Not just the Digital to Analog conversion that makes a difference but also how the Analog output of the DAC is implemented. IMHO up sampling contributes to a less transparent digital sound. Probably why most Ladder DACs sound more analog like.
Why most $3000 and lower DAC’s sound almost identical
I have a theory as to why all modern DACs essentially sound so similar these days, making it difficult to differentiate between them. IMO modern Delta Sigma chips have homogenized DACs into close to the same sound, making it very easy to take any DAC under $3000 and find it will sound good as another.
What I have discovered is that ladder R2R DACs and fully discrete DSD DAC’s are creating a better soundstage and less digital “glare”. An observation supported by countless others - nothing new. Anything with a Delta Sigma chip-based DAC that does oversampling will have less soundstage and more glare.
Nothing new so far - most of you will likely agree that that the above is a common consensus but here is the new bit, so read on if you are curious…
The dissatisfaction with this sound has led to a band-aid solution where Delta Sigma DAC manufacturers now offer a plethora of filters from sharp to smooth, linear phase to minimum phase. All of this is hand waving nonsense that offers a band aid to what is an absolutely fundamental design issue.
FUNDAMENTAL DESIGN ISSUE:
All oversampling with Delta Sigma offers superb measured spec at very low cost - it’s the logical choice for anyone using Precision test equipment to design a DAC. Typical chip filters use about 60 taps in their filters. They also ALL use Parks-McLellan filter designs (which has best “spec” and the short tap length is required for low-latency and easy processing). The result is a filter that has equiripple through the entire pass band. Mathematically it is a fact that an equiripple in the frequency domain equates to two echoes in the time domain - a pre-echo and post-echo. The “digital glare” heard is because of these echoes, likely the pre-echo is most audible. Our ears brain are processing the echos because unlike noise they are a complete reflection of the entire audio signal - low in level but lasting long enough to be detected by our acuity to locate the source of a sound. It is the same reason our speakers sound and image much better when moved out into the room and away from any close proximity to reflective surfaces. Despite these echoes being 60 db down from the primary signal, my listening sessions have convinced me of their audibility, particularly the echoes caused by the first 2x upsampling for 44.1 Redbook data (less so for higher resolution files).
CONCLUSION
Those who are trying MQA and various filters with typical Delta Sigma DAC’s are using band aids. A growing number of critical listeners have discovered that ladder R2R sounds better than typical DS DACs or, alternatively, that high precision conversion to DSD256 on a computer fed to a true one-bit discrete Delta Sigma converter (no chip) sounds equally great too.
Basically any conversion that eliminates oversampling/upsampling done on a chip is going to have less digital glare and better soundstage because of this absolutely fundamental design flaw in ALL Delta Sigma DAC chips.
83 responses Add your response
@dsnyder0cnn I believe the ESS ES9028PRO chip includes linear-phase filters (LPFs) as well. While LPFs can introduce pre-ringing artifacts, they maintain phase accuracy across all frequencies. In contrast, minimum-phase filters (MPFs) minimize or eliminate pre-ringing by allowing phase shifts, which can affect phase accuracy. This presents a trade-off between temporal precision and phase linearity. Without empirical data, it’s challenging to determine which effect—pre-ringing or phase distortion—has a more significant impact on perceived sound quality within the limits of human hearing. Have you experimented with LPFs in your listening tests? |
I completely concur with your experience... I used an old NOS dac which is better than many costlier dac with new technology... I tried and failed to upgrade my dac in the price category and the new product was completely artificial and bad sound...Even couple with a good power supply... As you said it well :
|
The OP raises a lot of detail that may indeed be relevant. But as is the case so often in audiophilia, the issue is not whether an effect merely exists, but whether it’s significant enough to be audible. IOW, an argument like this is credible only if it establishes quantitative, not merely qualitative, support. I realize that that’s not easy in our field, where even testing methodologies are hotly contested. Having said that, I do offer a counterexample that may add another wrinkle to the issue. I realize that I’m merely making an inference. But it is an inference that is worth considering and is apparently supported by empirical data generated by some of the world’s most respected DAC designers. T+A, one of Europe's ’s most highly regarded manufacturers of high-end DACs (well-known on the continent and now starting to establish a rep in the US) sells sophisticated $5-9000 dual-path DACs that process PCM with Delta-Sigma circuitry, but run DSD through an independent bespoke R2R DAC. The only reason I can think of for this complex design is that T+A engineers confirmed that, within the context of their design methodologies, each topology produced better results with one type of content. I doubt that this solution was adopted for reasons of cost or complexity. That suggests that the DS v. R2R controversy can’t be resolved conclusively by focusing on one, or even just a few, isolated factors. That is, "R2R is better across-the-board than DS in a certain price range." is too reductive, too conclusory, too Wikipedia/ChatGPT, to make me run out and replace my T+A (which, FWIW, is the best four-figure DAC I’ve heard. Check out Stereophile’s R 2500 R review, in which Tom Fine compares a stripped-down embedded version against his $20K reference DAC.) Nonetheless, the OP does raise interesting issues. Interesting enough, in fact, to make me want to go out and further research the topic. Thanks for starting this thread. |
It's important to note that the concept of "natural" sound is subjective and can vary based on individual preferences and experiences. Some listeners may equate naturalness with warmth and smoothness, while others may associate it with accuracy and detail. My preference for a sound that closely mirrors live performances, capturing the true character of instruments like the violin, is a valid and common perspective among audiophiles. |
+1 @curthuff T+A highly qualified seasoned engineers creating solid dependable high value components - consistently performs above their price points |
| Post removed |
| Post removed |
| Post removed |
Although Parks-McClellan-type FIR filters are common in delta-sigma DACs, they are not universally required, and many high-end or professional models intentionally avoid them. Also, several non-R2R DACs avoid using Parks-McClellan FIR filters by employing alternative filtering strategies that prioritize time-domain performance and minimize pre-echo. Chord Electronics uses custom FPGA-based designs in models like the Hugo 2 and DAVE, implementing long-tap minimum-phase WTA filters that do not rely on standard FIR algorithms such as Parks-McClellan. RME’s ADI-2 DAC FS, built on a delta-sigma architecture, offers user-selectable filter modes including minimum-phase and NOS-like options, which bypass symmetrical linear-phase FIR filters. Schiit’s Multibit DACs, like the Bifrost 2/64 and Yggdrasil, employ a proprietary DSP approach known as the "MegaComboBurrito" filter, specifically designed to avoid the pre-ringing associated with linear-phase FIR filters. PS Audio’s DirectStream DAC uses an FPGA-based delta-sigma architecture with fully custom up-sampling and filtering, steering clear of the Parks-McClellan method. Similarly, the Benchmark DAC3, based on the ESS Sabre chip, which eliminates symmetrical pre-ringing behavior. Not entirely sure at this moment how the mentioned units' relate to your $3000 price limit. |
Although Parks-McClellan-type FIR filters are common in delta-sigma DACs, they are not universally required, and many high-end or professional models intentionally avoid them. Also, several non-R2R DACs avoid using Parks-McClellan FIR filters by employing alternative filtering strategies that prioritize time-domain performance and minimize pre-echo. Chord Electronics uses custom FPGA-based designs in models like the Hugo 2 and DAVE, implementing long-tap minimum-phase WTA filters that do not rely on standard FIR algorithms such as Parks-McClellan. RME’s ADI-2 DAC FS, built on a delta-sigma architecture, offers user-selectable filter modes including minimum-phase and NOS-like options, which bypass symmetrical linear-phase FIR filters. Schiit’s Multibit DACs, like the Bifrost 2/64 and Yggdrasil, employ a proprietary DSP approach known as the "MegaComboBurrito" filter, specifically designed to avoid the pre-ringing associated with linear-phase FIR filters. PS Audio’s DirectStream DAC uses an FPGA-based delta-sigma architecture with fully custom up-sampling and filtering, steering clear of the Parks-McClellan method. Similarly, the Benchmark DAC3, based on the ESS Sabre chip, which eliminates symmetrical pre-ringing behavior. Not entirely sure at this moment how the mentioned units' relate to your $3000 price limit. |
Although Parks-McClellan-type FIR filters are common in delta-sigma DACs, they are not universally required, and many high-end or professional models intentionally avoid them. Also, several non-R2R DACs avoid using Parks-McClellan FIR filters by employing alternative filtering strategies that prioritize time-domain performance and minimize pre-echo. Chord Electronics uses custom FPGA-based designs in models like the Hugo 2 and DAVE, implementing long-tap minimum-phase WTA filters that do not rely on standard FIR algorithms such as Parks-McClellan. RME’s ADI-2 DAC FS, built on a delta-sigma architecture, offers user-selectable filter modes including minimum-phase and NOS-like options, which bypass symmetrical linear-phase FIR filters. Schiit’s Multibit DACs, like the Bifrost 2/64 and Yggdrasil, employ a proprietary DSP approach known as the "MegaComboBurrito" filter, specifically designed to avoid the pre-ringing associated with linear-phase FIR filters. PS Audio’s DirectStream DAC uses an FPGA-based delta-sigma architecture with fully custom up-sampling and filtering, steering clear of the Parks-McClellan method. Similarly, the Benchmark DAC3, based on the ESS Sabre chip, which eliminates symmetrical pre-ringing behavior. Not entirely sure at this moment how the mentioned units’ relate to your $3000 price limit. |
Unfortunately, the ESS chips all have pass-band equiripple just like other DS upsampling chips. This results in echoes (exact copy of entire audio at lower level just like a reflection) Ringing is not the same as echo. Ringing is a Gibbs phenomenon and mathematically it occurs at the corner frequency of the filter - unless you can hear 21KHz (typical corner) then pre or post ringing will be inaudible anyway. In truth it should not be on any digital audio mastered file/CD because those frequencies above 20 KHz should have been filtered out prior to A to D. The problem with any filter other than a sharp linear phase at 21KHz is that 1) It changes the phase of high frequencies compared to low frequencies which changes the timbre. 2) any smooth filter with slow roll off can be leaky and frequencies above nyquist can get through, resulting in intermodulation distortion. This is just from a technical perspective. Of course what sounds better to the listener trumps everything else.
|
T+A D200 DAC is incredible. I was wildly lucky to get one. It does have separate PCM (using a Burr Brown chip) and a discrete DS DAC that accepts DSD up to 1024. The DSD side of the DAC runs without upsampling and is the best aspect of this DAC, although PCM sounds pretty good. Another, similarly priced DAC is the Holo May KTE - this one supports DSD and NOS via an R2R DAC. It is equally highly regarded as the T+A D200. Both overcome the upsampling limitations of the short tap filters of a typical chip-based DS DAC. As mentioned, elsewhere in this thread, not all filters in DAC chips are optimized by Parks-Mcllelan - yet this approach yields the best specs - so it’s been almost a standard approach for years. However even those filters without this design will suffer from equiripple. Even “smooth” filters have equiripple. Only NOS R2R or 1 million+ taps upsampling (like Chord Dave) or super high precision conversion of PCM to high rate DSD on a computer (also using high number of taps) can sufficiently reduce pass-band equiripple and the echoes it generates. Only R2R can do so and not introduce latency.
|
I’ll have to check out that Holo DAC. Sounds interesting. Any technical papers you’re aware of that detail what’s under the hood? One thing that continues to amaze me is the lack of T+A awareness among street-level audiophiles. See, e.g., this thread itself. One reason might be T+A’s low profile at American shows. I was speaking to a colleague who visited two AXPONA booths that featured T+A gear last month, and he reported that neither setup produced SQ anywhere near what I hear at home from my T+A integrated, which is the sole source driving a pair of Harbeths through mid-fi (like $1K) cables. |
Holo May KTE measures superbly https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements But then again, almost every DAC does because every designer or chip designer uses Precision Analyzers. Listeners seem to rate this DAC very highly and there seems to be a growing community of R2R proponents since 10 years or more. The sad truth is that Precision Analyzers rely on frequency analysis and large windows of analysis to achieve their precision. And a pre or post-echo will not be detected at all because it’s just time domain distortion - it’s the exact audio signal at much lower level delayed or earlier than the main signal. (A true echo and a completely different animal from pre- or post-ringing at the Gibbs single frequency tone) It is NOT so much the shape of the upsampling filter (smooth etc) that affects what we hear but the equiripple added to pass-band. Anyone who thinks a very slight roll off at 15-18 KHz on a smooth filter is going to change much is mistaken. It doesn’t. What changes is the equiripple which is well audible as a pre or post echo that our hearing detects as fatiguing digital glare and makes stereoscopic interpretation (imaging/soundstage) more laborious and tiring (it’s why we tire of digital more quickly than analog) |
Then there is the issue that many say R2R DACs in the $2K to $3K range are soft sounding and don't have punchy bass, even if they have great natural tone and a nice body to that tone and a deeper soundstage. That often seems to be the main qualifier: That R2R DACs give a slightly wider and deeper soundstage - if it is in the recording. But others would ask, are they "creating" that or presenting it as an artificial "artifact", and it wasn't really what the recording engineer heard in his headphones? Choose your poison at whatever level of price you are willing to pay. Or go back to analog vinyl and be happy. All engineering design considerations (and the company bean counters) color the sound of all digital devices one way or another. DACs at nearly the same price points often do NOT sound the same. But I think we can all agree that today, that most all do sound much better than even 10 years ago, so we have that going for us. |
After my last adventure trusting reviewers who has no idea about a "natural organic timbre sound" who sell dac which are artificial sounding, i will stuck to my SPS dac (i succeeded to repair)... By the way all these people selling dac not one said that a dac must be well grounded to sound the best, not one...
By the way i had seen grounding box sold for many thousand bucks... I created mine for peanuts and total success... They sell product and do not inform most of the times....
|
I started out as a test engineer in the late 1970s and remember using, I think it was called, an Audio Precision System One, the standard high-end signal analyzer at the time, and probably the predecessor of the type of device you’re referring to as Precision Analyzer products. I remember that even back then, those analyzers could operate in both the frequency and time domain. I even remember using our System One to perform Fourier transforms. So I’m not sure about the "sad truth" you cite re:Precision Analyzer’s current product line operating in only the fr domain. I could certainly make the kind of measurements you describe with a pro signal analyzer 50 years ago. In fact, I half-remember someone like Atkinson routinely publishing such measurements years ago in the slick audiophile press (although I may be misremembering after all this time). To be clear: I’m not trying to start a debate. Just hoping to learn something about the current state of an art (of great interest to me) that I haven’t followed since retiring. Hey, thanks for the Holo/Stereophile link. I sorta recall reading that piece years ago, but will take another look, given the new context of this thread. Not sure how much of the articles conclusions still hold up half a decade later, but I'll check it out.
@mapman Don’t hate me, but if you’re looking to get defiitive buying recommendations from an authoritative source, IMHO, Audiogon is not the place. The value of this thread, e.g., is its discussion of under-the-hood D/A tech, which as I’m sure you know, doesn’t necessarily tell you anything definitive about SQ. The idea, I think, is that, if this thread increases your understanding of a technical issue, you’ll be able to ask more educated questions when you go shopping. IMHO, that’s a heckuva lot more valuable than plowing through dozens of "I really like this DAC!" messages from strangers who use a product with systems, rooms, cables, and power that may be nothing like yours. Over the years, I’ve found that the real value of fora like Audiogon -- other than providing a platform for masturbatory proclamations of conclusory opinions -- is more along the lines of "Teach a man to fish...". My 2c. Jeez, wouldn’t "MPoCO" be a great name for a band? |
All I am saying is that most analyzers tend to look at frequency response and do a lot of averaging to get precision. There are some capabilities to plot time domain signals like square wave or impulse response but largely the focus is towards applying known signals at input and measuring output and comparing output to input. An echo is not going to show up as distortion in any measurements - after all it is exactly like the input arriving early or delayed at much lower signal level. We don’t even have guidelines for phase distortion except that less than 2 msec is important and smooth gradual changes in phase are preferred - though obviously phase has to affect timbre no matter how small and phase distortion is added by guitarists to get bigger sound (it makes locating the sound that much harder) |
I suggest to look into Audalytic (Gustard) AH90. There is a firmware update to allow it to run DSD Direct. Anecdotal reports are that there are excellent results obtainable if you feed this DSD256 (needs high precision upsampling of source files on a PC with HQPlayer or similar and then feed that data via USB to this DAC). |
@shadorne thank you for this info! I had the Gustard R26 for a while and it was a nice sounding unit so I am interested in other technologies from the same group. I don't currently have a way to run DSD that I know of - I prefer streaming from Tidal than downloading giant files but I am not opposed to running DSD for my favorite stuff. How does this AH90 do with PCM signals? I honestly never tried the T+A with DSD files, just streaming Tidal in max quality and was absolutely blown away by albums like Bill Evans live recordings and anything with ambience was just so palpable. What are your thoughts on running standard files/streaming on this DAC vs the T+A? |
I stream Tidal as well as 6 Terabytes of personal CD collection using Roon. Roon can convert to DSD64 I highly recommend investigating on the fly upsampling to DSD on a PC and feeding that to a good DAC with true one bit conversion (avoid chip-based DACs unless the upsampling processing path can be avoided - for example all ESS have their characteristic glare due to forced internal upsampling on their chips) - the results can be mind boggling. |
While I am interested in the debate about different technologies, and why various approaches are superior or measure better, in the end, I make choices based on what sounds good to me, and that almost never correlates with measurements. Among the very best sounding DACs, to me, are Audio Note DACs that measure spectacularly poorly. Yet, they sound relaxed and natural and deliver a “saturated” sound, not sound that seems stripped of harmonics. Am I liking resonance/ringing? Perhaps, but I don’t care. What I own is not a tube DAC, but it is one whose sound I like. It is the DAC built into the Naim ND555 server. It delivers a rich sound for solid state. |
The human hearing cannot be understood by modeling it with a set of Fourier maps and calling it job done... It is way more complex... This is why the measures set of a dac is not telling all the tale and why you are right...
|
@acman3 thanks for your nice comment agreed it is a special Dac. And I agree with larryi that sound quality over everything. And chips aside both analog output stages and power supply matter just as much if not more than whether a Dac is DS or R2R. |
Absolutely agree about human hearing complexity. I am utterly convinced we hear sound as a complex interpreted amalgam of the different responses at both our ears analyzed over a short period of time. This is called the Haas effect - that the Haas effect exists is absolute proof we do NOT hear the transient instantaneous signal - we hear a processed result of some kind of analysis of approximately 40 milliseconds of sound - almost an eternity when you really think about it relative to a “transient”. I believe the differences we hear in modulators and filters for digital upsampling are almost entirely due to the equiripple in the pass band that creates echos in the time domain. Engineers believe that a tiny ripple at 70 db below the noise floor isn’t audible but they are forgetting that this ripple is across the entire frequency domain and leads to a highly correlated pre and post echo not random noise. And our ears brain are super sophisticated at locating sound origin - ESPECIALLY if the echo is occurring across the entire frequency domain (which is the case with equiripple from digital processing) - the entire original source sound is echoed exactly at a low level and we hear it - it directly affects timbre and soundstage.
|
The Haas precedence effect is one of the guiding principle in applied acoustics at all scale for Hall acoustics and as in our smaller room...We must takes it into account to adress the reflections points roles... I cannot add anything to your post with which i concur...
It is proven that human audition beat the Fourier uncertainty principle limit or the Gabor limit because Human hearing extract information working in his own non linear time domain... Then you are right... Timbre is way more than just a "color" or a taste but a mirror of the way ears/brain create music and extract objective information from the vibrating sound source. https://phys.org/news/2013-02-human-fourier-uncertainty-principle.html I also confirm that your idea about dac are mine too even if i had less experience with different dac design than you...
|