Is DSD a scam? — No. But the myths are.


1. Let's get one thing straight


DSD is DSD — there is no such thing as "fake DSD." Once a file is wrapped in the DSD format, it is DSD. Quality differences come from the SDM modulator's competence, not the format label.


There is only one thing worth discussing: noise shaping performance.


2. The technical essence of DSD


DSD is built on ΔΣ modulation (SDM) — combining an extremely low bit depth (1 bit) with extremely high sample rates (from 2.8224MHz), using noise shaping to push quantization noise into the ultrasonic band in exchange for high dynamic headroom in the passband.


This is neither mysticism nor a scam. It is fundamental signal-processing theory with nearly 60 years of academic accumulation. The only question is the quality of engineering.


3. Problems with those "DSD debunking" articles


The "DSD exposed" articles circulating online have several shaky points:


1. "Theoretical dynamic range is only 6dB"


The number is technically correct, but it ignores the noise-shaping context. It's like saying "a race car has no spare tire, how can it drive long distances?" — race cars are not measured by that metric. DSD's noise shaping is its core technology, not a bolted-on patch.


2. "SNR drops to 60dB above DSD128"


This contradicts measurements. DSD's noise-shaping characteristic is: the higher the sample rate, the lower the in-band noise. Going DSD64→128→256, every doubling adds more modulator headroom. SNR degradation can only come from poor modulator design, not the format itself.


3. "90% of DSD is upsampled PCM"


PCM-to-DSD conversion and "faking" are two different things:


  • Low-quality route: run 44.1/16 through a poor filter and wrap it as DSD → that's low quality, not fake
  • High-quality route: a high-order SDM modulator (taps=4096, order=7, oversampling=64) faithfully re-encodes PCM information into the DSD domain → this is a meaningful format migration

The latter removes the pre-ringing of PCM digital filters and gains a cleaner time-domain response. This is not placebo; it is quantifiable.


4. "Blind test recognition rate of 52%"


The 52% figure recurs in blind-test literature, but the tests themselves have several layers of problems:

  • Most DACs internally transcode DSD to PCM — they test the format label, not the format substance
  • Short ABX switching doesn't favor hearing SDM's time-domain characteristics
  • At small sample sizes, 52% has no statistical power

DSD's differences require a native DSD decoding chain + extended listening to manifest. This is not mysticism; it's a test-condition problem.


4. DSD's real shortcomings


Not making excuses — DSD has genuine limitations:


  • Editing chain: no direct time-domain slicing, fades or EQ; you must work in the PCM domain and convert back
  • File size: DSD64 already exceeds 24/96 PCM; DSD256 and above are larger still
  • Scarcity of native recordings: true all-DSD recordings are extremely rare and costly; most DSD material has passed through PCM
  • Demands on downstream gear: ultrasonic modulation noise requires amplifiers and drivers with good response characteristics; not every system suits it

5. Where noise-shaping quality shows up


1. Noise floor texture


Poor noise shaping leaves the audible-band noise floor uneven — hiss with coloration, grit and grain. Done well, the noise floor is clean and flat, the background is "black."


→ Most audible in quiet passages or the silences between movements. If a piano's decaying harmonics get rendered with a "sandy" quality — that's poor shaping.


2. Transient leading edge — pre-ringing residue


DSD's core advantage over PCM is in the time domain. But only if the modulator is well written.


Poor: transient leading edges blurred and softened. Good: crisp and clean, with defined attack placement and air.


→ Most audible in percussion, brass attacks, and consonant sounds (s/t/p) in vocals.


3. "Air" vs "digital harshness" in the highs


If noise-shaping energy isn't controlled well, the upper audible band (8k-20kHz) carries audible spurious components — the highs sound "hairy," "rough," "strained."


Done well, high-frequency extension is open and natural, with neither grit nor a "veiled" quality.


→ Most audible in female vocal overtones, high positions on violin, cymbal decay.


6. Conclusion


DSD is not a scam. DSD is a digital audio format with specific technical characteristics and genuine pros and cons.


The scam is when the market sells it as a totem.


  • "DSD crushes PCM" is a scam
  • "DSD is a scam" is also a scam

The value of a technology is defined neither unilaterally by its flaws nor by its marketing volume. It is defined by engineering quality, scenario fit, and your ears.


The yardstick is simple: it's not what format you use, it's what you make with it.


_Audio tool makers don't take sides on formats — they take sides on quality._