How do Chord DACs handle DSD, and does Rob consider DSD better or worse than PCM?
Short answer
Every Chord DAC Watts has described filters DSD into PCM inside the DAC instead of turning the 1-bit stream straight into analog Head-Fi, May 2018Head-Fi, Apr 2014. In Hugo 2, TT2, Blu Mk2 and the Hugo M Scaler (from 2017 to 2018 onward) that is one shared filter with better than 200 dB of rejection that outputs 705.6 kHz PCM Head-Fi, Jan 2018Head-Fi, May 2018Head-Fi, May 2018Head-Fi, May 2021; Mojo 2 (2022) uses the original Mojo DSD filter instead because the Hugo 2 filter would not fit Head-Fi, Feb 2022. On the formats themselves, his position since 2015 is that PCM is better: he agreed DSD is "fundamentally flawed" because its 1-bit noise shaper loses small signals and adds amplitude-dependent timing errors that cannot be removed afterward, while PCM is limited by the DAC, not the format Head-Fi, Sep 2015Head-Fi, Jan 2020Head-Fi, Nov 2024. In 2014 his view was already critical but more mixed: he credited DSD with better innate timing and said that without a WTA filter he would "prefer DSD every time", while also calling DSD "severely limited" and saying red-book, properly implemented, would outperform it Head-Fi, Apr 2014Head-Fi, Apr 2014Head-Fi, Jul 2014article, Aug 2014.
How Chord DACs handle DSD: the common approach
Watts's stated rule is that DSD is always filtered into PCM before conversion; of TT2 in May 2018 he wrote that it processes native DSD "like all my DAC's - in that it gets filtered" Head-Fi, May 2018. In April 2014 he explained that his simple single-stage analog output needs very low RF noise, and "DSD is at -20dBFS at 100kHz", so "the DSD is filtered, which converts it to regular PCM" Head-Fi, Apr 2014. In May 2019 he gave three reasons: re-running a DSD modulator is not transparent (so volume and crossfeed must be done in PCM); regular DSD cannot be converted to analog without adding significant distortion, noise floor modulation and small-signal degradation, so it is converted to PCM for the pulse array DAC; and DSD's HF and RF noise must be filtered, which he says is more transparent to do digitally Head-Fi, May 2019.
In September 2019 he said "It would be trivially easy for me to do a native DSD DAC conversion using pulse array", and declined because native DSD conversion is "horribly sensitive to clock jitter" and would need complex analog filtering that degrades transparency Head-Fi, Sep 2019. In June 2022 he called pure DSD DACs "pretty useless, as you can't do EQ or volume control digitally without converting to PCM" Head-Fi, Jun 2022. At CanJam NYC 2023 (late February; video published March 2023) he put it as "there's never an argument for keeping DSD and running it as a DSD only DAC" The Headphone Show 52:31.
He frames the DSD filter as doing the same job as his PCM filters, recovering the original analog signal "without the hideous out of band RF noise and nth order harmonics distortion" that DSD introduces (May 2018) Head-Fi, May 2018. After filtering, he says, Hugo 2 "treats it exactly the same as regular PCM" (January 2018) Head-Fi, Jan 2018.
By product and period
- Qute (described April 2014): a moving average filter; the signal was upsampled, not decimated, with 50 dB of rejection at 100 kHz Head-Fi, Apr 2014.
- Hugo (2014): because Hugo does volume and crossfeed, DSD is converted to 16FS PCM through a decimating filter with 140 dB of rejection (110 dB at 100 kHz); he reported it sounded "very much better" than the Qute-style filter in his AB tests Head-Fi, Apr 2014Head-Fi, Jun 2014. It ran on two dedicated custom DSP cores, and he described it as in essence converting DSD back to the original DXD master Head-Fi, Jun 2014Head-Fi, Jun 2014. Hugo accepts DSD via DoP only Head-Fi, May 2016. He keeps the 0 to -6 dBFS DSD overload margin and raises the volume internally by 6 dB to compensate Head-Fi, Feb 2015. In June 2018 he said this first decimation filter had aliasing performance better than -130 dB and that this was found to be audible; in May 2019 he gave it as -130 dB rejection with worst-case aliasing of -150 dB Head-Fi, Jun 2018Head-Fi, May 2019.
- DAVE (2015): two FPGA programs, because both would not fit at once: PCM+ (164,000-tap WTA with Hugo's DSD filter) and DSD+ (82,000-tap PCM, a non-decimating DSD filter using 64 DSP cores, and DSD512 support) Head-Fi, Oct 2015Head-Fi, Nov 2015Head-Fi, Nov 2015. At the May 2015 Munich show he said the non-decimating filters followed criticism of his decimating Hugo filter HIFICLUBE.NET JVH 3:20. In December 2017 he said DSD playback on DAVE measures identically to the digital-domain performance of the DSD64 signal Head-Fi, Dec 2017. In the M Scaler thread in July 2018 he wrote "I always set Dave to PCM+ mode", noting that with a 768k input the signal path is the same in both modes Head-Fi, Jul 2018.
- Hugo 2 (2017 to 2018): a new filter with better than 200 dB attenuation that converts to 705.6 kHz PCM, with aliasing better than -240 dB; he says the analog output measures identically to the original DSD64 test tones measured digitally, so performance is limited by DSD itself Head-Fi, Jan 2018Head-Fi, Jan 2018. He designed it to match the abilities of DAVE's DSD+ mode, and said it removes the need for a DSD+ mode and works for all formats up to DSD512 Head-Fi, Mar 2017Head-Fi, Jan 2018.
- TT2, Blu Mk2 and Hugo M Scaler (2018 onward): the same filter; TT2 needs no dedicated DSD mode "as the gap has now been closed" Head-Fi, May 2018Head-Fi, May 2018Head-Fi, Aug 2018. The M Scaler always outputs 705.6 kHz PCM from DSD64 to DSD512, and only via the dual BNC outputs Head-Fi, May 2018Head-Fi, Mar 2019. In May 2021 he said the M Scaler and TT2 DSD code is identical, and in July 2022 that M Scaler DSD handling is identical to Hugo 2 and TT2 Head-Fi, May 2021Head-Fi, Jul 2022.
- Qutest (2018 to 2019): in the Qutest thread in August 2018 he said "the DSD decimation filter actually sounds a bit better than the DSD+" (the post does not name Qutest explicitly) Head-Fi, Aug 2018. Over USB it takes native DSD via ASIO up to DSD512 and DoP up to DSD256 Head-Fi, May 2018; it decodes DoP over S/PDIF, with DSD64 the maximum on optical Head-Fi, Jul 2019.
- Mojo 2 (February 2022): he had hoped to use the Hugo 2/TT2 DSD filter, but "it would not fit", so Mojo 2 uses the original Mojo DSD filter, adding that Mojo 2's HF filter helps a great deal; DSD is converted to 705.6 kHz PCM and fed to the UHD DSP so EQ, crossfeed and volume work identically Head-Fi, Feb 2022Head-Fi, Feb 2022.
- Quartet M Scaler (podcast published October 2026): at the Quartet launch event in Romania (the date is not recorded), in an interview, he said the DSD decimation "follows the the decimation process that I've used in Quartet", going from 64FS to 16FS, and that DSD sources sound a lot better because of a much more powerful decimation Distorsiuni Interioare 56:10Distorsiuni Interioare 57:07. He gave no rejection figures for it in this archive.
Decimating versus non-decimating. From 2018 on he has consistently preferred the decimating Hugo 2/M Scaler filter to DAVE's non-decimating DSD+, because its attenuation above 88.2 kHz (220 dB against 130 dB) "certainly outweighs any possible loss that the decimation introduces" Head-Fi, May 2019Head-Fi, Nov 2018Head-Fi, Oct 2020. He repeated this in March 2023 and July 2024, adding in 2024 that with a Hugo M Scaler and DAVE the M Scaler route "easily sounds the best for DSD" Head-Fi, Mar 2023Head-Fi, Jul 2024. In June 2018 he said a non-decimating filter of the Hugo 2 filter's performance would be 32 times larger Head-Fi, Jun 2018.
Interfaces. DoP and native DSD carry identical data; native transmission via ASIO reaches DSD512, while DoP reaches DSD256, and ASIO is Windows only, so Linux gets DoP only Head-Fi, Sep 2019Head-Fi, May 2018. In February 2022 he said DoP up to DSD256 is fully supported on Linux and macOS, with DSD512 limited to Windows via ASIO Head-Fi, Feb 2022. In a TT2 thread post (dated only to between March 2025 and March 2026) he said DoP is supported on all inputs subject to their rate limits: optical at best DSD64, coax DSD64 and DSD128 Head-Fi, Mar 2025.
Is DSD better or worse than PCM?
2014: critical, with qualifications. In April 2014 he set out pros and cons: PCM has excellent small-signal resolution but innate timing limits that a long interpolation filter can fix, while DSD has poor resolution but "does not innately have timing problems" Head-Fi, Apr 2014. He added "if I were not using a DAC with a WTA filter, I guess I would prefer DSD every time", and advised "if the recording started as DSD, stick to DSD and vice versa" Head-Fi, Apr 2014. In June 2014 he compared CD and DSD versions of Pink Floyd's Wish You Were Here on Hugo and scored CD 2-0, while stressing "it's only my personal opinion based on one recording" Head-Fi, Jun 2014. In July 2014 he wrote "As to DSD - if this is the future, then we are in for trouble. DSD is severely limited in its technical performance", citing a loss in transparency and sound stage depth against the DXD originals on the 2L site Head-Fi, Jul 2014. In an August 2014 interview he said he had heard very good DSD, particularly '70s classics converted straight from analog tape to DSD, and also awful DSD, but was "sure that red-book, properly implemented, will out perform DSD" article, Aug 2014article, Aug 2014.
2015 onward: DSD is flawed by the format. In September 2015, agreeing with another poster, he wrote "DSD is fundamentally flawed, and unlike PCM where the DAC is the fundamental limit, its in the format itself" Head-Fi, Sep 2015. He gave two reasons. First, resolution: DSD64's noise shaper manages about 120 dB, and any signal below the noise shaper floor is lost, unlike dithered PCM Head-Fi, Sep 2015Head-Fi, Oct 2016. Second, an amplitude-dependent delay: small transients take longer to pass the 1-bit quantizer than large ones, which he saw in simulation as "several uS for DSD64" Head-Fi, Sep 2015. He called this timing problem one that "has never been talked about before, but its something I have been aware of for a long time"; it differs from his April 2014 statement that DSD has no innate timing problem Head-Fi, Sep 2015Head-Fi, Apr 2014. In March 2016 he said 44.1/16 is "easily better than DSD IMHO", adding that this applies on his DACs Head-Fi, Mar 2016.
He has given different sizes for that timing error over time: "tens of microseconds" in March 2017 Head-Fi, Mar 2017; "twenty or 30 microseconds" for a -60 dB signal at CanJam RMAF in October 2017 (video published December 2017) RMAF 38:21; "of the order of tens of uS" in September 2018 Head-Fi, Sep 2018; and "10 microseconds worth of per phase shift errors" at CanJam NYC 2022 (late February; video published March 2022) MFG Audio Reviews 53:59.
He ties these two problems to what he hears: "DSD sounds unnaturally soft (due to the timing errors) and flat (poor depth due to poor small signal resolution)" (August 2017) Head-Fi, Aug 2017. His fullest list (November 2018) adds clock-jitter sensitivity, signal-correlated switching activity and modulator instability causing noise floor modulation, which he says "can't be solved, and never can be solved" Head-Fi, Nov 2018. He says these losses cannot be recovered by any DAC: "Lost or distorted signals are not recoverable" (August 2018), and in May 2021 "it is impossible to repair the damage that DSD does" Head-Fi, Aug 2018Head-Fi, May 2021. In December 2017 he said DSD sounds as good on DAVE as it ever will, but that because DAVE handles PCM so much better, DSD's innate limitations become more obvious Head-Fi, Dec 2017.
2018: less vigorously against. In May 2018 he wrote that the Hugo 2 DSD filter was even better in performance, "so much now that I am not so vigorously against DSD as before", adding "But I still easily prefer redbook" Head-Fi, May 2018. In August 2018 he conceded "a bad PCM would likely to be not be as good as a perfect DSD (recorded straight to DSD, no PCM, no editing)" Head-Fi, Aug 2018.
On other DACs. He separates the format from the converter. In January 2020, agreeing with a poster, he said a DSD file will probably sound better than PCM on a modest non-WTA chip DAC, but not on Hugo 2, because "DSD is format limited, but PCM is conversion limited" Head-Fi, Jan 2020. At CanJam NYC 2022 he said "you put DSD on a crappy DAC, it sounds okay", while PCM on a capable DAC "is potentially unlimited in performance" MFG Audio Reviews 54:40MFG Audio Reviews 55:24.
DSD rates. He says higher rates help but do not fix the format. In May 2016 DSD512 "still sounds poorer than redbook on Dave" Head-Fi, May 2016. In May 2021 "all speeds have SQ problems": going from DSD64 to DSD256 improves soundstage depth and impact, "but it's nothing like as good as PCM via WTA" Head-Fi, May 2021. In February 2022 "good 16 bit 44.1 is better than DSD 512 via the M scaler", with the hedge that he has only one DSD512 track Head-Fi, Feb 2022. In June 2024 he called DSD256's noise at 400 kHz "still completely unacceptable" Head-Fi, Jun 2024. Most recently, after building test DSD modulators, he said they showed "how unacceptable DSD is from a technical POV, irrespective of DSD speed" (TT2 thread, March 2025 to March 2026) Head-Fi, Mar 2025.
What he credits DSD with. He says DSD's single element avoids the resistor mismatch of R2R DACs, so its distortion falls as the signal falls, and that DSD needs no sinc interpolation filter (CanJam NYC 2023) The Headphone Show 19:51The Headphone Show 22:11. He began his October 2017 RMAF DSD slide by saying he loved DSD, explaining that this referred to 1994, before he invented pulse array Head-Fi, Oct 2017RMAF 33:54. He used PDM DACs (which he says are the same as DSD256) before pulse array, and says he invented pulse array to solve DSD's problems Head-Fi, Feb 2019; in May 2018 he said he developed it because "I wanted the musical benefits of PDM or DSD, but with more transparency" Head-Fi, May 2018. In 2019 he said "DSD was actually a backward step in terms of how you convert from digital to analogue" Passion for Sound 6:13.
Current position (2024 to 2026). In May 2024 he told a poster "where I disagree with you is DSD - this sounds nothing like real acoustic instruments playing in space" Head-Fi, May 2024. In July 2024 he said he has a 4TB disk of DSD music but did not use it much, "as the soft sound with poor depth got to me" Head-Fi, Jul 2024. In November 2024: "Correct, I do see DSD as fundamentally flawed from a technical POV" Head-Fi, Nov 2024. In a podcast recorded in October 2025 (published March 2026) he called DSD "really poor as an archiving format" ecoustics 28:09, and in a podcast published in October 2026 (recorded at the Quartet launch event in Romania, date not recorded) he said its losses are baked into the DSD process, "And that's why I'm opposed to DSD" Distorsiuni Interioare 58:38.
What he recommends to listeners
Send the DAC the recording in the format it was mastered in: "If DSD, don't let the source convert it to PCM" (November 2024) Head-Fi, Nov 2024. For SACD, if the master is DSD stick with it, but since a DXD or other PCM master is more likely, then stick with PCM (January 2020) Head-Fi, Jan 2020. Do not convert PCM to DSD, which he says degrades depth, detail and timing Head-Fi, Feb 2017Head-Fi, Jan 2018, and do not re-modulate DSD64 to DSD128/256/512 before the M Scaler Head-Fi, Jan 2020. He prefers the DSD-to-PCM conversion done inside the DAC rather than in a PC Head-Fi, May 2019. In February 2023 he described DoP DSD as a quick bit-perfect test, adding "We're not recommending you listen to DSD" Dusty Audio 1:05:36.
What he has not said: this archive has no rejection or aliasing figures for the original Mojo DSD filter that Mojo 2 uses, no explicit statement that Qutest uses the Hugo 2 filter, and only one remark on how Quartet treats DSD Head-Fi, Feb 2022Head-Fi, Aug 2018Distorsiuni Interioare 56:10.
Quotes
“So I can't put raw DSD into the DAC, or it will sound very hard. So the DSD is filtered, which converts it to regular PCM.”2014-04-16 Head-Fi, Apr 2014
“DSD is fundamentally flawed, and unlike PCM where the DAC is the fundamental limit, its in the format itself”2015-09-25 Head-Fi, Sep 2015
“…I am not so vigorously against DSD as before. But I still easily prefer redbook…”2018-05-04 Head-Fi, May 2018
“my advice is always send my DACs the original recording in the format that it was mastered in. If DSD, don't let the source convert it to PCM”2024-11-11 Head-Fi, Nov 2024
Dated positions
| When | Products | Position |
|---|---|---|
| 2014-04 to 2014-08 | Hugo (original), Qute (original) | DSD has poorer resolution but no innate timing problem; he called it severely limited and said red-book properly implemented will outperform it, though without a WTA filter he would prefer DSD, and listeners should stay with the master's format. Head-Fi, Apr 2014Head-Fi, Apr 2014Head-Fi, Jul 2014article, Aug 2014 |
| 2014-04 to 2014-06 | Hugo (original), Qute (original) | DSD is filtered to PCM: Qute with a non-decimating moving average filter, Hugo with a 140 dB decimating filter to 16FS so volume and crossfeed can work. Head-Fi, Apr 2014Head-Fi, Jun 2014 |
| 2015-09 onward | Hugo (original), DAVE, Hugo 2, Hugo M Scaler | DSD is fundamentally flawed by the format itself: limited noise shaper resolution loses small signals and amplitude-dependent delays corrupt transient timing, neither recoverable. Head-Fi, Sep 2015Head-Fi, Nov 2018Head-Fi, Aug 2018 |
| 2015-10 to 2018 | DAVE | DAVE offers PCM+ (164k taps, Hugo DSD filter) or DSD+ (82k taps, non-decimating 64-core DSD filter, DSD512). Head-Fi, Oct 2015Head-Fi, Nov 2015 |
| 2017 to 2024 | Hugo 2, Hugo TT2, Blu Mk2, Hugo M Scaler | A new >200 dB decimating filter converts DSD64 to DSD512 to 705.6 kHz PCM; he prefers it to DAVE's DSD+ because better attenuation outweighs decimation losses. Head-Fi, Jan 2018Head-Fi, May 2018Head-Fi, May 2019Head-Fi, Jul 2024 |
| 2018-05 to 2018-08 | Hugo 2 | With the new filter he is less vigorously against DSD, still easily prefers red-book, and concedes a bad PCM may lose to a perfect native DSD recording. Head-Fi, May 2018Head-Fi, Aug 2018 |
| 2020-01 to 2022-02 | Hugo 2 | DSD is format limited and PCM conversion limited, so DSD may sound better on a modest chip DAC but not on his DACs. Head-Fi, Jan 2020MFG Audio Reviews 54:40 |
| 2022-02 | Mojo 2 | Mojo 2 uses the original Mojo DSD filter because the Hugo 2 filter would not fit, and converts DSD to 705.6 kHz PCM. Head-Fi, Feb 2022Head-Fi, Feb 2022 |
| 2024 to 2026 | Quartet M Scaler, Hugo TT2, Mojo 2 | DSD remains fundamentally flawed and unacceptable at every DSD speed; he is opposed to it and advises sending DACs the original master format. Head-Fi, May 2024Head-Fi, Nov 2024Head-Fi, Mar 2025Distorsiuni Interioare 58:38 |
Caveats and open points
- The size of DSD's amplitude-dependent timing error varies across his statements (several uS in 2015, tens of uS in 2017 and 2018, 20 to 30 uS at CanJam RMAF 2017, 10 uS in 2022); the RMAF and CanJam NYC figures come and should be checked against the recordings.
- His figure for DSD64 noise shaper performance also varies: about 120 dB in posts (2015, 2016, 2018), about 126 dB at CanJam NYC 2023 and 130 dB in a 2024 Korean interview The Headphone Show 22:11HIFICLUB 39:59.
- Head-Fi, Mar 2025 (the test-modulator remark and TT2 DoP limits) is dated only to between 2025-03-27 and the 2026-03-03 scrape.
- The Quartet DSD remark (October 2026) is a single passage and its wording is ambiguous about whether it describes the Quartet M Scaler itself.
- Head-Fi, Aug 2018 is in the Qutest thread but does not name Qutest; that Qutest uses the Hugo 2 DSD filter is not stated outright in this archive.
- Head-Fi, Mar 2017 (March 2017) says "DSD 256 benefits more than DSD 64, exactly like PCM" in a reply about Hugo 2's filter settings; it most likely refers to benefit from the HF filter, not to DSD256 sounding better, so it is not used as evidence on DSD rates.
- The CanJam RMAF 2017 talk was given in October 2017; its archived video is dated 2017-12-05.
- The Weekly ASCII article (2016) summarizing him as saying PCM is 'by far' better is a journalist's paraphrase in Japanese with machine translation, so it is not quoted as his words.
- Dating of talks: the CanJam NYC 2022 and 2023 videos were published in March, buts the talks were in late February (show dates unverified); the ecoustics podcast was recorded 2025-10-28 and published March 2026; the Romanian podcast was published October 2026 and recorded at the Quartet launch event on a date the archive does not record.
Topics: DSD: playback and DSD vs PCM