter. Head-Fi, Feb 2017
• 2017-02-13Watts defends his Mojo timing illustration as a simplification, targets non-linear timing uncertainty rather than ringing, says doubling taps halves the error, and notes WTA accommodates half-band decimation. Head-Fi, Feb 2017
• 2017-02-13Watts says interpolated transient timing depends on sample timing unless the filter is ideal, and plans a Davina null test with half-sample-delayed white noise to measure the error. Head-Fi, Feb 2017
• 2017-03-19Watts says quoted tap lengths cover only the WTA1 stage, so 49,152 taps apply to the 1FS to 16FS filter whether the 16FS or 256FS output is selected. Head-Fi, Mar 2017
• 2017-03-23Watts says the 49,152 WTA1 taps never change and exclude WTA2, refusing 'effective tap' marketing, and the HF filter matters more above 48 kHz in both filter modes. Head-Fi, Mar 2017
• 2017-04-27Watts says image upscaling like RAISR does not carry over to audio, where only transient timing must be recovered with longer filters, and plans Davina tests of 1M-tap losses. Head-Fi, Apr 2017
• 2017-05-16Watts notes the M scaler's 2.76 dB insertion loss and says 0.5M versus 1M taps, or video mode, changes depth greatly, hypothesizing coefficient truncation over reverb-length periods. Head-Fi, May 2017
• 2017-06-11Watts says DAVE sounds better via the M scaler not from a WTA fault, citing Verilog simulation and -301 dB tests, and video mode's two-thirds million taps shrinks the soundstage. Head-Fi, Jun 2017
• 2017-06-11Watts says a planned Davina test will give a measured number for the transient error, though not why taps above 0.66M improve the soundstage, which is perceptual. Head-Fi, Jun 2017
• 2017-07-25Watts says Hugo's WTA1 ran 1FS to 8FS at 26,368 taps; Hugo 2's goes 1FS to 16FS at 49,152 taps, then WTA2 to 256FS on white/green. Head-Fi, Jul 2017
• 2017-08-11Watts says the million-tap M scaler exceeds 16-bit accuracy, Hugo 2 takes 768 kHz dual BNC, better analog parts put DAVE ahead, and he flies with DAVE and Blu 2. Head-Fi, Aug 2017
• 2017-08-13Watts says Davina will test how many taps are enough and whether ADC bandwidth limiting matters, noting sinc reconstruction is required and his IIR decimation filter gives 300 dB rejection. Head-Fi, Aug 2017
• 2017-08-17Watts says chip DACs suffer noise floor modulation, jitter sensitivity, poor small-signal resolution and short filters versus Hugo 2's 49,152 taps, factors far outweighing passive part quality. Head-Fi, Aug 2017
• 2017-08-26Watts says audible timing errors are non-linear, not fixed delays; the M scaler's 1,015,808 taps match ideal sinc to 16 bits, Hugo 2 about 13; doubling taps adds a bit. Head-Fi, Aug 2017
• 2017-10-11Watts says WTA reconstructs transients regardless of origin; DAVE's 164,000-tap delay is about 0.2 s and 1M taps 1.4 s, and his RT60 explanation for depth is speculative. Head-Fi, Oct 2017
• 2017-10-11Watts says 16-bit accuracy means the WTA filter matches an ideal sinc to better than 16 bits, needing 1M taps, and the timing error is independent of source bit depth. Head-Fi, Oct 2017
• 2017-10-13Watts says reconstruction depends on two things, tap length and the coefficient algorithm, and explains that simply truncating a sinc function is a rectangular window. miceblue425 14:32
• 2017-10-13Watts explains what a tap is: in an FIR filter, data passes along a delay line and each coefficient multiplier taps into that line. miceblue425 19:44
• 2017-10-13Watts says doubling a Kaiser filter's tap length always improved bass, soundstage and timbre, but with no hope of more than a thousand taps he sought a better algorithm. miceblue425 21:21
• 2017-10-13Watts says his revised algorithm gave a tenfold improvement, a 256-tap WTA filter beating a 2000-tap Kaiser, yet going from 256 to 1000 WTA taps still helped. miceblue425 22:06
• 2017-10-13Watts says tap length is crucial and, with Moore's Law, tap counts rose: DAC64 1,000, QBD76 18,000, Hugo 26,000 and DAVE 164,000. miceblue425 22:51
• 2017-10-13Watts says every tap-length increase improved sound in the same ways: better bass, instrument separation and focus, soundstage and timbre. miceblue425 23:36
• 2017-10-13Watts says Blu 2's FPGA has 740 DSP cores and about 200,000 logic cells, enough for just over a million taps; he calls this first million-tap filter the M Scaler. miceblue425 24:38
• 2017-10-13Watts says he calculated that a 16-bit-accurate sinc function for CD needs about 700,000 taps, roughly a million, when microprocessors were primitive; he recalls a 2K oil-rig monitor. miceblue425 27:06
• 2017-10-13Watts says a million taps was then impossible, with no FPGAs, DSPs or PCs, so he saw it as digital's fundamental flaw and long planned for a million-tap filter. miceblue425 28:00
• 2017-10-13Watts says in DAVE's development he started from Hugo's 26,000-tap filter and designed a 164,000-tap one; two designs failed timing closure, which he explains. miceblue425 28:44
• 2017-10-13Watts says the DAVE filter took nine months and sounded only a bit better, so he expected little from the M Scaler and first built an easy half-million-tap version. miceblue425 29:37
• 2017-10-13Watts says the 512,000-tap filter compared with DAVE gave his biggest shock in 5 years, partly due to an improved WTA filter and moving from 8FS to 16FS. miceblue425 30:21
• 2017-10-13Watts says the FPGA could just fit a million taps; timing closure tested fine at double rate, but packing memory efficiently required a completely new filter architecture. miceblue425 31:06
• 2017-10-13Watts describes the million-tap architecture: about 132 modules, each with eight DSP cores, one data SRAM and one coefficient memory, since large memory blocks pack more efficiently. miceblue425 31:49
• 2017-10-13Watts says the modular design cut coding eightfold, and going from half a million to a million taps massively changed sound quality, in a way every M Scaler listener notices. miceblue425 32:31
• 2017-10-13Watts says the M Scaler deepens the soundstage enormously, which puzzles him because filter changes never altered depth before; depth depends on small-signal nonlinearity, which filters can avoid. miceblue425 33:17
• 2017-10-13Watts says a million taps gives 16-bit sinc reconstruction; 24-bit would need 256 million taps and minutes of delay, which will never happen and which he doubts is audible. miceblue425 40:07
• 2017-10-17Watts says taps are simply taps, rivals' short or minimum-phase filters misreconstruct transient timing, Blu 2 sounds profoundly better, and he predicts marketing pseudo tap counts. Head-Fi, Oct 2017
• 2017-10-19Watts confirms the WTA window forces coefficients beyond 1.4 seconds to zero, and the ignored sinc coefficients are much smaller than 16 bits, giving better than 16-bit accuracy. Head-Fi, Oct 2017
• 2017-10-20Watts says an infinite sinc needs infinite processing, more taps reconstruct timing better, yet the algorithm matters too; he spent much time adjusting WTA to beat conventional filters. Hifistatement 51
• 2017-10-20Watts says a 2000 tap Kaiser filter sounds worse than a 256 tap WTA filter, but with a given algorithm more taps always gave better sound in his tests. Hifistatement 1:43
• 2017-10-20Watts lists WTA tap lengths (DAC64 a thousand, QBD76 18 thousand, Hugo 26,000, DAVE 164,000) and says each increase improved bass definition, timbre, staging precision and note starts and stops. Hifistatement 0
• 2017-10-20Watts says the Xilinx XC7A200T, launched about three or 4 years ago, became buyable only about 18 months ago; he used it for the Davina project, and it allows a million taps for Blu 2. Hifistatement 0
• 2017-10-20Watts says sampling theory needs an infinite-tap sinc filter, that limited tap length gets reconstructed timing wrong, and that he calculated getting on a million taps for guaranteed 16-bit performance. Hifistatement 44
• 2017-10-20Watts says a million-tap filter was impossible in the 1980s, so he thought digital would fail, since only a million taps reconstructs timing with guaranteed 16-bit accuracy. Hifistatement 1:28
• 2017-10-20Watts says he did not expect a big sonic change from this project; moving DAVE from 26,000 to 164,000 taps took nine months and 3 redesigns over timing closure. Hifistatement 2:10
• 2017-10-20Watts says he expected only incremental gains, so designed a half-million-tap filter on the new FPGA in a couple of months, heard a huge change, and decided to use more taps. Hifistatement 3:36
• 2017-10-20Watts says the FPGA has only just enough processors and memory for a full million-tap filter, but he delayed the project to push on with a full million taps. Hifistatement 4:24
• 2017-10-20Watts says, coding alone, he designed blocks of 8 cores running as separate threads, and that going from half a million to a million taps made an enormous change. Hifistatement 1:22
• 2017-10-20Watts says longer taps improved bass definition, timbre and pitch as usual, but the million-tap M Scaler also opened up depth and soundstage beyond DAVE's. Hifistatement 2:03
• 2017-10-20Watts says the USB video mode, two-thirds of a million taps on the same architecture, collapses the soundstage, though architecture and small-signal work carried over from DAVE also matter. Hifistatement 2:57
• 2017-10-20Watts says small-signal integrity was verified by passing the -301 dB test, so he cannot explain why longer taps change depth, which he has tied to small-signal integrity. Hifistatement 3:43
• 2017-10-20Watts says depth concerns small-signal amplitude, not timing, and speculates that the million taps spanning 1.25 seconds, roughly a reverb time, may explain the depth change. Hifistatement 4:30
• 2017-10-20Watts hopes the Davina project will explain why extra taps change depth, calling the soundstage opening with the Blu an unexpected shock revealing a hole in his knowledge. Hifistatement 5:17
• 2017-10-20Watts concludes that the million-tap filter is by far the most advanced filter technology, others use a few hundred taps, and it gives 16-bit CD accuracy for the first time. Hifistatement 0
• 2017-10-20Watts says sinc ripples drop below 16-bit level near a million taps; doubling adds 1 bit. DAVE is about 13 bits, Blu beyond 16, NOS 1, MQA 3 or 4. Hifistatement 24
• 2017-10-20Watts rates a typical high-performance DAC at 6 or 7 bits accurate, Hugo 12, DAVE 13.5, Blu beyond 16, and says ringing is irrelevant because it comes from non-bandlimited impulses. Hifistatement 1:12
• 2017-10-22Watts says CPUs struggle even with DSD512 DoP, 'input related' tap figures are meaningless pseudo numbers, cascaded FIR stages differ from one flat stage, and only real taps count. Head-Fi, Oct 2017
• 2017-10-28Watts says going from Hugo's 26k to DAVE's 164k taps changed little until WTA algorithm tweaks did; for a given algorithm, doubling taps doubles accuracy against the ideal. Head-Fi, Oct 2017
• 2017-11-15Watts says the M scaler's WTA coefficients match an ideal sinc to 16 bits, needing about 1 million taps and the right algorithm, which PC up-samplers lack. Head-Fi, Nov 2017
• 2017-11-16Watts says he knew in the 1980s that a million taps would be needed to recover transients to 16-bit accuracy, then thought impossible. Head-Fi, Nov 2017
• 2017-12-05Watts says John Franks opposed Blu 2, but Watts wanted a million taps, expected a small gain, spent six months, and got a tremendous change, which only listening reveals. RMAF 58:11
• 2017-12-09Watts names DAVE, over 2 and a half years of work, and the M scaler on Blu 2, having envisioned needing 1,000,000 taps in the 1980s, as his hardest designs. article, Dec 2017
• 2017-12-11Watts says hi-res files need the same tap length, since doubling the sample rate cuts timing error area fourfold (2 bits) while the M scaler gives about a 16-bit improvement. Head-Fi, Dec 2017
• 2017-12-26Watts says his WTA filters roll off very fast: 49,152 taps reach -1dB within 75Hz of FS/2, the M Scaler's 1M taps within 4Hz, matching ideal sinc beyond 16-bit accuracy. Audiophile Style, Dec 2017
• 2017-12-27Watts says longer filters only help if accuracy against the sinc doubles with each doubling of taps; most algorithms differ too much from ideal sinc values to benefit. Head-Fi, Dec 2017
• 2018-01-12Watts says BluDave, his untuned reference, came from minimizing aberrations; going from 0.5M to 1M taps is audible, and Davina will show how far BluDave is from perfect. Head-Fi, Jan 2018
• 2018-01-14Watts says coefficient memory is the limit, so a mono mode would not double the taps, and Davina will establish how close the M scaler is to perfection subjectively. Head-Fi, Jan 2018
• 2018-01-16Watts says the ear/brain keeps proving more sensitive than expected; the M scaler gives better than 16-bit accuracy at 16FS, and Davina will show whether that is enough. Head-Fi, Jan 2018
• 2018-01-16Watts expects no further significant gains from the finely tuned WTA and M scaler, asking whether 20-bit (16,000,000 taps) or 24-bit (256,000,000 taps) accuracy is needed; 16M is extremely expensive. Head-Fi, Jan 2018
• 2018-01-16Watts says the M scaler is not 100% accurate but guarantees reconstruction of the original bandwidth-limited analog signal to better than 16-bit accuracy, or 99.9985%. Head-Fi, Jan 2018
• 2018-01-16Watts says filtering out aliased content does not help as the damage is done, and longer tap length benefits all recordings, from 1990s passive anti-aliasing to modern half-band ADCs. Head-Fi, Jan 2018
• 2018-01-16Watts says Dave would be about 99.988% accurate, yet connecting an M scaler to Dave is audible within 3 seconds, so these tiny errors are extremely important. Head-Fi, Jan 2018
• 2018-01-17Watts says the percentage figure was back-of-envelope, a different way of expressing 16-bit accuracy of the interpolation filter against the ideal sinc function, not to be taken too seriously. Head-Fi, Jan 2018
• 2018-01-18Watts says the 49,152-tap WTA 1 matches an ideal sinc to 12 bits, its coefficients diverging below that level; with the M scaler divergence occurs beyond 16 bits. Head-Fi, Jan 2018
• 2018-02-07Watts lays out the Hugo 2/Qutest signal path from 16FS WTA1 (49,152 taps) to 104 MHz pulse array, and explains noise-shaped truncation and the -301 dB test for depth. Head-Fi, Feb 2018
• 2018-02-21Watts says Hugo 2 is about 99.76% of an ideal sinc while conventional filters bear no similarity, yet adding an M scaler to Hugo 2 is a huge change. Head-Fi, Feb 2018
• 2018-02-23Watts recounts experimenting with FPGA interpolation filters from 1999, hearing gains from 128 to 256 to 2048 taps in bass definition, imagery, separation and timbre. MajorHiFi 8:38
• 2018-02-23Watts says 16-bit sinc accuracy needs a million taps, impossible in the 1990s, so he used listening tests to optimize his algorithm to recover timing more efficiently. MajorHiFi 9:17
• 2018-02-23Watts says a 256-tap WTA filter sounded better than a 2048-tap conventional half-band filter, at least a 10 to 1 improvement from the algorithm itself. MajorHiFi 10:05
• 2018-02-23Watts recounts DAC64 using four FPGAs and about 15 watts for a thousand taps, then Hugo reaching 26,000 taps with a huge gain; both were 8FS filters. MajorHiFi 10:27
• 2018-02-23Watts says a 10 times better FPGA let Dave reach 164,000 taps, and every tap-length increase improved sound quality. MajorHiFi 11:13
• 2018-02-23Watts says going from Hugo to Dave took nine months of coding yet gave an important but not night-and-day change, so he thought 164,000 taps neared the limit. MajorHiFi 11:29
• 2018-02-23Watts describes a Xilinx FPGA announced in 2012 but available about 18 months ago, with 740 DSP cores, 16 megabits of memory and 215,000 logic cells, enabling a million taps. MajorHiFi 12:02
• 2018-02-23Watts recounts calculating in the 1980s that a 16-bit-accurate sinc function for CD needs a million taps, then impossible, making a million taps a lifetime goal. MajorHiFi 14:33
• 2018-02-23Watts says after Dave's modest gain he chose 512,000 taps for the Blu upgrade as easy to place and route; it took about three months and sounded completely different. MajorHiFi 15:26
• 2018-02-23Watts says a separate, galvanically isolated M Scaler keeps that noise out of Dave, isolation took two prototypes, another M Scaler is in progress, and timing recovery reaches 16-bit accuracy. MajorHiFi 20:49
• 2018-03-06Watts says Hugo 2/Qutest use 67.5mS of data versus 1mS for conventional filters, minimum phase sounds soft, the M scaler uses 1.37 seconds, and more M scalers will follow. Head-Fi, Mar 2018
• 2018-03-06Watts says guaranteed 16-bit reconstruction of the original analog signal needs a sinc accurate to 16 bits, meaning a delay over 0.5S, whatever the DAC architecture. Head-Fi, Mar 2018
• 2018-03-08Watts gives sinc-equivalent accuracy as 12.2 bits for Hugo 2, 13.9 bits for Dave and 16.6 bits for the M scaler, below which WTA coefficients diverge from the ideal sinc. Head-Fi, Mar 2018
• 2018-03-17Watts says ideal reconstruction needs an infinite-tap sinc; DAVE's WTA matches ideal to 13.9 bits and the M scaler to 16.6 bits, an unexpectedly large audible step. Head-Fi, Mar 2018
• 2018-03-20Watts says complex USB gear adds RF brightness, optical is the benchmark, and gives WTA overload margin, M scaler latency (630 ms, video mode 104 mS) and 16.6-bit accuracy. Head-Fi, Mar 2018
• 2018-03-31Watts says he is an independent consultant, not a Chord employee, working with Chord since 1996; increasing taps improved accuracy from 8 to 9 bits in 1999 to 16-bit today. article, Mar 2018
• 2018-04-04Watts says that in the 1980s sampling theory told him a 1,000,000-tap filter was needed to guarantee 16-bit performance, then inconceivable; today the M scaler has 1M taps. article, Apr 2018
• 2018-04-04Watts says DAC chips have few FIR taps, perfect reconstruction needs infinite taps, and a brain-matched filter needs 1,000,000, realized last year in WTA M scaler technology (1,015,808 taps). article, Apr 2018
• 2018-04-04Watts lists tap lengths: DAC64 (1999) 1,024, Hugo (2013) 26,000, DAVE (2015) 164,000, Blu MKII (2017) 1,015,808; each increase plus WTA algorithm improvements raised sound quality. article, Apr 2018
• 2018-04-04Watts expects desktop models to stay ahead; on taps alone, Hugo 2 and Qutest have 49,152 versus Blu MKII's 1,015,808, which he calls 'a 20 times capacity advantage.' article, Apr 2018
• 2018-04-04Watts repeats that a filter matched to the brain needs 1,000,000 taps, says Blu MKII achieved 1,015,808, and says the longer tap length substantially improves sound quality. article, Apr 2018
• 2018-06-09Watts says doubling a WTA filter's tap length doubles reconstruction accuracy, but how much accuracy the ear needs is open; Davina will show how close the M scaler is. Head-Fi, Jun 2018
• 2018-06-09Watts says image removal alone is not enough: filters with near-identical frequency response sound different, closer to sinc sounds better, and 0.5 million to 1 million taps is clearly audible. Head-Fi, Jun 2018
• 2018-07-04Watts says 15M taps is impossible for lack of memory, 4M would be doable but costs thousands of USD, so a cost-efficient Artix successor is needed. Head-Fi, Jul 2018
• 2018-07-18Watts recounts WTA development from 1980s sinc ideals, finding sinc closeness beat image rejection alone, and wonders whether Davina's alias-free decimation will change the importance of tap length. Head-Fi, Jul 2018
• 2018-07-19Watts says modern ADCs' -6dB at FS/2 causes unrepairable aliasing, yet all recordings benefit from an M scaler; Davina will decimate without aliasing, and he will release randomly named files. Head-Fi, Jul 2018
• 2018-07-19Watts says it will identify losses from ideal 768 decimation then M scaling back up; the M scaler exceeds 16-bit recovery, but whether that suffices is open. needs context Head-Fi, Jul 2018
• 2018-07-21Watts says a sinc filter pre-rings and post-rings forever, needing infinite processing and delay, so one workaround is to shorten the sinc function. c w 12:28
• 2018-07-21Watts says conventional filters use only a few hundred taps because designers check only frequency response without listening tests, and he realized the algorithm could use resources better. c w 13:12
• 2018-07-21Watts says his WTA algorithm beat a conventional filter with a tenth of the taps, was fine-tuned through thorough listening tests, and needed changes as tap length grew. c w 13:58
• 2018-07-21Watts calls WTA a unique windowing function applied to the sinc, and starts a tap-length history: DAC64 in 1999 had a thousand taps using three FPGAs and 15 watts. c w 14:48
• 2018-07-21Watts lists QBD76 at 18,000 taps, Hugo 1 at 26,000 and Dave at 164,000, saying each increase improved timbre, bass pitch and left-right soundstage. c w 15:36
• 2018-07-21Watts traces the million-tap goal to 1981 at Cardiff University, where he read hearing physiology on transients and learned that finite-tap filters leave errors under Shannon's theory. c w 16:17
• 2018-07-21Watts says he reasoned the errors would be in timing, and that reconstructing 16-bit audio to 16-bit accuracy needs roughly a million taps by back-of-envelope calculation. c w 17:04
• 2018-07-21Watts says a million taps was impossible in 1981, so he favored vinyl then; WTA listening tests followed, and the Xilinx A200T now gives 740 DSP cores at 12 watts. c w 17:48
• 2018-07-21Watts says a million taps now run at the same power as DAC64's thousand in 1999, and at a million taps the WTA coefficients become identical to a sinc function. c w 18:36
• 2018-07-21Watts says the coefficients match sinc to the 16-bit level, guaranteeing reconstruction of the bandlimited signal to 16-bit accuracy for CD, then begins describing FIR problems on Dave. c w 19:24
• 2018-07-21Watts explains FPGA timing closure and placement, saying Dave began at 200,000 taps, failed to place and route, and finally worked at 164,000 (spoken as 'pounds'). c w 20:10
• 2018-07-21Watts says Dave's filter took nine months and was only modestly better than Hugo's; half a million taps on the new FPGA sounded hugely, unexpectedly better than Dave. c w 21:02
• 2018-07-21Watts says the transformational change and good timing closure led him to the full million taps, a goal dating 'back in 1991' (as spoken), requiring a new memory-efficient filter architecture. c w 21:47
• 2018-07-21Watts sums up: high-end DACs use a few hundred taps, the M Scaler has 10,000 times more processing, recovering 16-bit CD to 16-bit accuracy and transient timing to unprecedented accuracy. c w 27:04
• 2018-07-21Watts says past WTA changes never altered depth, and suspects the 1.3-second analysis span, similar to concert-hall RT60 times, explains it; he hopes Davina will show why. c w 28:33
• 2018-07-21Watts says the M Scaler recovers the original analog signal to better than 16-bit accuracy, then opens the floor to questions. c w 29:20
• 2018-07-21Watts says it is not job done: the gain from half a million to a million taps suggests going further, which Davina's native, bandlimited and decimated files will test. c w 31:47
• 2018-07-21Watts says three files will be published; if the difference is huge he will do 10 or 16 million taps, though part of him would like the project closed. c w 32:33
• 2018-07-21Watts says selecting 384 kHz halves the taps output; 768 kHz gives 1.3 uS resolution, while 1536 would halve the data window to 0.63 seconds; 768 is the listening-test optimum. Head-Fi, Jul 2018
• 2018-07-26Watts posts his CanJam presentation: transients cue placement, pitch and timbre; conventional filters mistime them; 1M WTA taps match sinc to 16.6 bits using 528 DSP cores. Head-Fi, Jul 2018
• 2018-07-26Watts says the M scaler reconstructs to better than 16 bits; after the big change from 0.5M to 1M taps he cannot say whether more taps help; Davina will show. Head-Fi, Jul 2018
• 2018-07-26Watts says Qutest, Hugo 2, TT2 and Dave share the post-scaler process (WTA2 to 256FS, third-order filter to 104 MHz), notes diminishing returns, and previews DX amps' transparency. Head-Fi, Jul 2018
• 2018-07-27Watts doubts vendors' software tap claims such as Audirvana's 500,000 taps, says PCs are noisy and poor at real time, and that filter structure alters sound. Head-Fi, Jul 2018
• 2018-08-02Watts says Blu 2 and M scaler filter code is identical, explains that upscaling recovers transient timing rather than adding information, and why pulse array needs DSD converted to PCM. Head-Fi, Aug 2018
• 2018-08-03Watts says a standard PC lacks the CPU power for 1M taps though a GPU PC could, but it still would not be the WTA algorithm, which matters most. Head-Fi, Aug 2018
• 2018-08-04Watts explains TT2's 98,304-tap 16FS WTA1, says WTA2 matches Hugo 2 and Dave, TT2 sounds warmer than Hugo 2, and describes fixing 192kHz crackles in Windows. Head-Fi, Aug 2018
• 2018-08-04Watts says Mojo's 768 kHz works on USB only; going from 0.5M to 1M taps is a big change, so audition the M scaler with Mojo against a Hugo 2. Head-Fi, Aug 2018
• 2018-08-05Watts says Mojo gains from 0.5M taps but upgrading Mojo may be better, and Mojo lacks an extra SPDIF decoder because power mattered and no M scaler existed then. Head-Fi, Aug 2018
• 2018-08-06Watts says the M Scaler is fundamentally 3 bits more accurate than the TT or DAVE, guaranteed to recover the original analog signal to better than 16 bits. Pursuit Perfect System 4:13
• 2018-08-19Watts says a claimed 10,000,000-tap filter has a very poor transition band that a few thousand taps could match; Mojo's WTA is 10 times better, and algorithm beats tap length. Head-Fi, Aug 2018
• 2018-08-19Watts says an 11,000,000-tap WTA filter would match the ideal sinc to 20-bit accuracy, so about 16M taps would be needed for better than 20 bits. Head-Fi, Aug 2018
• 2018-08-19Watts posts a quickly made 10M-tap conventional Kaiser filter with a transition band under 2 Hz, over 1000 times sharper than the test file. Head-Fi, Aug 2018
• 2018-08-31Watts says the original is the unsampled analog signal; the M scaler's million taps match ideal sinc to better than 16 bits, recovering transient timing regular filters get wrong. Head-Fi, Aug 2018
• 2018-09-01Watts explains that 16-bit accuracy refers to the ideal sinc, errors arise only with transients, NOS is crude sample-and-hold, and delta-sigma ADCs suffer half-band decimation aliasing. Head-Fi, Sep 2018
• 2018-09-02Watts says the M scaler matches a perfect sinc to 99.98%, the worst case for any signal; a 1kHz sine would be 99.9999996% accurate, transients giving larger errors. Head-Fi, Sep 2018
• 2018-09-24Watts says any varying transient timing error is audible however small, citing audible gains from half a million to one million taps and 350 dB over 330 dB noise shapers. Head-Fi, Sep 2018
• 2018-10-23Watts explains the M scaler matches sinc to 16.7 bits, fixing transient timing, but cannot repair aliasing from modern half-band ADCs, possibly why 1960s recordings sound amazing. Head-Fi, Oct 2018
• 2018-11-11Watts explains the M scaler's apparently slower tempo as reduced dominance effect, because 1M-tap WTA cuts program-dependent transient timing errors; 164,000 to 1M taps is very audible. Head-Fi, Nov 2018
• 2018-11-28Watts says a random timing error would need 24 bits beyond 1M taps, but it is program-correlated so matters more than dither; video mode uses an asymmetric filter. Head-Fi, Nov 2018
• 2018-12-13Watts says the algorithm strongly affects sound: his first WTA filter in 1999 used a thousand taps, the FPGA limit, versus 100 conventionally, and algorithm changes improved transients. Pursuit Perfect System 6:44
• 2018-12-13Watts says the WTA algorithm was refined from theory and hundreds of listening tests, so that a 256-tap WTA filter sounded better than a 2048-tap conventional filter. Pursuit Perfect System 7:30
• 2018-12-13Watts gives tap lengths of DAC64 (1,000), QBD76 (18,000), Hugo in 2013 (26,000) and DAVE (164,000), saying each increase improved bass definition and soundstage. Pursuit Perfect System 9:06
• 2018-12-13Watts says Mojo is described as having 500 times more processing power, gives its previously unpublished tap length as 38,000 taps, and credits that processing for a 400 pound device. Pursuit Perfect System 10:21
• 2018-12-13Watts recounts concluding from Shannon theory and psychoacoustics that conventional filters create audible timing errors, and estimating that 16-bit CD performance needs nearly a million taps, impossible with 1980s processors. Pursuit Perfect System 11:42
• 2018-12-13Watts says he felt digital would never sound good until a million taps were possible, and begins explaining why with a sinc response decaying to zero. Pursuit Perfect System 12:33
• 2018-12-13Watts says the sinc function falls below 16 bits at about three quarters of a million taps, so the 1 million tap M Scaler reaches 16.7 bits. Pursuit Perfect System 13:16
• 2018-12-13Watts attributes the M Scaler's stronger instrument power without cross-modulation to removing the timing modulation that past and future signal causes in conventional filters, restoring 16-bit accuracy. Pursuit Perfect System 16:26
• 2018-12-13Watts says longer taps improved bass, timbre and left-right staging but never depth, which he ties to small-signal accuracy, yet at half a million taps the soundstage opened up. Pursuit Perfect System 19:49
• 2018-12-13Watts says he cannot explain why going from DAVE's 164,000 taps to the M Scaler's 1 million opens the soundstage, suggesting its 1.3 second window matches concert-hall reverberation time. Pursuit Perfect System 20:30
• 2018-12-17Watts says 2 V is 3.5 dB and 1 V 9.5 dB below 3 V; non-WTA upsamplers cannot predict the M Scaler, as longer conventional filters add near-zero coefficients. Head-Fi, Dec 2018
• 2019-01-27Watts says the M scaler makes CD equivalent to a 705 kHz recording to 16 bits; with native 705 input both it and DAVE's 164,000-tap WTA1 pass through. Head-Fi, Jan 2019
• 2019-03-04Watts explains white/green adds a WTA 16FS-to-256FS stage replacing an IIR filter, why sinc accuracy needs long taps, and says preferring orange suggests an overly bright system. Head-Fi, Mar 2019
• 2019-03-21Watts explains what an FIR tap is, says the M Scaler matches an ideal sinc to better than 16 bits needing 1M taps, and doubling WTA taps doubles accuracy. Head-Fi, Mar 2019
• 2019-03-25Watts comments on upsampling experiments: stopband beyond -150 dB matters little, output rate should suit tap length, his Davina decimation reaches 250 dB, and truncation needs aggressive noise shaping. Head-Fi, Mar 2019
• 2019-04-03Watts says his 768k recording resembles M scaled 44.1k, doubling the rate cuts the error area fourfold, and R2R or 2.8 MHz noise-shaped DACs limit accuracy above 768k. Head-Fi, Apr 2019
• 2019-04-09Watts says ringing is inaudible because 22.05 kHz is inaudible and bandlimited signals carry no energy there, calling the industry's Dirac-impulse view a misunderstanding of sampling theory. Head-Fi, Apr 2019
• 2019-04-19Watts defines tap length, explains the WTA window leaves just over half the coefficients exactly sinc, guaranteeing reconstruction better than 16 bits, and notes a typo in Keith Howard's article. Head-Fi, Apr 2019
• 2019-05-02Watts gives the M scaler's 65 C thermal trip, says it recovers transient timing to 16-bit accuracy rather than adding information, and attributes optical's darker sound to RF, not jitter. Head-Fi, May 2019
• 2019-06-14Watts says Davina hit hardware and decimation-filter problems, argues audible timing errors are non-linear ones, and cites WTA 2 tests as evidence of sensitivity below the 7-4 uS interaural limit. Head-Fi, Jun 2019
• 2019-07-22Watts says the M scaler converts 48k to 768k with 16-bit sinc accuracy, each doubling of rate quarters the transient error area, and WTA2 timing gains above 768k remain audible. Head-Fi, Jul 2019
• 2019-09-09Watts says M scaler accuracy depends on input, far beyond 24 bits on sine waves, at least 16.7 bits on transients; DAVE 2 is unplanned and DAVE took 2.5 years. Head-Fi, Sep 2019
• 2019-09-10Watts says doubling WTA taps adds one bit of sinc accuracy, and 88.2/96k input runs the same coefficients as an 8-times filter, keeping accuracy and processing unchanged. Head-Fi, Sep 2019
• 2019-09-14Watts explains the trade-off between output rate and sinc accuracy at fixed taps, says 16FS is optimal for 1M taps, and rejects offline 16FS conversion over file size and metadata. Head-Fi, Sep 2019
• 2019-11-28Watts says a longer-tap scaler will not replace the Hugo M Scaler until much better FPGAs exist, with no sign from Xilinx, and expects pulse array ADC progress next year. Head-Fi, Nov 2019
• 2019-12-14Watts tells a TT2 owner every DAC upsamples, the M scaler matches sinc to 16 bits for accurate transients, blind tests are insensitive, and he uses Aeon Closed on planes. Head-Fi, Dec 2019
• 2019-12-19Watts says only a sinc function removes the error; in 1980 he calculated that 16-bit accurate reconstruction needs more than half a million, three quarters of a million, taps. Passion for Sound 10:57
• 2019-12-19Watts recalls that in 1980 there were no CDs, DSPs or FPGAs, only punch cards and an 8-bit microprocessor with 2K memory, making three quarters of a million taps impossible. Passion for Sound 11:25
• 2019-12-19Watts says the timing error degrades timbre, soundstage and bass, which improve as taps increase; Qutest's 48 thousand taps are about 12-bit sinc accurate. Passion for Sound 19:15
• 2019-12-19Watts says adding an M Scaler gives the full million taps, with coefficients identical to sinc to 16.7 bits, guaranteeing better than 16.7-bit reconstruction accuracy. Passion for Sound 19:56
• 2019-12-19Watts says 16-bit Red Book arguably needs timing reconstructed only to 16 bits, and that 16.7-bit filter accuracy is not DAC accuracy: a steady 24-bit sine wave comes out 24-bit. Passion for Sound 20:18
• 2019-12-19Watts says it is wrong to claim that a filter only 16-bit accurate yields only 16 bits of output. Passion for Sound 21:10
• 2019-12-19Watts says required accuracy is found by making filters more sinc-like and listening; his first WTA filter, DAC64 in 1999, had 1,000 taps, the M Scaler one million. Passion for Sound 22:05
• 2019-12-19Watts says WTA was shaped by the maths, removing image aliasing, and hundreds of listening tests tuning the taper; his first test pitted 256-tap WTA against 2,000-tap Kaiser. Passion for Sound 24:25
• 2019-12-19Watts says the 256-tap WTA filter sounded much better than the 2,000-tap Kaiser filter, at least a tenfold improvement in transient reconstruction. Passion for Sound 25:10
• 2020-03-27Watts says the M scaler gives a natural sense of flow and less fatigue because loud instruments no longer modulate transient timing, about 10 times more accurate than DAVE alone. Head-Fi, Mar 2020
• 2020-04-09Watts says mains conditioners neglect RF, a single-blind test tied a new HF filter's warmth to noise floor modulation, and DAVE is about 13-bit accurate, M scaler over 16. Head-Fi, Apr 2020
• 2020-04-09Watts says USB decoding and galvanic isolation work down to 3.3v VBUS, and the gain from better-than-16-bit over 13-bit timing varies by listener, but he needs the M scaler. Head-Fi, Apr 2020
• 2020-06-25Watts names 1 million taps and the M Scaler, though a thumb-sized battery running 12 watts for eight hours would make a tiny device very hot. Part-Time Audiophile 30:56
• 2020-06-25Watts says doubling the sample rate quarters the timing error, only 2 bits, while the M Scaler matches a sinc function to better than 16 bits to reconstruct timing. Part-Time Audiophile 39:38
• 2020-11-11Watts compares 1M-tap rectangular sinc, Kaiser and WTA filters by measurement and listening, explains how the WTA equation was tuned, and concludes algorithm matters more than tap length. Head-Fi, Nov 2020
• 2020-11-14Watts says an infinite-tap sinc guarantees correct intermediate values for bandlimited signals, and WTA is far more accurate per tap, with doubling taps doubling accuracy; improvement has not yet converged. Head-Fi, Nov 2020
• 2020-11-27Watts advises never upsampling before his DACs and says M Scaler research at 1M taps showed the WTA algorithm is 50 to 500 times more effective than conventional filters. Head-Fi, Nov 2020
• 2021-01-03Watts says the ear/brain responds to errors roughly as log(log), citing 164k to 1M taps and depth error steps, and ranks interpolation filter, DAC design and analog design. Head-Fi, Jan 2021
• 2021-05-23Watts says PGGB uses an apodizing, slow roll-off filter rather than sinc, so long taps cannot help its transient timing, and true offline sinc sounds like the M scaler. Head-Fi, May 2021
• 2021-06-02Watts says PGGB uses a slow roll-off apodizing filter for 44.1k, so many taps cannot fix transient timing; samples he heard sounded flatter, softer and less separated. Head-Fi, Jun 2021
• 2021-06-02Watts quotes PGGB's FAQ on apodizing and says a slow roll-off interpolation filter cannot reconstruct bandlimited signals and seriously degrades transient timing, making long taps pointless. Head-Fi, Jun 2021
• 2021-06-04Watts says what matters is how many taps are Whittaker-Shannon, about 256M for 24-bit recovery; windowed or apodized sinc fails, and 64-bit float causes noise floor modulation. Head-Fi, Jun 2021
• 2021-07-24Watts says resolution means resolving small signals, not fixed noise (21.3 bits for Dave); Dave resolves -186dB; and 1M taps' 16.7-bit accuracy now looks insufficient for transients. Head-Fi, Jul 2021
• 2022-01-05Watts says benefits of longer taps diminish markedly at higher sample rates, 768 kHz seems the sweet spot, so future more costly M scalers will stay at 705/768. Head-Fi, Jan 2022
• 2022-01-06Watts says the M scaler is 1M taps at 705/768k so only half the taps are used at 352/384, and that tape bias and amplifiers also degrade transient timing. Head-Fi, Jan 2022
• 2022-01-31Watts ties transient timing to the WTA filter: 1 million taps in M Scaler, 40 thousand in Mojo 2 using forty DSP cores per channel; samples are 22 microseconds apart. Darko.Audio 9:24
• 2022-01-31Watts says the change is evolutionary: original Mojo was 38 thousand taps, he improved the filter and updated its internal architecture, and the improvements are subtle. Darko.Audio 12:58
• 2022-01-31Watts says two or three years ago he found amplitude-dependent phase shift is audible as suppressed depth, while investigating why 1 million taps beat DAVE's 164 thousand for depth. Darko.Audio 24:34
• 2022-01-31Watts says DSP core speed is irrelevant and taps count; Mojo 2 has four fifths of Hugo 2's tap length, no WTA2, and taps are only part of sound quality. Head-Fi, Jan 2022
• 2022-02-11Watts says his high-end M Scaler uses a Choral (DAVE) chassis and is not 4M taps, algorithm outweighs taps, and gives SINAD of 125dB (Mojo 2) and 127dB (TT2). Head-Fi, Feb 2022
• 2022-03-19Watts says the WTA filter makes the best of limited processing, began in 1999 with DAC64 at a thousand taps, and explains what a tap is. MFG Audio Reviews 21:09
• 2022-03-19Watts says doubling taps with the same algorithm doubles reconstruction accuracy, the ear is extremely sensitive to timing errors, and the WTA filter was optimized by many listening tests. MFG Audio Reviews 22:03
• 2022-03-19Watts says each doubling of sample rate cuts the error area fourfold, 256 times less at 705 kHz, the 16FS sweet spot, and WTA2 is less tap-length sensitive than WTA1. MFG Audio Reviews 52:02
• 2022-04-26Watts says at 44.1 kHz samples are 22 microseconds apart, the interpolation filter must reconstruct transient timing between them, and perfect reconstruction needs infinite taps. Audio Force 9:59
• 2022-04-26Watts says typical 64 or 128 tap filters make transients constantly shift early or late; Mojo 2's WTA uses 40 DSP cores per channel and over 40,000 taps. Audio Force 10:44
• 2022-04-26Watts says the WTA filter minimizes timing modulation but cannot eliminate it, nor can an M Scaler; his algorithm maximizes accuracy with limited taps, and Mojo's FPGA is custom coded. Audio Force 11:37
• 2022-05-14Watts says WTA taps split between future and past; M scaler video mode uses 98,304 future and 507,904 past taps (606,208 total); ASIO native USB allows DSD512, with fewer pops. Head-Fi, May 2022
• 2022-05-23Watts says DAVE's DX outputs carry 705/768 kHz from its 164,000-tap WTA1 filter and should sound better than Qutest's 48k taps; Qutest ignores the embedded volume data. Head-Fi, May 2022
• 2022-07-14Watts says the M Scaler is essential, PGGB sounded like an apodizing filter, and the M Scaler reconstructs transient timing beyond 16 bits, passing -301 dB via dual BNC. Head-Fi, Jul 2022
• 2022-08-19Watts says a back-of-envelope sum showed that reproducing transient timing to 16-bit accuracy, with 22 microseconds between CD samples, needs getting on for a million taps of a sinc function. AVstore 3:16
• 2022-08-19Watts says a million taps would give 16-bit accuracy, calls this digital's major flaw, and says it was impossible in the early 1980s, when he programmed with punch tape. AVstore 3:56
• 2022-08-19Watts says he used FPGAs with off-board filters until 1999, when a Xilinx FPGA allowed 1,000 taps versus about 128 commercially, and he began developing the WTA filter. AVstore 5:30
• 2022-08-19Watts says as FPGAs grew he increased tap length and improved the WTA algorithm, each doubling of tap length improved the sound, and the 200T device arrived around 2015. AVstore 6:11
• 2022-08-19Watts says going from DAVE's 164,000 taps to the Hugo M Scaler's 1 million made a bigger difference than expected, opening the soundstage, and that 16-bit accuracy is not enough. AVstore 6:55
• 2022-08-19Watts says a disclosed scaler to match DAVE, three or four years in work and priced like DAVE, won't replace the Hugo M Scaler; more taps help, but algorithm matters. AVstore 7:51
• 2022-08-19Watts says that as tap length increases the sound improves, and there may come a point where more taps give no audible improvement, meaning the job is done. AVstore 12:02
• 2022-08-19Watts says required accuracy came out a very silly number, and his ADC project will let him decimate 768 kHz recordings to 48 kHz, rescale to 768 kHz, and compare. AVstore 12:31
• 2022-08-19Watts says every upsampler contains a filter, even NOS (a first-order hold), and the two key factors are the number of taps, which sets reconstruction accuracy, and the algorithm. AVstore 14:39
• 2022-08-19Watts says the algorithm calculates the coefficients and is free to design, while tap count is limited by hardware, since each tap needs memory, a multiplier and an accumulator. AVstore 15:27
• 2022-08-19Watts agrees, saying lots of taps with a conventional algorithm will not sound good. AVstore 20:56
• 2022-08-21Watts says WTA uses five variables tuned by thousands of listening tests, the algorithm matters more than tap length, and a two-sample linear interpolation filter would be a disaster. Head-Fi, Aug 2022
• 2022-08-22Watts explains a tap as a sample times a WTA coefficient, summed in an accumulator; the M Scaler uses 63,488 original samples per interpolated sample, going 1FS to 16FS. Head-Fi, Aug 2022
• 2022-08-29Watts says filters make transient timing errors program dependent, which is why 100k and 1M taps sound different, that audibility does not fade with age, and that prototypes are delayed. Head-Fi, Aug 2022
• 2022-09-18Watts says an NVIDIA device's negative video delay lets him run the M Scaler at 1 million taps, a bigger M Scaler has one-second delay working, and he dislikes surround. The Source AV 4:38
• 2023-01-16Watts turns DAVE's HF filter off with the M scaler, doubts a Hugo 2-priced portable scaler-DAC by 2032, sets crossfeed to max, and says more taps add depth. Head-Fi, Jan 2023
• 2023-02-04Watts says he worked out in the 1980s that reconstructing to 16-bit accuracy needs more than half a million taps, which then seemed impossible. Dusty Audio 39:52
• 2023-02-04Watts says 1990s FPGAs let him use 1000 taps against 128 in off-the-shelf filters, explains that taps measure processing, and says it sounded a lot better. Dusty Audio 40:34
• 2023-02-04Watts says bass pitch is perceived from the pluck transient, and going to 1000 taps improved bass pitch, timbre variation and the starting and stopping of notes. Dusty Audio 41:22
• 2023-02-04Watts says 1000 taps fell far short of the million needed for 16-bit accuracy, so he looked at changing the coefficient algorithm instead of standard Kaiser or other windows. Dusty Audio 42:07
• 2023-02-04Watts says others use standard math and trust their brand; with the M Scaler's million taps he can guarantee 16-bit reconstruction accuracy between samples. Dusty Audio 43:46
• 2023-02-04Watts asks whether 16-bit accuracy is enough, says he has spent five years exploring how many taps are needed, and is building his own ADC because ADC quality limits recordings. Dusty Audio 44:38
• 2023-03-26Watts says most perceptual cues come from transient timing, which needs a sinc filter whose coefficients only converge on zero, so the filter must keep doubling for accuracy. Eric's Hi-Fi 1:26
• 2023-03-26Watts says reconstructing 16-bit data to 16-bit accuracy needs more than half a million taps, hence a million, a target based on calculation rather than listening tests. Eric's Hi-Fi 2:11
• 2023-03-26Watts says the limit is reached when doubling taps is inaudible; during Covid he investigated it while working on a scaler to match DAVE, stressing reconstruction accuracy over taps. Eric's Hi-Fi 3:09
• 2023-03-26Watts says investigating why the windowing was so sensitive showed how many taps are needed before it stops mattering: ten to the power of seventeen. Eric's Hi-Fi 4:45
• 2023-03-26Watts says that tap count means a WTA function spanning two billion years past and future, which is impossible; he found an improvement to reveal with the new scaler. Eric's Hi-Fi 5:21
• 2023-03-26Watts says his listening tests show the tiniest transient reconstruction error is audible, so reconstruction needs extreme accuracy, a harder problem than he expected in the 1980s. Eric's Hi-Fi 6:55
• 2023-03-26Watts says he was the first to talk about large tap counts, but what matters is the algorithm itself, not the taps. Eric's Hi-Fi 9:13
• 2023-04-03Watts says it will certainly have more taps, and that the one benefit of COVID was two years of not running around, which let him spend time on fundamental research. needs context Amplify Audio 1:41
• 2023-04-03Watts says the question he asked himself was how many taps are needed before reconstruction is perfect. Amplify Audio 2:18
• 2023-04-03Watts declines to give a tap count because thermal considerations may change it, but says it is substantially more than the M Scaler's. needs context Amplify Audio 2:29
• 2023-04-03Watts says what else has gone into it matters more, because it is not just about taps but about the algorithm itself. Amplify Audio 2:49
• 2023-04-03Watts says he has researched why these things matter and how many taps are needed for the ultimate sound, and it turns out to be an awful lot. Amplify Audio 3:01
• 2023-04-03Watts says before the M Scaler he thought depth depended only on small-signal amplitude; raising WTA tap length had never changed depth before, suggesting timing matters for depth. Amplify Audio 7:04
• 2023-09-19Watts says only one filter set minimizes transient timing error, and the ESS9038 Pro's 128 taps use only 8 original samples, inadequate for transient reconstruction. Head-Fi, Sep 2023
• 2024-04-08Watts rebuts a claim that more taps make transients collide, saying longer sinc filters make transients independent, and each tap doubling improved sound. Head-Fi, Apr 2024
• 2024-05-10Watts says this is 4 million taps, followed by an unclear phrase transcribed as 'there's like two million big'. Reactcore 18
• 2024-05-10Watts says it is not about bandwidth but about timing accuracy: getting transients to be reconstructed. needs context Reactcore 24
• 2024-05-10Watts says no simple number can be given because two issues are going on, one derived from the sinc function and another problem. needs context Reactcore 54
• 2024-05-10Watts begins to say that for the sinc-function-derived issue you can put a number on it. needs context Reactcore 1:02
• 2024-05-10Watts says the sinc function issue can be given a number, '16 bit kind of as long as it's 20 bit', but the other problem differs. Reactcore 1:05
• 2024-05-16Watts says the project was also research on interpolation; Hugo M Scaler had sinc performance to 16-bit, a target he thought unreachable when he identified the flaw in the 1980s. ConvinceMeAudio 7:44
• 2024-05-16Watts says 16-bit is an arbitrary target; his research found even infinitely small timing errors are perceptible, which led him to improve the interpolation filter in Quartet. ConvinceMeAudio 8:31
• 2024-05-16Watts says development took long because of engineering complexity and because it was a research exercise into interpolation filters: how many taps are needed to perfectly reconstruct. needs context The Speaker Shack 45
• 2024-05-17Watts says the algorithm matters much more than the number of taps, and it has been optimized to reconstruct timing much more accurately than before. AVstore 4:40
• 2024-06-18Watts answers that it is named Quartet because it has four million taps. needs context article, Jun 2024
• 2024-06-18Watts says the M Scaler is guaranteed better than a 16-bit reconstruction and Quartet better still, and explains how the brain uses transient timing for separation, location, timbre, note edges and bass pitch. article, Jun 2024
• 2024-07-12Watts says in the 1980s he calculated that 16-bit timing accuracy needs nearly a million taps for a 16 times oversampled filter, the key to the Hugo M Scaler. HIFICLUB 9:54
• 2024-07-12Watts describes programming with paper tape and says a million taps was then impossible, so he concluded digital would always sound bad until DSPs and FPGAs arrived. HIFICLUB 10:47
• 2024-07-12Watts says FPGAs could run interpolation filters by 1999 and a million taps by 2015, leading to the Hugo M Scaler; his role was more taps and a better algorithm. HIFICLUB 11:31
• 2024-07-12Watts says thousands of listening tests found the WTA variables' sweet spot, which suits everyone's brain, and the work has been more taps plus tuning toward 16-bit performance. HIFICLUB 13:56
• 2024-07-12Watts says the M Scaler's 1 million taps improved pitch, timbre and staging; Quartet uses more FPGA power for 4 million taps and an improved filter. HIFICLUB 14:45
• 2024-07-12Watts says Quartet taught him that the Hugo M Scaler's better-than-16-bit reconstruction is not enough; much better is needed, with effects too small for measurement equipment. HIFICLUB 20:23
• 2024-07-18Watts says 1M taps gave 16-bit accuracy but any timing error is perceptible, describes the isolated 75 ohm coax output, and says flight-ready Quartet needs better FPGAs. Head-Fi, Jul 2024
• 2025-05-12Watts introduces his slide on tap length history. Audiophile Junkie 13:28
• 2025-05-12Watts lists tap counts: DAC64 1,000 (1999), QBD 18 thousand, Hugo 26 thousand (2013), DAVE 164 thousand, Hugo M Scaler 1 million; more taps and WTA improvements each improved sound. Audiophile Junkie 13:35
• 2025-05-12Watts lists benefits of more taps and WTA tweaks (bass, staging, separation, fatigue, note starts and stops); at 1 million taps the Hugo M Scaler reaches identical sinc values. Audiophile Junkie 14:19
• 2025-05-12Watts says the sinc coefficients now have 16-bit accuracy, guaranteeing reconstruction better than 16-bit accuracy, while the best DACs available reconstruct to only two or 3 bits. Audiophile Junkie 15:08
• 2025-05-12Watts asks what 1 million taps gives you. Audiophile Junkie 15:34
• 2025-05-12Watts says reconstruction exceeding 16-bit accuracy transforms perception of pitch, melody, timbre and space, and its biggest benefit is a better emotional connection to the performance. Audiophile Junkie 15:41
• 2025-05-12Watts explains the delay: the more of the audio the filter sees, and the more coefficients are sinc values, the more accurately it reconstructs transient timing. needs context Audiophile Junkie 17:21
• 2026-03-17Watts says a DAC must rebuild a continuous waveform with accurate transient timing, which improves timbre, separation, location and pitch, but the filters needed enormous processing impossible in the 1980s. ecoustics 2:18
• 2026-03-17Watts says he was right that reconstruction is crucial but wrong to think 16-bit reconstruction accuracy would give perfection because recordings were 16-bit. ecoustics 3:13
• 2026-03-17Watts says he now knows reconstruction accuracy must be much better than 16 bits. ecoustics 3:42
• 2026-03-17Watts explains that an interpolation filter places a transient between samples slightly early or late, modulating back and forth and confusing location, timbre and pitch; he guessed 16-bit accuracy sufficed. ecoustics 4:05
• 2026-03-17Watts says 16-bit reconstruction accuracy leaves an error below 16 bits, but he later found that even the tiniest timing reconstruction error has audible consequences. ecoustics 4:49
• 2026-03-17Watts says reconstruction must fill in everything between samples to match the original analog from the microphone into the ADC, within 16-bit accuracy. ecoustics 6:10
• 2026-03-17Watts says all 2,048 created samples must match the original analog to 16-bit accuracy. ecoustics 9:03
• 2026-03-17Watts says leaving the data unfiltered gives a distorted signal with stepped edges unlike a 20 kHz sine; in 1980 he calculated 16-bit reconstruction needed well over half a million FIR taps. ecoustics 10:35
• 2026-03-17Watts says FIR filters did not then exist, analog reconstruction filters were very poor because they cannot see the future, and he thought digital was fatally flawed. ecoustics 11:41
• 2026-03-17Watts explains an FIR filter delays samples, multiplies each by a coefficient and sums them, and that a million-tap WTA filter reconstructs to 16-bit accuracy. ecoustics 12:22
• 2026-03-17Watts says WTA is his filter type guaranteeing 16-bit reconstruction, and that 'tap' comes from tapping old analog delay lines, though memory is used today. ecoustics 13:12
• 2026-03-17Watts says only sinc coefficients reconstruct perfectly, a million-tap sinc filter guarantees better than 16-bit reconstruction, and the Hugo M Scaler achieved that. ecoustics 13:56
• 2026-03-17Watts says DAVE's filter is 164 thousand taps; an M Scaler adds a million taps, converting 48 kHz to 768 kHz (16 times), and DAVE then upsamples to 2048 times. ecoustics 23:40
• 2026-03-24Watts says each of the WTA equation's five factors had to be set by ear to within ten parts per million, and he did not understand why a million taps was so sensitive. Audio Force 10:45
• 2026-05-03Watts says the algorithm is much more important than the number of taps. Audio Force 8:03
• 2026-05-03Watts says the WTA algorithm alone is designed for transient timing accuracy, unlike filters built for frequency response or stopband; for a given algorithm, doubling taps halves timing error. Audio Force 8:16
• 2026-05-03Watts says each doubling of taps halves the error, improving location, separation, timbre and low-frequency pitch, and that emotional engagement with music matters most. Audio Force 9:10
• 2026-08-30Watts says that after the Hugo M Scaler, more taps (1,000 in 1999) and a better WTA kept sounding better; a million taps gives 16-bit reconstruction, but more is needed. The Source AV 6:00
• 2026-08-30Watts says research into reconstructing transient timing perfectly showed the number of taps needed is much bigger than 1 million, and asks the host to guess. The Source AV 6:47
• 2026-08-30Watts says a filter with 28 zeros after the four would have a delay of 900 trillion years, 66,000 lifetimes of the universe, so that level is unreachable. The Source AV 7:27
• 2026-10-02Watts says every module must reproduce a -301 dB signal within a thousandth of a dB or depth suffers; in 2017 he took Blu 2 from 0.5M to 1M taps. HiFi Buys 1:24:04
• 2026-10-02Watts says going from half a million to 1 million taps gave a profound depth change, whereas earlier WTA improvements had affected timbre, separation and imagery, never depth. HiFi Buys 1:25:05
• 2026-10-02Watts says a tap is part of an FIR, finite impulse response, filter. HiFi Buys 1:48:43
• 2026-10-02Watts says one tap is one access to memory, a name from early analog delay lines that were tapped into to change the delay. HiFi Buys 1:48:50
• 2026-10-02Watts says each delayed signal is multiplied by a coefficient and summed with the others, and adding all the taps gives the new interpolated sample. HiFi Buys 1:49:21
• 2026-10-02Watts says more taps approach the ideal infinitely sharp sinc; claims that its ringing is wrong misunderstand sampling theory, which requires zero energy at FS/2. HiFi Buys 1:50:29
• 2026-10-02Watts says ironically a band-limited impulse rings more through a short-tap-length filter than through a long one. HiFi Buys 1:52:00
• 2026-10-02Watts says this filter rings for three seconds with three seconds of pre-ringing because it looks at six seconds of music, yet a properly band-limited impulse shows no ringing. HiFi Buys 1:52:19
• 2026-10-11Watts says his research from 2018 asked how many taps give subjectively perfect reconstruction, and why his five-variable WTA coefficient formula had a sweet spot. Distorsiuni Interioare 17:04
• 2026-10-11Watts says the WTA filter handles two conflicting factors together, and that the taps needed came to 4 with twenty-six zeros, with a delay of 900 trillion years. Distorsiuni Interioare 19:19
• 2026-10-11Watts says that delay is 66 thousand times the universe's lifetime, so tap length could not keep growing; his Blackbird WTA processes the two factors separately. Distorsiuni Interioare 20:07
• 2026-10-14Watts says low tap count interpolation filters reconstruct transients slightly too early or too late. Part-Time Audiophile 27:59
• 2026-10-14Watts says more taps give better reconstruction, all else equal; a sinc filter has infinite taps; a tap multiplies a buffered sample by a coefficient. Part-Time Audiophile 28:54
• 2026-10-14Watts explains the name came from tapping into analog delay lines, and that each tap multiplies a sample by a precomputed coefficient, all summed to give the interpolated output. Part-Time Audiophile 29:43
• 2026-10-14Watts says more taps approach an ideal sinc, but the algorithm matters far more: a billion taps with the wrong algorithm would sound worse than a thousand. Part-Time Audiophile 30:14
• 2026-10-14Watts says the coefficient algorithm matters most, but with a given algorithm, doubling the taps and adjusting accordingly improves sound quality. Part-Time Audiophile 30:58
• 2026-10-14Watts says ESS chips have half a dozen filters and at most a few hundred taps, versus 4 million on Quartet and, he thinks, about 44 thousand on Mojo 2. Part-Time Audiophile 33:10
• 2026-10-14Watts says Quartet began about 8 years ago after the 1 million tap Hugo M Scaler, which guarantees 16-bit accurate reconstruction; 6 years of research went into improving interpolation filters. Part-Time Audiophile 52:36
• 2026-10-17Watts says after the M Scaler he wanted to know how many taps give the best sound, and whether it would take 10 million or 100 million. article, Oct 2026
• 2026-10-17Watts describes tuning the five WTA equation variables to within 10 parts per million by listening, finding a sweet spot with the million taps. article, Oct 2026
• 2026-10-17Watts says he put a number on when the effects disappear, used it to compute the taps needed for a good approximation to the original signal, and asks how many that is. article, Oct 2026
• 2026-10-17Watts says the answer is much, much bigger than four million taps. needs context article, Oct 2026
• 2026-10-17Watts says the delay would be 900 trillion years, or 66,000 universe lifetimes, and that this only reaches good reproduction, not the best. needs context article, Oct 2026
• 2026-10-17Watts says Blackbird WTA improves sinc reconstruction accuracy tenfold, eliminates the other problem and cuts the delay from 900 trillion years to tens of seconds; the algorithm, not four million taps, is the story. article, Oct 2026
• 2026-10-17Watts clarifies that the age-of-the-universe figure is the time for the signal to get through the filter. needs context article, Oct 2026
• 2026-10-17Watts says yes, still, because the limit is now tens of seconds, and an Ultima scaler will be a future project. needs context article, Oct 2026
• 2026-10-17Watts says companies claiming more taps use FFT convolution, whose Fourier frequency resolution degrades transient timing, and he is skeptical because FFT sizes are not published. article, Oct 2026
• 2026-10-17Watts says the FPGA gives needed precision and a direct FIR filter without FFT tricks, and the algorithm matters far more than tap count, as Blackbird and WTA show. article, Oct 2026