PCM vs DSD?

I was originally going to post this under the thread Delta Sigma: An Inconvenient Truth, but decided to make it its own thread.

Basically, I am wondering if direct DSD into a D/S DAC is capable of competing with R2R DACs.

I agree wholeheartedly with the majority of the community that R2R is a superior sounding technology to D/S when it comes to PCM. I love my R2R DACs: I own the Bifrost 2/64 and Gustard R26, both have that addictive, organic, realistic, and natural sound and are especially awesome in NOS mode with HQPlayer doing oversampling to 176.4/192 kHz.

...but what about PCM vs DSD? Real DSD tracks are few and far between, but HQPlayer and/or Roon unlock the capacity to convert all of your music into DSD on the fly, with vastly superior filters to those found inside D/S DACs, for a fraction of the cost of your high-end DAC (unless you're into a lifetime Roon sub). There's also a free modified version of sox capable of converting offline files to DSD, although the filters and modulators aren't as good as HQPlayer.

The only real D/S DAC that I own is my DAP*, Shanling M8 (2x AKM4499EQ). PCM sounds pretty damn good, but if I use sox to convert PCM files to DSD, the difference is absolutely night and day: DSD is vastly superior to PCM in every way. Is it hazy, thin, or any of the other descriptors that people usually apply to D/S DACs? No, it's quite solid, extremely detailed and dynamic, with focus like getting a new pair of glasses. Maybe it isn't quite as lush as R2R, but it's definitely clear and clean af while lacking any of the typical negative descriptors of D/S sound.

Getting a bit into things I don't quite understand, I do get that DSD has a bit of an edge over PCM being that it is 1-bit and therefore maintains 'perfect linearity' as opposed to multi-bit PCM which... y'know what, I'm not going to pretend I understand this. Let's just say that I've read that DSD should, in theory, be superior to PCM by virtue of being 1-bit (altho we all know that theory only goes so far in this hobby).

So.... maybe there are members out there, who are more experienced in this than I am, that could answer my question: can DSD direct into D/S DACs compete with R2R sound?** Now that I've unlocked HQPlayer's capacity to convert everything to DSD, should I be looking to trade in my Bifrost 2/64 for a D/S DAC that can play DSD direct, or trade my Gustard R26 in for an A26?

--

* My Gustard R26 has a built-in discrete 1-bit DSD DAC, however, I get a hiss when I play DSD so I just ignore the feature.

** I am aware that this is a sweeping generalization, and actual performance is dependent on things like the DAC chip itself (I understand some ESS DACs will even oversample/reprocess DSD), topology of the circuit, and quality of the analog output stage as well. Personal preferences are also going to play a role here.

 

That speaks to the player's own operation more than the format right? If the source was PCM it's not like you're finding better information out of nowhere. Do you think SOX has a better process or maybe you're bypassing an upsampler that isn't ideal?

 

Yes, I think the difference is in the quality of the audio processing. Going to DSD bypasses the internal oversampling filters in the DAC. Digital oversampling filters like those in sox or HQPlayer do a far better job than analog oversampling filters built into DAC chips.

 

For the last three days or so I have been upsampling all PCM to 176.4/192 kHz in Roon ("precise" {which completes the math I think?} linear phase) before sending it to YggyA2. I think I hear a little less over-thickness in bass. Thing is I'm not sure if we know if at this sample rate the Super Combo Burrito (or whatever it's called) filter/upsampling in Schiit dac's really gets out of the way...or perhaps we do?

Do we know this? My understanding is limited, but I though a DSD "filter" is just a low pass filter applied to attenuate the ultrasonics (megahertz range if memory serves) of DSD which is the primary product of DSD - the audio range being an almost a 'byproduct'.

 

Yeah, when using the closed-form filters in HQPlayer, which should be the digital equivalent of Schiit's filter or close to it, the bass is a little too claustrophobic and overly aggressive for me. You need to use NOS mode to fully bypass the oversampling filter.

My experience has been a resounding yes: when properly applied, digital oversampling filters are superior to the analog ones built into DACs. This is true for both PCM and DSD conversion. My understanding is that PCM is converted to DSD inside of a D/S chip before conversion to analog anyway, and the night-and-day difference between DSD and PCM playback suggests that the stock oversampling filter(s) are to blame as this is the only real difference. At a basic level, digital filtering, as processed by a computer, can throw way more processing power at the problem than the analog filter built into most units.

 

I am not sure Schiit DAC's (in general, or any specific model) have a NOS mode as such...Has Jason/mike said this explicitly? The Yggdrasil (and every other Schiit MB DAC I'm aware of) has a 176.4/192 kHz limit on all inputs (even USB) true, but do we know that further up sampling/digital filtering is not occurring? I uderstand that Jason/MM keep the actual design of the "super mega combo burrito filter" a trade secret...I could be wrong of course.

I think there is at least one conflation here - that between "analogue" filtering (always a low pass filter if I am not mistaken) and "digital" up sampling, which involves a mathematical "reconstruction filter" that to us laymen mostly is about "phase" - the relationship in time between all the frequencies in the audio band (e.g. "minimum" -what should be called "out of phase" vs linear phase). Real Engineer's please step in and correct/expand upon this please.

I have heard it asserted that for the modern SD chip, it is not really a limitation of processing power but rather the lazy implementation of (by today's standards) poor upsampling technique (that is nonetheless the "standard" in the industry) by the big consumer dac suppliers (ESS, AKM, CS) that is the limiting factor and could be the reason many believe they hear large differences when the upsample with HQPlayer, etc. However I think I remember reading Jussi (the owner/creator of HQPlayer) say that the use of his product with SD chips is limited (i.e. they have no real NOS modes usually, etc.)....maybe someone can correct/expand on this as well.

 

No No No No No No No No No No No No No No No No No No No No No No No

 

Yes, Schiit multibit DACs have a NOS mode. It's a relatively recent feature, you may need a firmware update, but you basically hold the input button down for a couple of seconds and let go. The light should start 'breathing'. This is NOS mode.

I believe certain AKM chips are able to process DSD without applying any oversampling. What rates it can play natively depends on the chip itself, but there are chips out there that can do this "natively".

Care to elaborate?

 

Thanks, I have not been paying attention Looks like I would need to send it back to Schiit going through this process:

https://www.schiit.com/products/yggdrasil-upgrade

as mine is an older A2 unit. I probably won't end up doing this, though I have ordered Holo's Cyan 2 DAC that should arrive by the end of March and this is a Real-True NOS R2R Dac and I learned from the Spring 3 KTE loaner that I enjoyed upsampling to 176.4/192 kHz more than NOS mode (I just used Roon as I did not take the time to set up HQP). I will give DSD I try as well, though it would have to be something to permanently commit to the computing power (which will spin the fans on my Roon installation)...

 

Topic been beaten to death. Real reason was why DSD was brought out when Sony/Philips promised us Perfect Sound Forever back in the 80s?

 

I will try to answer you question with a couple of anecdotes:

A few of my friends, (who are actually SBAFers, but are semi-retired and don't come on here often these days) recently discovered the goodness of DSD using the HQPlayer. Some of their DACs include the Holo May, T+A DAC200 etc and they have their own favorite filters and modulators. The May owner used to use PCM on his May prior to that, and the T+A DAC owner used to own a Wavedream and had an R26 on loan.

While this was happening, another friend got to hear a modded PS Audio's DirectStream DAC (which is a DSD DAC) with a more recent firmware and he immediately bought one and got it modded and he has finally quit chasing DAC upgrades. He used to own Wavedream, Abbas etc. in the past. Since then, a few of my friends have also gotten on to the modded DirectStream bandwagon. Most of them went from R2R DACs such as Wavedream, Yggdrasil, R26, Abbas etc.

Some things to note: Almost all of them who got converted are speaker users and don't use headphones much these days. We don't know yet if DSD makes much of a difference on headphones. Even on speakers, not everyone felt that the DSD (either in the form of HQP or DirectStream DAC) was an upgrade in all aspects. Some missed the inner detail and texture of their former DACs and some missed the sheer authority and macro swings. As always I guess it is going to come down to your preferences and your chain.

Between the Bf2 and R26, you could retain the one you like the most. Friends report that apart from the choice of the filter and the modulator, the higher sample rate makes a considerable difference as you go higher like 512 and 1024. So if you have a powerful enough PC that can run the long tap filters at DSD1024, you may want to consider the Cyan 2 or even a used Spring 3.

 

Regarding the question of upsampling in general: It depends a lot on the algorithm. Personally, as I never had a DAC that swallowed DSD, I only stayed in the PCM realm, but here’s what I found:

• Upsampling with Roon made the RU6 quite a bit better. On the Meier Daccord FF, Roon upsampling using “precise” minimum phase to my ears was identical to what the DAC did itself, with small differences when switching to linear phase.
• Upsampling with HQPlayer is a completely different matter. It has the simple algorithms that Roon uses, too, but it also has much better ones, and those make an easily audible difference on my Meier DAC. It’s really like carrying a portable speaker from a room with quite some reverb to a room with dry acoustics.
• My guess is that it takes a good DAC, probably especially a good clock for this to make that large a difference, because on the RU6, the gains from HQPlayer vs. Roon upsampling were small.

(Upsampled to 352.8/384 kHz for RU6 and to 192 kHz for the Daccord as it neither supports 176.4 kHz nor anything above 192 kHz.)

 

Right, so no recent first-hand experiences with the implementations being talked about in this thread. It's not just about the original format in the 80s it's about upsampling PCM to DSD and the potential benefits from doing so.

 

Why we are doing this in the first place?

I did it a few times and I realized that converting PCM to DSD was a giant waste of time. All of my crap is in the PCM world, including my DACs. As I gotten older, I mostly given zero fucks about DSD. Personally PCM to DSD didn't do anything for me. The DSD format itself didn't do anything for me.

What does it for me is mastering of an album. Could be PCM or DSD, if the mastering sucks then who gives two shits about PCM and DSD.

 

Because, lately, some people have been hearing benefits from a few different methods of PCM -> DSD upsampling. I think a few people here have a few new counterarguments worth sharing.

Of course mastering is most important, but that's not why this thread was created.

 

There is no benefit.

 

Maybe you don't like the math but these aren't forums about pure math. We're here to talk about listening to things. Some people have been liking what they hear.

 

Yeah, but they have Force Lightning though...