Almost every digital recording you've ever heard is PCM — the format behind
CDs, WAV and FLAC files, and every streaming service. DSD is the other one:
the format on Super Audio CDs, sold as .dsf and .dff files, and prized
by a slice of the audiophile world. They solve the same problem — store a sound wave as numbers —
in almost opposite ways.
PCM: measure the height, many times a second
PCM (Pulse Code Modulation) is the intuitive approach. Many thousands of times a second it measures the instantaneous height (amplitude) of the wave and writes that number down. Two numbers define the quality:
- Sample rate — how often it measures. CD is 44,100 times a second (44.1 kHz); high-res files run at 96 kHz or 192 kHz. The sample rate sets the highest frequency that can be captured — half the rate (the Nyquist limit), so 44.1 kHz reaches just over 22 kHz, comfortably past human hearing.
- Bit depth — how finely each measurement is graded. 16-bit gives 65,536 possible levels; 24-bit gives over 16 million. More levels means a lower noise floor and more dynamic range (the gap between the quietest and loudest sound).
6.02 × bits + 1.76 dB. So 16-bit lands near 96 dB and 24-bit
around 144 dB on paper. In the real world no DAC or microphone reaches 144 dB; the
best converters top out near 120 dB. 24-bit's real value is headroom while mixing, not
144 dB you'll ever hear.
DSD: a torrent of single bits
DSD (Direct Stream Digital) throws out fine-grained levels entirely. Each sample is a single bit — just a 1 or a 0 — but it fires them extremely fast. DSD64, the SACD rate, runs at 2.8224 MHz: 64× the speed of a CD. Higher variants (DSD128, DSD256, DSD512) double from there.
Here's the part that's almost always explained wrong. A 1 does not mean "louder" and a 0 "softer" — that would be a different, cruder scheme called delta modulation. DSD uses delta-sigma modulation, where loudness is carried by the density of the 1s. A run that's mostly 1s represents a high part of the wave; a roughly even mix of 1s and 0s is the middle; mostly 0s is the bottom. Squint at the bitstream and the local proportion of 1s traces the original wave.
A single bit is a brutally coarse measurement, so this generates an enormous amount of quantisation noise. DSD's trick is noise shaping: the modulator shoves almost all of that noise up into the ultrasonic range, above ~20 kHz where you can't hear it. The result is excellent fidelity across the audible band — about 120 dB of dynamic range for DSD64 from 20 Hz to 20 kHz — at the cost of a steeply rising wall of noise above it. That ultrasonic noise is real, and it's why DSD and aggressive analog hardware sometimes don't mix.
Why DSD gets called "analog-like"
The strongest practical argument for DSD is what happens on the way back to analog. In its simplest form a DSD stream can be turned back into sound with little more than a gentle analog low-pass filter to strip the ultrasonic noise. PCM playback, by contrast, runs through digital reconstruction filters inside the DAC. Done well, those filters are inaudible; done poorly, a steep "brick-wall" filter can add pre-ringing and smear sharp transients.
So the "DSD sounds more analog" impression is less about the bits and more about the filtering — DSD nudges designers toward gentle filters. (Modern PCM DACs offer gentle filter options too, which is part of why the gap is smaller than it's often made to sound.)
The plot twist: PCM is born from a DSD-like stream
Here's the fact that reframes the whole debate. Almost every modern analog-to-digital converter is itself a delta-sigma converter. It doesn't measure neat multi-bit PCM straight off the microphone — it first produces a high-speed, low-bit pulse-density stream, then runs it through a decimation filter that trades that speed for bit depth to make PCM.
In other words, PCM is a processed derivative of a DSD-like stream. DSD is roughly that raw stream saved before the decimation step. Neither format is "closer to the original analog wave" in some pure sense — they're two destinations from the same starting point.
Side by side
| Format | Bits | Sample rate | Data rate (stereo) | Dynamic range |
|---|---|---|---|---|
| CD (PCM) | 16 | 44.1 kHz | ~1.4 Mbps | ~96 dB |
| Hi-res PCM | 24 | 192 kHz | ~9.2 Mbps | ~120 dB (real) |
| DSD64 (SACD) | 1 | 2.8224 MHz | ~5.6 Mbps | ~120 dB |
| DSD128 | 1 | 5.6448 MHz | ~11.3 Mbps | ~120 dB |
| DSD256 | 1 | 11.2896 MHz | ~22.6 Mbps | ~120 dB |
So which is better?
The honest answer: for the same well-made recording, far less than the marketing implies. In careful blind listening, high-res PCM and DSD are both effectively transparent — most "DSD sounds better" experiences trace back to a different master or a different DAC filter, not the format itself.
There's even a respectable engineering case against 1-bit DSD. Because a 1-bit quantiser can't be properly dithered, it can produce faint idle tones and far more ultrasonic noise than a modern multi-bit converter. Engineers like Benchmark's John Siau argue that today's multi-bit delta-sigma designs are measurably cleaner. They have a point — and DSD still sounds wonderful in practice. Both things are true.
The takeaways that actually matter:
- Mastering beats format. A great master in 16-bit PCM beats a mediocre one in DSD256, every time.
- Buy the music you want in whatever format it was done best in. Many classic recordings have superb DSD/SACD transfers — that's a fine reason to choose DSD, format theory aside.
- Your DAC and your files matter more than the DSD-vs-PCM label. Pick a player that plays both bit-exact, and you never have to choose.
Next: what "bit-exact" means and why it matters — the difference between hearing your files and hearing your operating system's idea of them. Or jump to playing DSD on an iPhone.