Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[rovo]

Hello all,

The DSD format intrigues me and so I'm playing around with it to obtain a better understanding.

I've been doing some PCM to DSD conversions and I noticed that often, the DSD music files are often given a slightly higher dynamic range value... See screenshot.

Can someone explain to me what'd going on? I have trouble rationalizing it. My ears don't hear the difference btw.



-RoVo

[rovo]

Redbook conversion. Same thing.

[AlanDistro]

Dynamic range is not the best tool, it's just a tool.

The dynamic range reading only cares about peaks and valleys, not what the material actually sounds like.

This is why sometimes when you convert a redbook file to a lossy mp3, the lossy mp3 shows a higher dynamic range, because the lossy codec is creating slight variations in the transients and the dynamic range tool is reading those new lossy transient peaks, even though the mastering is the same.

The same idea is happening here, just at much much higher resolutions. DSD does not work the same way PCM works, so the transients are handled differently and it's tricking the DR meter by 1 dB on some tracks.

It sounds like you understand this already, but you can't make something have more dynamics just by converting it to a different format.

Also, just so you know, you're not gaining anything by converting your PCM to DSD, you're just upscaling what's there. Same as if I took a 800x800px lossy jpg and saved it as a 1600x1600 lossless png. Yes the files size and dimensions are larger, but there's not actually any more detail or information in the larger file, its source is still limited to the information that was in the 800x800, the software just guessed at filling in the extra pixels.

[rovo]

Thank you. From my understanding, the PCM2DSD conversion is not lossless. Noise is handled differently for one. The effect has been described as euphonic by some.

I'm not suggesting the conversion will add bits that weren't there to begin with, which is why I was having trouble interpreting the DR results. Your reply makes sense though.

For completeness sake, I include the image with the column headers (that were missing on the previous images I shared).

[RD James]

DSD was an intermediate format that was technically superior to 16-bit 44kHz PCM at the time, and is inferior to high bit-depth, high sample-rate PCM.
The resurgence is only because there are lies about it being "magical" due to its 1-bit nature and being "the closest thing to analog tape" etc.
Yeah, being difficult to work with and sounding inferior to PCM is what makes it "close to analog tape". Audiophiles sure love to make things expensive and inconvenient, and convince themselves that the inconvenience makes things sound better though.
 
Don't worry, they still have "HD" PCM tracks they can sell you if you don't fall for their lies - despite 16-bit 44.1kHz basically covering the entire range of human hearing, and 24/48 absolutely covering it.
The only time you should be able to tell the difference between the two would be bad mastering/conversion. 24/48 is already overkill.
Playback devices need to be >16-bit if you are going to use digital volume control or processing such as room correction though.
 
You want high bit-depth, high sample-rate PCM for audio recording/mastering, but it's pointless as a distribution format.

[rovo]

I don't think it is quite as clear cut as you seem to believe. One of todays' problems, which makes it increasingly difficult to get to facts on just about any subject is that there are so many people who feel the need to state opinions as facts...

I've read stories from both opponents and proponents of the DSD format. The opponents usually seem quite passionate, but are unfortunately almost always the least informed....though there are exceptions.

As I have said, I am intrigued by the format. I am 51 and my ears won't hear anything above 16KHz... In fact, my ears can't tell the difference between High Res music and CD quality (or even lossy compressed for that matter). This doesn't stop me from enjoying it. It also doesn't cause my inquisitive mind to shut off.

[RD James]

I don't think it is quite as clear cut as you seem to believe. One of todays' problems, which makes it increasingly difficult to get to facts on just about any subject is that there are so many people who feel the need to state opinions as facts...
I've read stories from both opponents and proponents of the DSD format. The opponents usually seem quite passionate, but are unfortunately almost always the least informed....though there are exceptions.
As I have said, I am intrigued by the format. I am 51 and my ears won't hear anything above 16KHz... In fact, my ears can't tell the difference between High Res music and CD quality (or even lossy compressed for that matter). This doesn't stop me from enjoying it. It also doesn't cause my inquisitive mind to shut off.

Here's the noise spectrum of an unfiltered DSD track, converted to 352.8kHz PCM:

 
I've marked -96dB, -144dB, and 25kHz points.
-96dB is the noise floor for undithered 16-bit PCM.
-144dB is the noise floor for undithered 24-bit PCM.
25kHz is the point at which the noise floor rises above -144dB.
 
So if we make a naive comparison, you could say that DSD has lower noise than PCM in the audio band (20-20K), and its frequency response would match approximately 50kHz PCM. (sample rate must be twice your highest frequency to avoid aliasing)
Or if you work in marketing, you would say that DSD's frequency response extends to ~54kHz, since that's approximately where it crosses the -96dB mark, which you could say is approximately 96kHz PCM.
 
But that is compared to undithered PCM.
With flat TPDF dithering (well, a 3dB tilt as is typical) the noise floor for 16-bit PCM drops to about -140dB, and 24-bit PCM drops to about -190dB.
With noise shaped TPDF dither, the noise floor for 24-bit PCM drops to below -240dB.
Higher sample rates allow for lower noise in the audio band, since you can shift the noise into even higher frequencies. Here, I used 192kHz.

 
DSD also has problems with an uneven noise floor. It's not flat like PCM, since there are insufficient bits to fully dither the signal.

 
While it may be around -190dB or so at the lower end of the spectrum, it rises to about -160dB around 7-8kHz, which puts it somewhere in-between 16-bit and 24-bit PCM if you consider the entire audio band. That's why it's compared to approximately 20-bit PCM.
 
But this is all unnecessary anyway, because no audio system in existence has a noise floor that low, and even if such a system were to exist, it would be masked by environmental noise anyway.
Not to mention that if you're listening to music so loud that you can hear even a -96dB noise floor from undithered 16-bit playback, you will be deaf in no time.


 
Using 24-bit in production allows you to record at lower volume levels which means that the audio is far less likely to clip/saturate when things get loud.
For release, the gain will be normalized so that all the audio information is well above -96dB and thus there is no need for 24-bit as a distribution format.
 
High sample rates are used in production because it means that spurious tones will be sampled correctly instead of causing aliasing, which allows them to be filtered out when converted to a 44.1/48kHz format for release.
If you're buying the high-res tracks, all you're getting is additional noise that is beyond the range of human hearing, which has the possibility of causing aliasing in your playback system - and that will be audible.
Some of these tracks just have continuous or repeating tones around 30kHz or so, and the rest of it is just noise. If you filter out anything below 20kHz or so, and then pitch-down the high frequency content so that we can hear it, there's nothing of value there.
 
The reason I have DSD tracks in my library is because they're the only way to get 5.1 mixes for some albums, or because I don't want to buy them again just to get a DVD-A/Blu-ray version of the same thing.

[rovo]

Thank you.

The error you are making is comparing a 'PCM from DSD' signal with a pure PCM signal. This is not an honest comparison as you compare the former, which is constricted to the worst of both worlds with the latter, which has not suffered from any conversion.

Also, you leave out DSD128 etc. which have a dramatic effect on the noise shaping and the filter used. SACD players have always upsampled to DSD128, so the noise floor only becomes an issue when playing DSD64 files on your PC, for which the format was never intended. I think Both formats have their place and can sound terriffic...if played natively. This however, is seldomly the case, as native DSD ADC/DAC’s are rare and native PCM ADC/DAC’s are even rarer, and so comparisons become pointless, especially if the ADC/DAC’s used are “hybrid” types.

Your second mistake is that you seem to think that hearing is a purely mechanical phenomenon of wave propagation. It is much more than that. It is very much also a sensory and perceptual event.

Furthermore, the opponents of DSD (and high res PCM) are in the unenviable position of having to prove a negative. While there are more or less generally accepted limitations to human hearing, we still do not understand the human auditory system sufficiently to replicate it mechanically (perfect prosthesis).

Furthermore, there is evidence stemming from research for the perception of ultrasonic content in music. There is continuously ongoing research that continually push the boundaries of our knowledge about what and how we hear. So much work still needs to be done before we can have confidence in asserting what can be heard and what cannot be heard.

These studies partly fall in the realm of psychoacoustics; the psychological and physiological responses associated with sound, i.e. cognitive psychology and the effects that personal expectations, prejudices, and predispositions may have on listeners' relative evaluations and comparisons of sonic aesthetics and on listeners' varying determinations about the relative qualities of various musical instruments and performers. The expression that one "hears what one wants (or expects) to hear" comes to mind. This aspect cannot be dismissed: we are no androids. It is these 'flaws' if you want to call them that, that for no small part, determine what we hear....or perhaps even, on a subconscious level: what we choose to hear, or choose not to hear.

This the reason why I said "I don't think it is as clear cut as you may believe". No amount of noise spectrum images will tell me that I cannot prefer one format over the other. Or that euphonics cannot allow me to prefer high resolution audio over cd quality audio... We are no machines. You cannot apply machine rules to an analog being.

-rovo

[RD James]

The error you are making is comparing a 'PCM from DSD' signal with a pure PCM signal. This is not an honest comparison as you compare the former, which is constricted to the worst of both worlds with the latter, which has not suffered from any conversion.

It's a 24-bit 352.8kHz PCM file. You're not losing anything in that conversion. I've since tried a 32-bit conversion (it defaulted to 24-bit) and there was no difference in the result.
The conversion from DSD to PCM is generally considered "lossy" because it is not a reversible operation.
It is not reversible because DSD is a 1-bit format which has a lot of ultrasonic noise that must be filtered out for playback (the "lossy" part) and conversion from PCM back to DSD adds more noise on top of the original source, since it is the DSD64 format which is adding that noise.
 
As it was the point of this demonstration, I did not use a filter on the ultrasonic noise inherent to the DSD format, and obviously it was not converted back to DSD afterwards.
So what you're seeing is as "honest" a comparison as one could make, since it is the closest thing you can get to displaying the "raw" signal with a DSD64 source.
 
I suppose that is the "appeal" as an archival format - it's like an analog master tape in the sense that you cannot do anything to it non-destructively. Frankly that just seems like a terrible idea to me, since "doing anything" also includes playback as you have to filter out that noise.
You would never want to use it for recording/mastering, since you cannot do any editing to it non-destructively, and the file will be flooded with high frequency noise.

Also, you leave out DSD128 etc. which have a dramatic effect on the noise shaping and the filter used. SACD players have always upsampled to DSD128, so the noise floor only becomes an issue when playing DSD64 files on your PC, for which the format was never intended. I think Both formats have their place and can sound terriffic...if played natively. This however, is seldomly the case, as native DSD ADC/DAC’s are rare and native PCM ADC/DAC’s are even rarer, and so comparisons become pointless, especially if the ADC/DAC’s used are “hybrid” types.

First of all, you're going to have to cite a source to back up the claim that all SACD players upsample to DSD128, because that's news to me.
Second of all, that doesn't do anything about the noise which is already in the file as a result of being encoded to DSD64.
It is the encoding process which adds that noise, not playback. Upsampling can't change it - you need to filter out the noise.
 
Looking at the only "DSD128" album I have here, it appears to be a conversion from an undithered 24-bit PCM track - the noise floor bottoms out at -144dB, and is actually quite a bit higher than that in the lower frequencies.
Since it's double the sample rate, the ultrasonic noise starts around 50kHz rather than 25kHz. Otherwise, it's largely the same as DSD64.

Furthermore, there is evidence stemming from research for the perception of ultrasonic content in music. There is continuously ongoing research that continually push the boundaries of our knowledge about what and how we hear. So much work still needs to be done before we can have confidence in asserting what can be heard and what cannot be heard.

It's all just noise and spurious tones up there. The only thing to be found in the 20-50kHz region in this example (a randomly picked SACD track) is not music:



 
If you go by the assumption that we can detect or respond to ultrasonic frequencies somehow, looking at the data present in "HD" audio tracks shows us that it needs to be filtered out!
Best-case scenario for that, is that all it does is add some distortion in the audio band rather than us being able to detect it through some other means.

[rovo]

You've invalidated your arguments against DSD and are just continuing to go down the same road some more. DSD and PCM are different formats. They have their own strengths and weaknesses. As soon as you convert one format into the other you throw away the strength of the format you're converting from (DSD), and cannot automagically have the signal make use of the strenth of the format you're converting to (PCM).
This is why I said you're comparing one signal that is constricted to the worst of both worlds with one that hasn't suffered from any conversion.

You cannot convert DSD to PCM and then compare the result with a native PCM signal and say 'it's as honest as it gets'. It is in fact totally bogus.

Let me rephrase this. You cannot compare the objective strengths (which you are not interested about) and weaknesses of two completely different and almost entirely incompatible signals by converting one (DSD) into the same format as the other one (PCM), discarding almost 90% of the strength of the DSD signal and pretend it is an honest comparison.
Just because you lack knowledge and tools to compare the native DSD with native PCM doesn't mean it is an honest and valid comparison.

The other way around would be just as unfair. You're just taking shots at DSD, while shooting from the hip...and hoping people won't notice your aim is terrible....as long as they hear the sound of your gun.

-rovo

[rovo]

As for your argument against High res music. You're picking a random track from HDTracks (which is known to sometimes sell upconverted CD quality music) and pretend it is an honest representation of all high res audio.

If you were truly honest you would pick a track which is known to have superior sound quality, because every effort has been made during the entire production chain to preserve as much of the original signal as possible.
The way you go about makes you appear you're hell bent on discrediting all high res music.

-rovo

[RD James]

You've invalidated your arguments against DSD and are just continuing to go down the same road some more. DSD and PCM are different formats. They have their own strengths and weaknesses. As soon as you convert one format into the other you throw away the strength of the format you're converting from (DSD), and cannot automagically have the signal make use of the strenth of the format you're converting to (PCM).
This is why I said you're comparing one signal that is constricted to the worst of both worlds with one that hasn't suffered from any conversion.
You cannot convert DSD to PCM and then compare the result with a native PCM signal and say 'it's as honest as it gets'. It is in fact totally bogus.
Let me rephrase this. You cannot compare the objective strengths (which you are not interested about) and weaknesses of two completely different and almost entirely incompatible signals by converting one (DSD) into the same format as the other one (PCM), discarding almost 90% of the strength of the DSD signal and pretend it is an honest comparison.
Just because you lack knowledge and tools to compare the native DSD with native PCM doesn't mean it is an honest and valid comparison.
The other way around would be just as unfair. You're just taking shots at DSD, while shooting from the hip...and hoping people won't notice your aim is terrible....as long as they hear the sound of your gun.

Are you aware that you don't get a waveform from DSD without conversion? It's a 1-bit PDM format (pulse density modulation) so it needs to be low-pass filtered.
The DXD format - 24-bit 352.8kHz PCM - was created specifically for the Pyramix DSD workstation and is considered to be "transparent" even if it is not technically lossless, since the conversion is not reversible.
Again: it is not a losslessly reversible conversion due to DSD limitations, not PCM limitations. The data rate if DXD is three times that of DSD64. The issue is that there is no lossless way to convert audio back to DSD.
There is nothing magical about DSD - it's still trying to represent the same waveform as PCM. Its 1-bit nature brings a lot of limitations with it however. Everything is multi-bit these days for a reason.
 

As for your argument against High res music. You're picking a random track from HDTracks (which is known to sometimes sell upconverted CD quality music) and pretend it is an honest representation of all high res audio.
If you were truly honest you would pick a track which is known to have superior sound quality, because every effort has been made during the entire production chain to preserve as much of the original signal as possible.
The way you go about makes you appear you're hell bent on discrediting all high res music.

Check your high res files and see what you find.
The upconverts have a steep rolloff around 22.05kHz and don't extend the frequency response beyond that, instead of a smooth rolloff.



It's the "true high res" files which have noise and spurious tones in the 20-50kHz region.
Not all of them - some of them do have a clean (though largely empty) high frequency response, but at least 50% if not more of the "high resolution" tracks I see are either upconverts or would be better served by filtering out the HF junk that is inaudible by itself but may cause problems in the audible band when it's played back due to distortion.

[rovo]

Presenting a DSD signal after converting it to PCM tells us nothing about the original, native DSD signal. It strictly tells us about the conversion algorithms. There is no transparent conversion possible from DSD to PCM.

If we convert from DSD to DXD, that is 1-Bit at 2.822MHz to 24-Bit at 352.8kHz – we need to throw away 87.5% of the time domain information of DSD..
(Conversely, if we convert from 24-Bit at 352.8KHz (DXD-PCM) to 1-Bit at 2.822MHz (DSD) – we need to throw away around 99.96% of the amplitude information the PCM format is capable off, while we are only having 12.5% of the time domain information that the DSD system is capable of.)

So in effect we get the worst of both formats, rather than the best of one.

I do however agree that there is a lot of junk being sold in the high res market.

-rovo

[rovo]

...this is why it is silly to have MC bitstream your PCM files as DSD... You gain nothing; you only lose. PCM should remain PCM and DSD should not be converted to PCM.

-rovo

[Fitzcaraldo215]

I am a lover of hi Rez audio.  I have not bought and I have scarcely listened to anything not natively recorded in hi Rez for the last 10 years, mostly in Mch.

But, I also dislike conversions. I agree, and there is no appeal to me for a conversion of PCM to DSD.  One problem has always been trying to get an accurate playback level match between a DSD and a PCM version of the same recording.  This might also manifest itself in differences of apparent dynamic range.  So, I could never honestly be sure that one was superior to the other.

OTOH, I do prefer converting DSD to 176k PCM in order to use DSP for speaker distance correction, bass management and Room EQ.  In spite of the level match issue, that sounds best to me in obvious ways.

[RD James]

Presenting a DSD signal after converting it to PCM tells us nothing about the original, native DSD signal. It strictly tells us about the conversion algorithms. There is no transparent conversion possible from DSD to PCM.
If we convert from DSD to DXD, that is 1-Bit at 2.822MHz to 24-Bit at 352.8kHz – we need to throw away 87.5% of the time domain information of DSD..
(Conversely, if we convert from 24-Bit at 352.8KHz (DXD-PCM) to 1-Bit at 2.822MHz (DSD) – we need to throw away around 99.96% of the amplitude information the PCM format is capable off, while we are only having 12.5% of the time domain information that the DSD system is capable of.)
So in effect we get the worst of both formats, rather than the best of one.
I do however agree that there is a lot of junk being sold in the high res market.

I'm not sure that you understand the difference between the way that PCM and PDM encode data.
There is no "time domain information" being lost in this conversion. Is it not clear enough from this image that everything above 176.4kHz is all noise?
DSD requires such a high sample rate because it is only 1-bit and everything is noise-shaped.
Even ignoring the fact that it's all noise from the noise-shaping process, are you suggesting that humans can perceive audio from 176.4kHz to 1.4112MHz?
 
Converting to DSD is a lossy process because the format is only 1-bit, which insufficient to eliminate distortion. You need at least a 3-bit system to fully dither any signal and eliminate distortion.
Even if you assume the claims made about the DSD format are true, for half the bandwidth of a DSD signal an 8-bit 176.4kHz noise-shaped PCM signal will outperform it in every way.

...this is why it is silly to have MC bitstream your PCM files as DSD... You gain nothing; you only lose. PCM should remain PCM and DSD should not be converted to PCM.

Virtually all DACs these days - even those supporting "native" DSD playback - are multi-bit converters.
True 1-bit converters are a rare thing. (for good reason)

[rovo]

I'm gonna end this discussion. You keep countering my replies with false information. Or, perhaps I should say, alternative facts. This is, after all, the age of trump, where truth don't matter and facts are overrated.

You either don't know what you're talking about or, which I suspect, willingly and knowingly come up with statements that have no basis in facts. Every reply you've made seems to suggest your goal is to discredit high res PCM and DSD. You use bogus comparisons, counter my logic with a nonsensical statements, and fail to properly explain yourself.

I have better things to do with my time than to play pigeon chess. I will leave you to your crusade against anything high res. I'll just put on another SACD in my player and do something truly outrageous: Lean back, close my eyes and enjoy the music.

-rovo

[RD James]

I'm gonna end this discussion. You keep countering my replies with false information. Or, perhaps I should say, alternative facts. This is, after all, the age of trump, where truth don't matter and facts are overrated.
You either don't know what you're talking about or, which I suspect, willingly and knowingly come up with statements that have no basis in facts. Every reply you've made seems to suggest your goal is to discredit high res PCM and DSD. You use bogus comparisons, counter my logic with a nonsensical statements, and fail to properly explain yourself.
I have better things to do with my time than to play pigeon chess. I will leave you to your crusade against anything high res. I'll just put on another SACD in my player and do something truly outrageous: Lean back, close my eyes and enjoy the music.

Here's an AES paper from 2001 to start with, if you are going to suggest that I am making things up: https://sjeng.org/ftp/SACD.pdf
I have provided examples to back up my statements while you've offered nothing more than calling it "false information".

[rovo]

Back to my original question about  dynamic range of DSD vs PCM.

@AlanDistro
It would appear you actually can influence dynamic range by converting into a different format. Converting PCM to DSD (or DSD64 to DSD128) influences (improves) the noise floor. Since dynamic range is measured from noise floor upwards it follows that when you lower the noise floor this automatically improves dynamic range.

See: http://samplerateconverter.com/content/dsf-oversampling-d64-vs-d128-real-advantage-sound

-rovo

[rovo]

OTOH, I do prefer converting DSD to 176k PCM in order to use DSP for speaker distance correction, bass management and Room EQ.  In spite of the level match issue, that sounds best to me in obvious ways.

I may seem to contradict some of my earlier statements but I do actually convert PCM to DSD128. I have a Calyx M. DSD128 sounds superior to anything else on it. I especially like the sound of the Calyx M on the AUX connection in my truck.
I obviously only convert music that I feel benefits most... For instance, I have some Manu Katché, Steely Dan, Jerry Granelli and a few others that sound nothing short of breathtaking, especially when played loud The utter and complete blackness of silence in between sounds on the Calyx M is almost scary.

Everything else (pop, rock) that isn't CD quality already gets downsampled and stored on my Calyx M as 24/44.1 or 24/48. I could almost certainly get away with 16bit for those; I'd have to try it out some day.

-rovo

[kr4]

I may seem to contradict some of my earlier statements but I do actually convert PCM to DSD128. I have a Calyx M. DSD128 sounds superior to anything else on it. I especially like the sound of the Calyx M on the AUX connection in my truck..................................................
............................. The utter and complete blackness of silence in between sounds on the Calyx M is almost scary.

You must have an unusually quiet truck.

[rovo]

You must have an unusually quiet truck.

Heh, yeah I suppose I was rambling on a bit about my Calyx M. Obviously I wasn't referring to silence while driving. I have several pairs of excellent headphones that make its exceptional quiet background quite noticeable.

-rovo