PCM vs DSD?

Against my better judgment I'll chime in briefly on this one. Back when I was archiving my vinyl collection with my Tascam DA-3000, I tried PCM and DSD. Yes there was a difference in sound and loudness, almost like playing with a filter or a loudness button... Which one was better was up for debate depending on the ripped track. Ultimately I decided to archive and convert everything to PCM for the sake of sanity, and convenience.

If someone wants to play with filters and software that's cool and I have no issue with that. I prefer a more simple straightforward approach when listening to music, to each their own.

 

DSD is a lossy PCM storage format that takes up more space than 16-bit, 44.1 kHz PCM. It is always much more distorted than the PCM file from which it was made.

The reason many SACDs sounded different than the contemporary CD was the Scarlet Book has strict modulation requirements to not blow up most playback systems using only the 50 kHz, 30 dB per octave recommended analog filter. EDIT: and so the laser could track it. The recommend metering for these correlates more to the signal peaks than simple 16-bit, 44.1 kHz sample levels. Even then, they had additional audible distortion compared to the contemporaneous overmodulated CD versions over the entire length of the signal versus keeping brief moments of high distortion within the legal bandwidth to what were brief transients. Of course the cleanest way to play back the signal encoded within the 1-bit, 2822.4 mHz format is to feed it into a normal DA converter and filter out the massive noise. The overmodulated CD audio on the dual layer SACDs could simply be played back. A 44.1 kHz, 16-bit PCM file that meets those audio modulation requirements will be cleaner than the 1-bit 2822.4 mHz one because the entire DSD signal will have quantization distortion correlated to the encoded signal and all positive samples are by definition clipped in a 1-bit signal.

Foolish audiophiles on the internet have neither any knowledge of the modulation requirements nor how to make an audio signal conform to them.

 

I'm with @Psalmanazar, that short of a theoretical advantage of DSD which I understand it does not have, what's the point of recommending DSD over PCM in some general "best practices" sort of way? Perhaps @EagleWings was not meaning to do this. We all know (or should) that it is all in the implementation, or should. In a real sense, "PCM vs DSD" is the wrong question, and the only meaningful answer is "depends".

@GoldenOne recently did a review of the TA DAC200 (a high end $6k dac) where he thought the PCM was easily bested by many lesser priced dac's, but that the DSD implementation in it was real good. Ok, as long as you go into such a set up with eyes open.

As far as taking the time/effort/$ in setting up computers/software/dac chain's for HQP>DSD>some specific DAC well ok, but let's be honest that even in the audiophile niche, this is not going to be for most.

In the past when I messed around with Roon/HQP DSD sent to various SD dacs, I thought the sound was a bit smoothed over, reminding me of a clean, clear, yet somewhat squashed high feedback amp. The dacs were not optimal however. One thing fur sur, I'd rather have a TOTL PCM dac (such as the wavedream) than a more convultoed PC/HQP/DSD dac chain based simply on cost and complication.

 

I believe the more complex (but highly customizable) way is to build out your own hardware to run something like HQPlayer to do the PCM -> DSD conversion before sending the upsampled audio to a DAC that can accept it. The easier way is to have a DAC that automatically upsamples everything to DSD before outputting (though you don't have the ability to play with how much you want to upsample nor what filters you want).

I thought I was firmly on team r2r multibit after moving up the Schiit and Rockna ladders but I ended up trying a modded version of the latest Direcstream with the latest firmware (which automatically upsamples everything to DSD256) and I had to admit it was more than just a side-grade to my Wavedream Signature by many DAC metrics. Similar ability to resolve detail but much better decay, timbre, spatial stuff, etc. I think there's something great that happens to spatial resolution from the upsample that you might not get without it. It was a clear level up from my WD Sig. I haven't tried a ton of DACs but this is a hunch.

I'd love to hear more hardcore setups (DSD512+) to see what's possible. I came into this without any sort of bias toward DSD and was pleasantly surprised.

I get it-the math says otherwise but I'm here to listen to music. Forest for the trees and all that.

 

DSD cannot overcome the mathematical limitations of the 1-bit quantizer.

The technically more correct way to mess around would be to feed the da chip an already massively upsampled and dithered fixed bit pcm signal on a supposed pin to bypass the internal upsampling and anti-image filter but I gave doubts on if that would even matter theoretically given the massive noise introduced and then shaped and filtered away and the chips generally have less harmonic distortion at 44.1 and 48 kHz.

The DACs utilizing asynchronous sample rate conversion for jitter reduction are already doing this and hitting the chips with an optimal sample rate that cannot be easily determined by the end user. Their designers and manufacturers thought of this years ago. Benchmark and Crane Song at 211 kHz. Weiss at 195 kHz. Apogee at an unspecified sample rate. The modern chipsets make this easy. The only thing you gain for playback by feeding them a different sample rate is the anti-imaging filter is placed higher up but with adequate filters this scarcely matters.

 

I know. I'll pay for shipping both ways if you can send me a better DAC than what I have. I've had people over that have heard both Wavedream Signature and modded Direcstream mk2 in my room as well as a blind A/B session with both DACs side by side. Happy to give a DAC that is better at math an equal shot and get at least as many ears on it.

I'm all for shitting on useless things like MQA or mpingo discs for acoustic treatment but, as of today, the best DAC I've ever owned upsamples everything to DSD256.

 

I'm wondering if the APS permalloy OPTs have anything to do with it?

 

Famish had time with a stock DS mk2 and said it was the first DAC he seriously considered swapping out his Wavedream for. He was one of a few people that put it on my radar, actually.

 

Both have output transformers though? That's what I'm getting at.

 

Yeah, they're partly being used as filters. "...the directstream transformers act as a low pass filter - to filter out all the unwanted high frequency noise from the DSD signal to only let pass the relevant (audio) frequencies. This is where the digital signal gets truncated to act as an analog output line level voltage signal."

 

I know my friend you are being serious, but this stuff hurts my head so much that I won't listen to music tonight but instead put some ribeyes on the grill and smoke a cigar.

 

If you aren’t interested in the conversation then don’t participate in it.

 

Negative. I am calling out how absurd this topic is in 2024. This is borderline OAFAS talk.

Time for some steak, yum.

 

I feel like this is one of those topics that pertains too much to specifics. The specific DSD/PCM implementation in the DAC and the shaping algorithms you use for either, is 90% of the impact on the DSD/PCM difference, not the actual format itself.

 

The distortion of DSD is a direct result of the 1-bit quantizer, keeping it 1-bit, and trying to use an steep analog filter. PCM is linear except for a fixed noise-floor and the Nyquist limit. DSD is just the output of a dysfunctional 1-bit delta-sigma PCM ADC taken prior to the anti-alias filter and decimation.

In "pure" 1-bit DSD DACs, the 1-bit stream is just converted to analog by low passing it around 50 kHz with a steep analog filter, always minimum phase of course, causing further phase shift and not eliminating the high frequency noise.

Feeding the DSD into a PCM DAC will often bypass the upsampling, apply a digital minimum phase filter at 50 kHz, then run it through the usual DAC's processing with multi-bit quantizers and such but with phase shift compared to playing back PCM.

Filtering and decimating the higher sample rate, 1-bit DSD stream to lower sample rate, multi-bit PCM and then running it into the PCM DAC as PCM cuts out more the high frequency noise and has less phase shift. The cleanest way to play back a 1-bit PCM file, already screwed over by the 1-bit quantifier, is to finish the conversion to multi-bit PCM and play it back as such. LOL.

https://www.analog.com/en/resources/technical-articles/sigmadelta-adcs-tutorial.html

 

Nice. So it's cool for you to deflect, but not somebody like Colt? Fix your head:

https://www.betterhelp.com/

Seriously, OAFAS? It's f'ing ORFAS you dumb shit stain.

 

that's a different discussion independent of the inadequacies of a 1-bit quantifier that correlates more to "what can humans build for cheap?"

you can get far cooler distortions than DSD conversion and playback provides without any chirping to pollute your blackground or excessive high frequency noise to stress your tweeters: limiters, pushed tube amplifiers, transformer hysteresis, etc.

 

On paper, pcm dacs all the way. In the listening room, dsd. As time goes on and things improve that may change.

 

There are plenty of things that could be cleared up a bit. Delta sigma DACs already have "digital oversampling filters" although they are often called "interpolation filters". Interpolation filters are digital filters, both the input and the output is digital. The analog filters built into the DAC are analog as the name suggests and they start to cut off well over the nyquist frequency. The analog filter in the DAC, usually called the "reconstruction filter" doesn't require any processing power at all. Both the input and the output of the reconstruction filter is analog. The part about the delta sigma modulator converting to essentially DSD is correct though. Something to note is that these modulators can be multi-bit even in audio DACs so the output of the modulator is not always the typical 1bit DSD.

On the topic of phase, the interpolation filter used in DACs (and ADCs) could be either linear phase or minimum phase. Linear phase interpolation filters are virtually never used in applications where real-time audio is a requirement as they introduce large delays compared to minimum phase filters.

This delay means they aren't even an option for most professional applications. Try to talk/sing/play an instrument while you're hearing yourself back with >10~20ms delay and you'll understand what I'm harping on. I can't think of such a DAC being used outside of maybe mixing and mastering but I doubt that the typical studio would buy a separate DAC for that when they probably already sit on some kilobuck 50in - 50out audio interface.

If you looked up (and believed) the manufacturer's published specs about their interpolation filters I don't see why you would call these implementations poor. The linear phase ones usually don't touch the pass-band at all, and they attenuate the images adequately. There are some that cut off a bit too early but from what I've seen they are the exceptions. Minimum phase filters alter the pass-band which could be objectionable given how easy it has become to create a well behaved linear phase filter though.

About upsampling with HQplayer, you can upsample with it of course but you won't bypass the upsampling built into the DAC (if there's one, some DACs don't have it). HQplayer upsamples to some rate, maybe 384kHz if the DAC even accepts that and after that, the DAC will still use its own internal upsampling algorithm because it works with much higher sample rates internally anyways.