DSD: DoP versus Native-DSD (and Why Some "Native" DACs Aren't)

If you’ve shopped for a DSD-capable DAC in the last decade you’ve seen the same two acronyms on every spec sheet: DoP (DSD over PCM) and Native DSD. They’re usually presented as a quality choice — DoP for compatibility, Native for purity. The truth is more interesting and more annoying: at the wire, both transports carry the same audio. The thing that actually decides whether you’re hearing “real” DSD is what happens inside the DAC chip after the bits arrive — and several of the most popular chips (and several of the most popular implementations of those chips) do not preserve a 1-bit signal path even when fed native DSD.

This post is about that gap between marketing and silicon, with named hardware.

One bit, very fast

DSD (Direct Stream Digital) is the modulation Sony and Philips chose for SACD: a 1-bit stream sampled at 64x the CD rate (DSD64 = 2.8224 MHz), or its multiples — DSD128, DSD256, DSD512. Each sample is just “is the analog signal currently above or below the running average?” The price of one bit per sample is enormous ultrasonic noise, which a delta-sigma modulator pushes well above the audio band. The benefit is that the bitstream maps almost trivially to a switched analog output — no big multi-bit DAC array required.

That last part is the whole point of DSD: a 1-bit stream is supposed to drive the analog stage directly, with nothing in between. The moment you decimate it, filter it as PCM, or apply a digital volume control implemented as a multiply-and-truncate, you’ve stopped doing DSD and started doing math on a representation of DSD.

What’s actually inside a DoP frame

DoP (DSD over PCM) was published in 2012 by Andreas Koch (Playback Designs), Andy McHarg (dCS), and Rob Robinson (Merging Technologies) so that DSD could ride over existing PCM transports — USB Audio Class 2, S/PDIF, AES3 — without anyone having to rewrite drivers (dsd-guide.com).

The encoding is dead simple. For each 24-bit PCM frame:

The DAC watches for the alternating 0x05 / 0xFA pattern. If it sees the pattern hold for several frames, it switches its internal datapath from “PCM” to “DSD” and starts feeding the lower 16 bits straight into its DSD modulator. If the pattern breaks, it falls back to PCM. That’s the entire spec.

Because each 24-bit PCM frame carries exactly 16 DSD bits, the underlying PCM sample rate is DSD rate ÷ 16:

Eight of every 24 bits on the wire are pure marker overhead — about 33% of the link bandwidth is metadata. That overhead is real but it is lossless: the receiving DAC reconstructs the original 1-bit stream sample-for-sample.

What “native” looks like at the USB endpoint

“Native DSD” is the marketing name for several different things:

• USB Audio Class 2, alternate setting with FORMAT_TYPE_RAW_DATA — the modern, driver-free path on Linux and macOS. The interface descriptor advertises a non-PCM format and the host sends the 1-bit stream without DoP framing.
• ASIO 2.1 DSD mode — the Windows path. The ASIO driver exposes a DSD endpoint and the player writes raw DSD samples into it. Each driver vendor (XMOS, Thesycon, Amanero) implements this slightly differently.
• Vendor-specific control transfers — early DSD-capable DACs from Mytek, Playback Designs, and others used proprietary out-of-band signaling to switch the endpoint into a raw mode.

In all three cases the wire format ends up being the same 1-bit stream that DoP carried inside its PCM frames. The only difference is whether the marker bytes are present.

Key insight: A DAC that supports both DoP and Native DSD will, after de-framing the DoP marker bytes, present the same exact 1-bit sequence to its modulator. There is no possible audible difference between the two at the DAC input. Anyone who tells you DoP “sounds rolled off” or “sounds digital” relative to native DSD is hearing driver behavior, jitter, or — much more likely — a different processing path inside the DAC that the DAC switches to based on which transport it sees.

That last clause is where things get weird.

The chip decides, not the cable

A DAC chip’s published spec sheet will list things like “DSD64/128/256 native support”. What that phrase does not tell you is whether the chip’s internal datapath actually carries the 1-bit stream all the way to the analog output, or whether it decimates, filters, and re-modulates it along the way. There is no industry term for this distinction, which is why marketing has been able to elide it for over a decade.

Three things commonly happen to “native” DSD inside a “native DSD” DAC:

1. Decimation to high-rate PCM for processing. Most modern DAC chips run their volume control, channel mixing, and FIR filtering as PCM math. The DSD stream is decimated (often to something like 352.8 kHz / 32-bit PCM), processed, and then re-modulated back to DSD by the chip’s own delta-sigma modulator before hitting the analog stage.
2. Non-decimating processing in DSD domain. A few chips can apply Scarlet Book–compliant 50 kHz low-pass filtering and a non-decimating volume control while keeping the signal in 1-bit form. Cirrus Logic’s CS43198 is the textbook example (CS43198 datasheet): its “DirectStream Digital processor” specifically advertises “non-decimating volume control with soft ramp, and 50 kHz filtering as recommended by Scarlet Book.”
3. True bypass. The DSD bitstream goes from input pin to analog output without touching any digital stage. This is what DSD purists actually want, and it is much rarer than the spec sheets suggest.

The same physical DAC chip can offer all three modes, with the choice of mode controlled by I²C registers that the DAC manufacturer (not you) configures at boot.

Where the spec sheets stop being honest

Topping E70 Velvet: the one Topping had to admit

The Topping E70 Velvet is built around AKM’s flagship two-chip combo: the AK4191EQ digital processor/modulator and the AK4499EXEQ current-output analog stage. AKM specifically markets this pair as enabling a DSD bypass that keeps the 1-bit stream away from the PCM-style filter chain. Topping marketed a “Pure DAC mode” that strongly implied this bypass was being used.

It isn’t. From the Head-Fi review thread and Topping’s own subsequent clarification:

Many people were duped into believing that the E70 Velvet’s pure DAC mode bypasses the AK4191EQ (which is where digital volume control is processed) and sends DSD directly to the AK4499EX, but Topping confirmed that this specific DAC does not do this. The pure DAC mode simply disables digital volume control. The DSD path is not the ‘direct’ path; it is the ‘processor’ path that includes FIR filter, volume control and then re-modulation.

In other words: every DSD stream you feed an E70 Velvet — whether as DoP or Native — is decimated, filtered as PCM, has volume math applied, and is re-modulated back to DSD before hitting the analog stage. The “Pure DAC” toggle changes whether the volume math runs at unity or at your dial setting; it does not change the path.

Other AK4191/AK4499EX DACs (reportedly higher-cost models from manufacturers like SMSL and Cayin) do enable the bypass, so the chipset is capable — Topping just chose not to wire it up that way.

SACD players, basically all of them since ~2005

A November 2025 Headphonesty piece (source) summarized what SACD engineers have been quietly saying for years: most SACD players, including units in the four-figure price range, convert DSD to PCM internally the moment you enable any of the features customers actually want — bass management, channel trims, speaker distance, room EQ, multichannel mixing. All of those operations are dramatically easier in PCM, and the chips that perform them only run on PCM. The DSD-to-PCM conversion is so routine that there are commercial FPGA cores sold specifically for this job (Audiopraise).

The exceptions are players that explicitly market a “Pure Direct” mode and disable every form of DSP when you engage it. The early Sony reference units — SCD-1, SCD-777ES — are the canonical examples because they simply didn’t have a PCM path to fall back to. Modern units that take DSD straight to analog without any processing are a small and shrinking population.

ESS ES9038PRO: same chip, different OEM, different DSD

The ES9038PRO datasheet (Mouser) describes auto-detection between PCM and DSD inputs. What it doesn’t say loudly is that the chip’s internal volume control, de-emphasis, and DPLL all live on the PCM side — so any DSD source that you want to apply digital volume to gets routed through the PCM path, and the “DSD direct” registers only do what their name implies if the host MCU sets bypass on every adjacent stage. Different OEMs configure these registers differently. The same chip in a Gustard X18 and a Topping D90 may behave very differently for DSD.

AKM AK4499: the bypass exists, almost nobody wires it up

The AK4499 family does support a true DSD direct path — “with DSD as the source, the comparatively simple interpolators and modulators of the DAC chip are bypassed” — but this requires the OEM to configure the chip into bypass mode and to give up digital volume control while doing so (AK4499 datasheet, Euphonic Review). Most consumer-priced AK4499 implementations do not, because customers want a remote with a volume knob.

Chord DAVE, Hugo, Qutest: their own FPGA, their own rules

Chord doesn’t use an off-the-shelf DAC chip at all. The DAVE, Hugo, and Qutest run a Xilinx FPGA that implements Rob Watts’s pulse-array DAC topology (Stereophile). The FPGA hosts two separate signal paths — a heavily oversampled WTA filter for PCM and a non-decimating filter for DSD — and the two never share processing stages. This is the most compelling answer to the “is your DSD really DSD?” problem currently shipping at the high end, and it costs accordingly.

So which one matters

DoP solved a real driver-compatibility problem and solved it elegantly. It is not a workaround you should feel bad about using.

Native DSD is real on the wire and worth supporting where it works. It is not a guarantee that the bits stay 1-bit after they leave the USB endpoint.

The actually-interesting variable — and the one that nobody puts on a spec sheet — is what your DAC does with DSD after it arrives. The Topping E70 Velvet incident is valuable not because Topping was uniquely dishonest, but because it’s one of the few cases where the manufacturer was eventually forced to clarify on the record. Most DACs marketed as “native DSD” have never had to answer the question.

If you want a quick field test: turn your DAC’s digital volume off unity and listen for truncation. A true 1-bit direct path won’t produce it. Almost everything else will.