Any upsampler or upsampling process that matches native DSD quality?

[joelha]

I realize the answer I'm likely to get is a resounding "No", but I thought I'd ask anyway.

 

Whether it's a hardware upsampler (DCS, etc.), a player with real-time upsampler (JRMC, HQPlayer, etc.), or whether it's a process for pre-upsampling files (Korg, etc.), I'm wondering whether anyone has found a way to take redbook files or even higher resolution files and get close to the sound of native DSD.

 

Joel

[MetalNuts]

I can set my Esoteric D-07 DAC to upsample the source to 2x, 4x or DSD. The different upsampling setting of course will sound differently but I have not compared it to any native DSD (provided my equipment remains constant to run the test) by audition.

[barrows]

Joel: I think your question needs some qualifiers:

 

1. What do you mean by native DSD quality? Are you assuming that DSD is superior in accuracy to PCM formats? If so, I must point out that this is just a subjective belief, and that there are quite a few folks with lots of experience, who feel that PCM at and above 24/96 is more accurate than DSD. DSD has a lot of high frequency noise, and it can even be speculated that this noise creates a euphonic coloration which may actually be the sonic signature of DSD.

 

2. Or, are you asking if native redbook quality level recordings can be up/oversampled to equal the apparent resolution of DSD (which is equal to about 20 bits and 88.2 kHz PCM in most parameters)?

 

If you want to achieve the apparent sonic coloration of DSD, then DACs which convert PCM-DSD for conversion should meet your needs. The ones I know of for sure which actually do this, and then do the D/A conversion at a single bit are the Meitner and EMM labs DACs. Everything else I am aware of converts DSD using a multibit D/A, (dCS, and Playback Designs included).

 

Another way to achieve the sonic coloration of DSD would be to convert PCM files to DSD via a good SRC (Saracon, etc), and then play them back over DoP through a true DSD DAC like the Meitner, EMM Labs, or exD/Sonore.

[joelha]

Hi Barrows,

 

Thanks for your reply.

 

While I didn't intend for this thread to be controversial (but then in our hobby, usb cables can be controversial), my experience of DSD is that it is far superior to 24/192. That's why I'm interested in the question of upsampling to DSD quality.

 

And while I know that redbook (and higher resolution files) can be upsampled to DSD and even DSD X2, upsampling to DSD and matching the quality of a non-upsampled DSD file are not necessarily the same thing.

 

That's why I'm asking for specific preferred methods, assuming any exist, to match DSD quality.

 

Joel

[dallasjustice]

my experience of DSD is that it is far superior to 24/192.

You have found this to be the case with which DAC? My experience is that all DACs have their sweet spot and it's impossible to make absolute judgments about which format is better simply by listening to single DAC.

[joelha]

I'm using the Mytek DAC for DSD playback and my judgement was anything but absolute. As you quoted me in your message, I had said "in my experience".

 

Joel

[Miska]

If you want to achieve the apparent sonic coloration of DSD, then DACs which convert PCM-DSD for conversion should meet your needs. The ones I know of for sure which actually do this, and then do the D/A conversion at a single bit are the Meitner and EMM labs DACs. Everything else I am aware of converts DSD using a multibit D/A, (dCS, and Playback Designs included).

 

What apparent sonic coloration? Any facts or figures about this coloration? Multibit delta-sigma, like in ESS Sabre at modulation depths around 90% produce equally high amount of noise.

 

What I've been looking at 5 MHz band spectrum output from various multi-bit delta-sigma converters, the noise output is not much different from noise output of DSD128...

 

I would say that the SDM -> PCM -> SDM chain used in modern PCM production adds much more apparent sonic colorations due to ringing and such that doesn't exist with native DSD at all...

[Miska]

I'm wondering whether anyone has found a way to take redbook files or even higher resolution files and get close to the sound of native DSD.

 

Of course you can try it out.

 

But I've found that upsampling DXD files (for example from 2L) to DSD becomes quite close to native DSD.

 

Native DSD just lacks the sonic colorations of modern PCM production chain with SDM -> PCM -> SDM conversions. With DXD the amount of this conversion coloration is at such low level that it doesn't bother me anymore, of course depending on the particular algorithms still.

[barrows]

Well, while there are those, such as Miska, who believe that DSD is more accurate than PCM, this is only an opinion. Others, such as highly respected recording engineer Keith Johnson are on the opposite side of the fence. My point, is that the assumption inherent in this question is not an established "fact" in any way.

Personally, I do not "know' yet whether I prefer DSD to PCM, as I have not done enough fair comparisons to have a legitimate opinion. Both can certainly sound excellent, and it is also important to consider that a lot (but certainly not all) all DSD has gone through a PCM step during mixing, so it often still will include the PCM "colorations" referred to be Miska.

Interestingly, I do find a fairly consistent sound with DSD, a certain slight softness and sweetness to the high frequencies-which seems to always be present in any playback file which has had a DSD encoding step along the way. I find this sound to be quite pleasant, but it does make me wonder if it is truly accurate...

[Miska]

Well, while there are those, such as Miska, who believe that DSD is more accurate than PCM, this is only an opinion.

 

At least ESS, Burr-Brown, Cirrus Logic, Wolfson and AKM seem to also believe, given that they all design delta-sigma converters these days. And those exceed performance of the multi-bit converters of past with fraction of the cost.

 

Others, such as highly respected recording engineer Keith Johnson are on the opposite side of the fence.

 

Just that you won't find any (?) new PCM audio ADCs anymore if you are looking for any new equipment. All the new "PCM" ones I know are delta-sigma, and thus practically equivalent of DSD ADC with internal conversion to PCM.

 

And most PCM DACs on the market these days essentially convert to equivalent of DSD for their actual conversion stage.

 

Both can certainly sound excellent, and it is also important to consider that a lot (but certainly not all) all DSD has gone through a PCM step during mixing, so it often still will include the PCM "colorations" referred to be Miska.

 

Why would it go through PCM step during mixing? I don't have trouble mixing DSD without resorting to PCM. Sonoma should be able to do it too. And for example in case of Pyramix the intermediate step is at DXD rate which, as I said earlier, becomes mostly transparent in terms of these PCM colorations.

 

Most of my DSD material has minimal amount of mixing and that has been in most cases done in analog domain.

 

Interestingly, I do find a fairly consistent sound with DSD, a certain slight softness and sweetness to the high frequencies-which seems to always be present in any playback file which has had a DSD encoding step along the way. I find this sound to be quite pleasant, but it does make me wonder if it is truly accurate...

 

That is most likely due to noise shaping it has got, especially if the conversion is to 24-bit. You can get similar improvement to PCM by using suitable noise-shaping algorithm as part of upsampling. I'm doing it always with HQPlayer when I perform PCM-to-PCM upsampling. It improves DAC linearity by removing some of the low-level non-linearities of DAC-chip output when the signal enters analog filtering stages.

 

If I convert native DSD to RedBook PCM, it sounds like RedBook PCM after the conversion. If I convert to 176.4 kHz PCM it sounds like noise-shaped 176.4 kHz PCM. When converted to DXD it is staring to come close to the original and doesn't sound like PCM so much anymore.

[barrows]

well, hmm. First of all, the idea that there are not any true PCM DACs available is certainly not true, off the top of my head I can think of these: Wadia, Naim, Resolution Audio, Total DAC, MSB, and Phasure NOS. As far as ADCs go, that may be a little trickier, but if a serious recording engineer really felt the need for straight R2R style ADC, one could be commissioned.

 

As to "why" a PCM step might be included in the recording process, I will let the recording engineers speak for themselves on that one, but I will add, if one spends any time at the pro audio forums, one will find many professional recording engineers who believe that PCM even as low as 24/96 is "transparent", and if they believe that, it is pretty easy to understand why they might choose to mix in pro tools (PCM). Not that I would agree with them.

 

Here is a quote from issue 63 of Hi Fi + regarding the production of the Genesis catalog on SACD:

 

"How were the Genesis mixes done?

 

The multi-tracks were transferred at 192 to Pro Tools and those files were mixed to the Sonoma system in DSD..."

 

Now, there are plenty of "audiophile" recordings around which are straight DSD, but much of the SACD content out there is not.

 

I think everyone here makes a good point that which format (given native recordings in each) one may prefer will be highly dependent ont he converter itself. I have yet to hear of anyone doing really comprehensive listening tests which might "prove" the superiority of PCM or DSD, and I find measurements at these levels of "differences" to be entirely unconvincing.

I would like to participate in the following style of listening test:

 

1. Get really good DACs for PCM (say an MSB or Total DAC) and a really good, true single bit, DAC for DSD (say EMM Labs).

 

2. Get some recordings which were captured directly both in DSD and 24/192 pcm, using native ADCs (say EMM and PM Model 2)

 

3. Audition the results through the same system, only chaning the relevant DAC, being sure to use amps and speakers capable of not adding too much, say an Ayre MXR and a speaker with no crazy phase issues, like a Vandersteen Model 7.

 

A test such as this would likely point out what the real differences between DSD and PCM are.

 

In the mean time, I would suggest the OP look into trying some minimum phase slow roll off filters for PCM, maybe Miska could suggest an approach in HQPlayer which might help him to get closer to the sound he is looking for from his PCM files?

[Julf]

As to "why" a PCM step might be included in the recording process, I will let the recording engineers speak for themselves on that one, but I will add, if one spends any time at the pro audio forums, one will find many professional recording engineers who believe that PCM even as low as 24/96 is "transparent", and if they believe that, it is pretty easy to understand why they might choose to mix in pro tools (PCM).

 

From a digital signal processing technology point of view, PCM is much easier to handle than DSD, especially for more complex operations (most "pure DSD" editing software is limited to simple cut/paste/fade/crossfade mixing functionality). That is why the recording industry is moving to DXD (digital eXtreme Definition). Here is the description on DXD from Wikipedia:

"Digital eXtreme Definition (DXD) is an alternative audio encoding scheme for professional use that was developed for editing high-resolution recordings recorded in DSD, the audio standard used on Super Audio CD. Contrasted with DSD-Wide or DSD Pure which offers level, EQ, and crossfade edits at the DSD sample rate (64fs, 2.822 MHz)^[1][2], DXD is a PCM signal with 24-bit resolution (8 bits more than the 16 bits used for Red Book CD) sampled at 352.8 kHz – eight times 44.1 kHz, the sampling frequency of Red Book CD. The data rate is 8.4672 Mbit/s per channel – three times that of DSD64. DXD utilizes the vast array of plugins also available to PCM based digital audio workstations, such as Cubase, Logic, Digital Performer, etc."

 

It also contains a link to a pretty good white paper, "The Advantages of DXD for SACD".

 

Looks like most editing of DSD will continue to happen in PCM.

 

[Miska]

From a digital signal processing technology point of view, PCM is much easier to handle than DSD, especially for more complex operations (most "pure DSD" editing software is limited to simple cut/paste/fade/crossfade mixing functionality).

 

I've never wanted to do easy things...

Easy things are just boring.

 

And those editing functions suffice for classical music, which is currently majority of DSD content anyway.

 

It also contains a link to a pretty good white paper

 

That's just a sales pitch in itself. You can also take a look at alternative view.

 

Looks like most editing of DSD will continue to happen in PCM.

 

But at minimum at 352.8 kHz rates. No point at all going any more downwards from there. My point was that if you have to do PCM, don't go below that point. Ever, if you can choose. With more CPU power, I could bet that rates will just go up from there, not down. In the DXD recordings made with DXD ADCs you can also see the familiar ultrasonic noise floor lift, coming from the delta-sigma ADCs.

 

When you convert to DSD128 you save some bandwidth while avoiding less good delta-sigma modulators built into DAC chips. DXD consumes 2.0 megabytes per second while DSD128 consumes just 1.3 megabytes per second for stereo. Good DSD DAC can be significantly cheaper than good PCM DAC, much simpler.

[Miska]

First of all, the idea that there are not any true PCM DACs available is certainly not true, off the top of my head I can think of these: Wadia, Naim, Resolution Audio, Total DAC, MSB, and Phasure NOS.

 

Now you are listing just end-products, not converter chips. There are just few old multi-bit chip designs being manufactured anymore. No new ones being designed. Naim and Phasure use old PCM1704, MSB is selling discrete OEM modules.

 

Number of newer delta-sigma designs, even from Burr-Brown has long exceeded the performance of PCM1704 (it is still very good multi-bit ladder design though).

 

1. Get really good DACs for PCM (say an MSB or Total DAC) and a really good, true single bit, DAC for DSD (say EMM Labs).

 

2. Get some recordings which were captured directly both in DSD and 24/192 pcm, using native ADCs (say EMM and PM Model 2)

 

I'm doing it in more realistic way. My own A/D/A design has two parallel ADCs, one 192/24 PCM and another one is DSD128. Both modern design, sharing the same analog front-end.

 

DAC is a single chip that can do 192/24 PCM with multi-bit delta-sigma and true native DSD128.

 

I can switch the data stream from either ADC to the DAC on the fly. This way, the only difference is in actual converters, not in analog stages or such.

 

3. Audition the results through the same system, only chaning the relevant DAC, being sure to use amps and speakers capable of not adding too much, say an Ayre MXR and a speaker with no crazy phase issues, like a Vandersteen Model 7.

 

Your way, you would be listening partially to different analog stage designs too, not just converters.

 

In the mean time, I would suggest the OP look into trying some minimum phase slow roll off filters for PCM, maybe Miska could suggest an approach in HQPlayer which might help him to get closer to the sound he is looking for from his PCM files?

 

Well, it is possible to try out both minimum phase upsampling to both PCM and DSD128, using Mytek DAC for example.

 

1) Select PCM output and "poly-sinc-short-mp" and "NS9" at 192 kHz

2) Select SDM/DSD output and "poly-sinc-short-mp" and "DSD5" at 5.6 MHz

 

PCM input can be anything from RedBook to DXD... Then just compare...

Result is still DAC specific, not universally applicable.

 

(these days, I use only DSD128 on Mytek, nothing else)

[Julf]

I've never wanted to do easy things...

Easy things are just boring.

 

Good point!

 

But at minimum at 352.8 kHz rates. No point at all going any more downwards from there. My point was that if you have to do PCM, don't go below that point. Ever, if you can choose. With more CPU power, I could bet that rates will just go up from there, not down. In the DXD recordings made with DXD ADCs you can also see the familiar ultrasonic noise floor lift, coming from the delta-sigma ADCs.

 

When you convert to DSD128 you save some bandwidth while avoiding less good delta-sigma modulators built into DAC chips. DXD consumes 2.0 megabytes per second while DSD128 consumes just 1.3 megabytes per second for stereo. Good DSD DAC can be significantly cheaper than good PCM DAC, much simpler.

 

Sure, and I agree that can be an argument for DSD128 as a distribution and storage format. But intermediate processing will probably mostly happen in (high-resolution) PCM.

[barrows]

Yes, but how many recording engineers are currently using DAWs which support 384/352.8 rates?

My point is that in the real world we are not often going to get recordings which are edited at those rates.

If one is only concerned with reproducing recordings which they make themselves, and, perhaps, recordings from classical specialists and audiophile labels that is one thing... But many on these forums appear to be listening to SACD rips of pop/rock/jazz titles from the major labels-I suspect that a great many of these SACD rips have been processesd as PCM somewhere along the path.

 

Miska: Once again you are referring to a very specific set of circumstances which exist only for yourself. For instance, I gave examples of true commercial PCM DAC designs because those are available for purchase to readers of this site, with the expectation that few here are going to be purchasing a DAC chip (1704, PCM63, etc) and building their own DAC.

I agree with you that the comparison approach I suggested will be subject to the different output stages in the two DACs, but this cannot be avoided by those who only have access to available products: of course having two converters in one chassis, able to take advantage of the same output stage would be even better.

I am not so sure that that TI, AKM, AD, etc. has shifted all focus to SDM converter chips strictly because of the higher performance they are able to achieve with them-I suspect the cost of manufacture is likely a much bigger reason for the shift to SDM. Granted that current SDM chips offer better numbers than that available from (old) R2R designs, but who knows what might be available now in R2R chips if development on them had continued? I am not sure I believe MSB's claims about their discrete converters, but if one does, apparently there is plenty of perfromance there. In any case, DNR beyond 120 dB is a moot point as far as I am concerned, as even the very best systems in a real room will never reach this level of performance.

[Miska]

Yes, but how many recording engineers are currently using DAWs which support 384/352.8 rates?

 

Almost any new DAW software version supports it, of what I've seen.

 

Miska: Once again you are referring to a very specific set of circumstances which exist only for yourself. For instance, I gave examples of true commercial PCM DAC designs because those are available for purchase to readers of this site, with the expectation that few here are going to be purchasing a DAC chip (1704, PCM63, etc) and building their own DAC.

 

I'm talking about percentage of multi-bit ladder DAC chip models of all available audio DAC chip models. There are several hundred different audio DAC chip models on the market, while less than five of those are multi-bit ladders.

 

We could also talk about percentage of market share between multi-bit ladder DACs and delta-sigma DACs, also counting in all the AVRs and other audio equipment like CD/DVD/BD players. Also there, the amount of market share of ladder DACs is really tiny.

 

This is what I'm interested in.

 

I am not so sure that that TI, AKM, AD, etc. has shifted all focus to SDM converter chips strictly because of the higher performance they are able to achieve with them-I suspect the cost of manufacture is likely a much bigger reason for the shift to SDM.

 

I'm not sure how much ESS and others are counting manufacturing costs for their top of the line DAC chips. But of course it matters too. It doesn't make much sense for IC manufacturer to design or manufacture any chips that sell less than something like 20-100k pieces per year.

 

If you can achieve better performance with delta-sigma design for $10 per piece than multibit-ladder can do for $50 per piece, why would you make those multibit-ladders? Competition would just beat you. These things are generally considered in very objective way.

[Jud]

Others, such as highly respected recording engineer Keith Johnson are on the opposite side of the fence.

 

No doubt I'd be able to find it myself after some looking, but happen to have a link for a lazy man? ;-)

 

I'll mention that the DAC designer Mike Moffat is apparently of the same opinion, but note that he also doesn't think much of USB input for DACs.

 

Interestingly, I do find a fairly consistent sound with DSD, a certain slight softness and sweetness to the high frequencies-which seems to always be present in any playback file which has had a DSD encoding step along the way. I find this sound to be quite pleasant, but it does make me wonder if it is truly accurate...

 

What I'm about to say is likely indicative of nothing, because I've never yet heard native DSD. Best I can do at the moment is listen to DSD decimated to 24/88.2 PCM. Having compared that to Redbook of the identical performance (upsampled to 24/192) of various classical pieces conducted by Jordi Savall, in particular the Sixth Brandenburg Concerto, I found that yes, violins did sound "sweeter" in the (converted) DSD version, but to me the sound was closer to that of real instruments in concert.

[joelha]

Miska,

 

I followed your instructions for DSD128 up sampling using my Mytek DSD DAC and I'm getting a slow stuttering playback of my redbook file.

 

I'm starting HQPlayer with ASIO.

 

What am I doing wrong?

 

Thanks,

 

Joel

[Miska]

I followed your instructions for DSD128 up sampling using my Mytek DSD DAC and I'm getting a slow stuttering playback of my redbook file.

 

You can try setting buffer time in settings to 100 ms. But one possibility is that you are running out of CPU time, what kind of computer do you have? You can try setting output sampling rate to DSD64 (2.8 MHz) and see if you get proper playback that way, it takes half of the CPU time compared to DSD128.

 

As an example, older MacMini with Core2 Duo is able to do DSD128 upsampling on a good day, but not always, it's really on the edge. Latest MacMini with 2.3 GHz Core i5 can do it without any trouble.

 

You can use Windows Task Manager to see if any of the CPU cores is maxing out.

[joelha]

Thanks Miska,

 

Here's what I can tell you.

 

I'm at the office and therefore away from home audio system.

 

However, I can tell you that I am using USBPAL 1.33.10 and a CAPS 2.0 with 4GB of RAM.

 

When you refer to the buffer setting, are you talking about the ASIO buffer size?

 

If so, it seems that I can only go up to 64ms.

 

How do I go up to 100ms?

 

Does the USBPAL sample rate setting affect anything in this regard?

 

I have been able to remote in to my system and can tell you that when playing, I'm using 52% of my CPU resources and 46% of my memory.

 

Thoughts?

 

Joel

[joelha]

Well, it turns out that one of the cores on my CAPS 2.0 server is being maxed out not only when upsampling to DSD128 but even when playing DSD64.

 

I have two CAPS 2.0 servers. If I like using HQPlayer (the upsampling feature in particular), I'd hate to think I've got two servers which can't run the player, at least in their current configuration.

 

Suggestions?

 

Thanks,

 

Joel

[Miska]

I've attached screenshot of resource usage for the RedBook -> DSD128 case on a 2.3 GHz Core i5 MacMini (8 GB of 1333 MHz DDR3 RAM). This should give some kind of ballpark figure of how much CPU power is needed for the case.

 

I've tried to make the algorithms as fast as possible without giving up anything in terms of quality. If there's demand, I can consider adding some lighter weight filters for the SDM upsampling to the next version.

 

[Sam Lord]

Good discussion! I think we should review the DSD and DXD history, though sadly mine is not reliable. But as I recall, there was a great demand by recording and archival engineers for a really good digital medium, and DSD was developed for them. Well prior to the marketing of SACDs, Sony made a demo using IIRC 8 channels of a prototype DSD recording system of a space shuttle launch. Sony's chief classical engineer told me about that demo in astonished tones. Sadly the market didn't embrace the format for many reasons. Over at co-designer Phillips, their engineers (notably then-young Bruno Putzeys) developed DXD mainly as a mastering tool. It was designed to be completely transparent in at least two conversions, i.e. DSD to DXD and back to DSD. This of course was only possible with extremely good circuitry on both ends.

 

A while later DSD128 came out with vastly better noise performance than DSD64, and today there is some preference in the mastering community for pure DSD128 recordings and playback above all else, save perhaps tape. See:

Korg DSD

 

And the best monolithic ADCs today are hybrids, having five- or six-bit modulators running at high speeds. The raw output of these can be noise-shaped to give astounding performance, a bit like the approach of the playback side used in Phasure DACs.

 

But the best ADC performance today might be achieved in DSD64 form by the Grimm ADC1, a design mostly by the aforementioned Bruno Putzeys. He has been in the very center of experiments with with various ADC architectures. About 4 years back he told that the best multibit ADCs are monolithic, but that a continuous-time ADC could be better than those if it were discrete, so that's what he built. Continuous-time circuits change values all the time, not just during clock transitions, and this yields a big noise advantage. Bruno is otherwise pretty agnostic about the best possible delivery formats. For signal processing long words are necessary in any case.

 

I think we should keep in mind the actual circuits involved in these comparisons, since only a handful of designs are competing for the very best sound. There are also the gain-ranging units made by Prism (Dream AD1) and Stagetec, as well as the new breed of gain-ranging digital output level controls.

 

Cheers

[joelha]

Thanks for the information and screenshot, Miska.

 

I guess I'll have to try the upsampling feature of HQPlayer on a more powerful (but not music server optimized) PC just to better understand what the player sounds like.

 

On a different but related note, I did try upsampling a redbook file to DSD128 using Korg's Audiogate. I was surprised that I could not tell the difference between the files on my Mytek DAC.

 

I'm hoping for better results with HQPlayer.

 

Joel