PCM vs. DSD: Encoding Method Accuracy, SINAD, and What Actually Limits Modern DACs

[AUDACC]

Now that there are two high value DACs, the Topping D50iii and SMSL D200, and high resolution music files available at reasonable prices, there are HiFi newcomers and new HiFi oldcomers that have an interest in comparing music encoding formats. Here are some of the questions that might differentiate DSD from PCM.

Does the encoding method (PCM vs. DSD/PDM) change how accurately 20Hz–20kHz audio is delivered? Which one measures better on SINAD and how bit depth/sample rate and noise-shaper order actually affect accuracy? Do modern DACs treat the two formats equally. The short answers follow, details and sources are in attachments. The questions are mine. Claude helped with the research.

• For the same master, a well-engineered PCM file (16/44.1 or higher) and a well-engineered DSD file (DSD64 or higher) can both deliver the 20Hz–20kHz content with distortion and noise far below audibility. Neither file is inherently "more accurate" in-band; the achievable accuracy is set by the recording/mastering bit depth/sample rate for PCM or modulator order/oversampling ratio for DSD, not by the encoding format per se.
• On SINAD specifically, well-implemented PCM almost always measures better than DSD in modern DAC tests, mainly because high-order noise-shaped 1-bit DSD pushes a lot of noise just above the audio band, and that noise tends to be captured in the measurement bandwidth of the filter output. In contrast, for multibit PCM less noise above the audio band is passed through the DAC processing to the analog reconstruction filter. Real DAC measurements (cited in attachments) consistently show this.
• PCM accuracy scales with bit depth per the standard 6.02N + 1.76 dB relationship; sample rate by itself doesn't improve in-band SINAD, it only moves the Nyquist frequency and noise-shaping/dither headroom out further.
• DSD accuracy is set by noise-shaper order and oversampling ratio (i.e., DSD64/128/256/512), not "bits" in the PCM sense, since DSD is 1-bit. Counter-intuitively, real-world measurements show DSD256 often measuring worse than DSD64/DSD128 on common ESS-based silicon — a modulator/implementation artifact, not a fundamental DSD law.
• Modern DACs do not process PCM and DSD with equal accuracy in practice. Most popular delta-sigma DAC chips (ESS Sabre, etc.) run an internal multibit PCM-like path; DSD is converted/expanded internally and generally measures a few dB worse than the chip's best PCM SINAD on the same hardware.
• DAC SINAD does not keep climbing with higher PCM sample rate. It plateaus (and sometimes regresses) beyond 96–192kHz because the analog noise floor — not the digital format — is the bottleneck.
• DSD can sound different from PCM of the same recording due to several measurable DAC-level mechanisms: differing reconstruction/low-pass filter characteristics (typically gentler in DSD mode), the character of ultrasonic DSD noise interacting with amplifiers and speakers, and the choice of delta-sigma modulator oversampling rate. These are real, measurable, hardware-level differences — not mystical properties of the 1-bit format.

 

[Ricardus]

Yes.

 

[HalSF]

Valid and interesting topic, but the intellectual integrity here is totally ruined for me by the use of AI slop instead of well-informed human reasoning, research, and expression.

 

[AUDACC]

Valid and interesting topic, but the intellectual integrity here is totally ruined for me by the use of AI slop instead of well-informed human reasoning, research, and expression.

All the conclusions in original post are backed up with factual data. They can be viewed in the PCM DSD zip file attachment.

 

[danadam]

PCM accuracy scales with bit depth per the standard 6.02N + 1.76 dB relationship; sample rate by itself doesn't improve in-band SINAD, it only moves the Nyquist frequency and noise-shaping/dither headroom out further.

If we have fixed bandwidth, e.g. 0-20k, then SINAD in that bandwidth improves with sample rate, because noise is 3 dB lower for 2x higher sample rate.

 

[AUDACC]

If we have fixed bandwidth, e.g. 0-20k, then SINAD in that bandwidth improves with sample rate, because noise is 3 dB lower for 2x higher sample rate.

DAC PCM Performance plateaus, and sometimes degrades, beyond a certain sample rate. The theoretical reason is the sample rate moves the Nyquist point, not the quantization noise floor; once dither/noise-shaping has already pushed quantization noise inaudibly low, and the DAC's analog noise floor becomes the bottleneck, more samples per second has nothing left to improve. Archimago's testing found that going from 96kHz to 384kHz PCM test signals did not yield a cleanly scaling SINAD improvement, and that DSD512's noise floor was essentially identical to PCM out to 192kHz — i.e., once operating with a clean noise floor, going higher in rate buys flat performance, not continued gains. On ASR's comparative DAC charts, the highest-SINAD units are differentiated by analog design and implementation quality, not by sample rate.

 

[krabapple]

ok, and...?

 

[danadam]

the sample rate moves the Nyquist point, not the quantization noise floor; once dither/noise-shaping has already pushed quantization noise inaudibly low, and the DAC's analog noise floor becomes the bottleneck, more samples per second has nothing left to improve.

Sure, if the noise floor in the signal is already lower than the device noise floor then using higher sample rate won't change much, but the same is true for bit-depth. Using higher bit-depth also won't change much in that situation.

Archimago's testing found that going from 96kHz to 384kHz PCM test signals did not yield a cleanly scaling SINAD improvement

If he started with 16-bit/96k signal then both increasing bit-depth and increasing sample rate would show improvement, in a fixed bandwidth. 6 dB per bit and 3 dB per doubling sample rate, respectively.

 

[AUDACC]

Sure, if the noise floor in the signal is already lower than the device noise floor then using higher sample rate won't change much, but the same is true for bit-depth. Using higher bit-depth also won't change much in that situation.


If he started with 16-bit/96k signal then both increasing bit-depth and increasing sample rate would show improvement, in a fixed bandwidth. 6 dB per bit and 3 dB per doubling sample rate, respectively.

Thanks for clarifying Archimago’s test results.

The goal of HIFI is to deliver the sound of a live music performance as accurately as possible to an individual’s remote listening space. Reviews in the last couple of years show reasonably priced amplifiers to be essentially transparent. The information complied onto this thread indicates that the recorded music delivery mechanism (streaming/CD/digital download) and the DAC have also become essentially transparent at very reasonable prices.

Now we can move our effort and resources (money) to areas where transparency can be improved such as recording/mastering, speakers, room acoustics. My Hi Fi system, thanks to high speed internet and the recent purchases of a modern amplifier and DAC is now good enough to warrant paying more for better speakers, room treatment or a premium for a great recording.

 

[DSJR]

I'd respectfully suggest that the goal of a HiFi or monitoring system is to reproduce as faithfully as possible, what the mastering or cutting engineer wanted us to hear in the commercial end result that we stream or purchase. We can't go back as end-users to the studio or live venue as it was recorded sadly. Apologies for the pedantics

With digital today, the speakers/room interface can be the most important thing by miles, but with vinyl sources (for those who still use the format as I often do), the source-first hierarchy is still valid to a large degree, although the quality of sound available from even humble players has improved so much with a decent stylus fitted, surprisingly so to an old hand like me...

 

[AUDACC]

I'd respectfully suggest that the goal of a HiFi or monitoring system is to reproduce as faithfully as possible, what the mastering or cutting engineer wanted us to hear in the commercial end result that we stream or purchase. We can't go back as end-users to the studio or live venue as it was recorded sadly. Apologies for the pedantics

With digital today, the speakers/room interface can be the most important thing by miles, but with vinyl sources (for those who still use the format as I often do), the source-first hierarchy is still valid to a large degree, although the quality of sound available from even humble players has improved so much with a decent stylus fitted, surprisingly so to an old hand like me...

The insight you bring is important. The low hanging fruit in the vinyl mastering/cutting engineering world is gone. But now many digital sourced home music systems are transparent enough to hear the difference. Paying a premium for what the end user considers an outstanding recording indirectly encourages the recording/mastering engineer community to improve the finer nuances of their techniques or equipment on future recordings.

 

[Mart68]

Now we can move our effort and resources (money) to areas where transparency can be improved such as recording/mastering, speakers, room acoustics.

I'd suggest that's been the case since CD arrived. Leaving out recording/mastering which we can't do anything about anyway.

 

[MaxwellsEq]

Paying a premium for what the end user considers an outstanding recording indirectly encourages the recording/mastering engineer community to improve the finer nuances of their techniques or equipment on future recordings

I would REALLY love to agree with you, but unfortunately, in my experience this is not how the commercial aspect works in the slightest.

I worked in professional recording studios. Even before digital recordings, with Dolby SR analogue desks and magnetic tape at 15 IPS could sound amazing. Yes, without noise reduction, tape hiss was occasionally audible in the quietest classical passages, but the frequency response was pretty good from 30 Hz to 10kHz and the dynamic range excellent.

Unfortunately this is never what gets shipped. In the LP era, deep bass and high dynamic range could not be cut and high frequency is all resonance. But even in the 1990s when dither made CD SNR better than your ears can withstand and DC to 20kHz was flawlessly encoded, all commercial content is compressed relative to what comes out of the mixing desk.

Very high dynamic range simply will not sell. Buyers will complain that the quiet bits are inaudible (which they will be in any domestic room).

 

[Redacted]

Perhaps it's a betamax v. VHS moment, and DSD will go the way of betamax for similar reasons.

 

[Sir Sanders Zingmore]

The information complied onto this thread indicates that the recorded music delivery mechanism (streaming/CD/digital download) and the DAC have also become essentially transparent at very reasonable prices.

We knew that already. Has something new come to light in this thread?

 

[AUDACC]

I would REALLY love to agree with you, but unfortunately, in my experience this is not how the commercial aspect works in the slightest.

I worked in professional recording studios. Even before digital recordings, with Dolby SR analogue desks and magnetic tape at 15 IPS could sound amazing. Yes, without noise reduction, tape hiss was occasionally audible in the quietest classical passages, but the frequency response was pretty good from 30 Hz to 10kHz and the dynamic range excellent.

Unfortunately this is never what gets shipped. In the LP era, deep bass and high dynamic range could not be cut and high frequency is all resonance. But even in the 1990s when dither made CD SNR better than your ears can withstand and DC to 20kHz was flawlessly encoded, all commercial content is compressed relative to what comes out of the mixing desk.

Very high dynamic range simply will not sell. Buyers will complain that the quiet bits are inaudible (which they will be in any domestic room).

Well I guess that leaves speakers and room treatment.

I resent having to exercise the volume remote between quiet and loud classical passages and do support a bit narrower dynamic range with low THD, IMD and without clipping.

 

[AUDACC]

We knew that already. Has something new come to light in this thread?

No new breakthroughs in this thread. I just wanted to make available in a compact package the data and reasoning that shows well-done digital encoding, file delivery and decoding could be transparent and not marketing hype. HiFi newcomers it might save a lot of research time in selecting their Hi Fi components if they read this thread.

 

[Redacted]

Where did clipping come into it?
I think newcomers are just going be either bored or confused and will buy PCM recorded material and play it back on commonly available and affordable kit.

 

[MaxwellsEq]

I resent having to exercise the volume remote between quiet and loud classical passages

That's the problem - so do many other people. It's been possible for several decades to give people content that, its loudest would deafen them whilst at its quietest would be quieter than any domestic room's background noise. But consumers won't accept this. The quiet bits must still be audible whilst vacuuming the carpet .

 

[DavidEdwinAston]

That's the problem - so do many other people. It's been possible for several decades to give people content that, its loudest would deafen them whilst at its quietest would be quieter than any domestic room's background noise. But consumers won't accept this. The quiet bits must still be audible whilst vacuuming the carpet .

Anybody who vacuums the carpet, while listening to music, should be burned at the stake!