Every audio format has a noise floor. Analog tape has hiss. PCM has a flat dither floor: around -96 dB for 16-bit, and around -120 dB in practice for 24-bit, limited by the converter's analog stage. DSD has one too, but shaped rather than flat. That difference in shape explains most of what listeners measure, hear, and occasionally worry about.

DSD is a 1-bit format, and representing music in a single bit produces a large amount of quantization noise. That noise is not added by any process; it is the unavoidable byproduct of the modulation itself. At zero signal, the raw bitstream is nothing more than an alternating pattern of 1s and 0s. What makes DSD work is where the noise ends up: the modulator's design places it above the audio band, leaving an in-band floor below -120 dB, quieter than 20-bit PCM, with the noise energy rising steeply in the ultrasonic range. The literature calls this noise shaping, and you'll meet the term in every DSD tool and reference. For DSD64, the floor sits around -120 dB at 20 kHz and rises to roughly -60 dB by 50 kHz.

This shape is a property of the format, not a defect in any particular file. If you point an analyser at any DSD file, you will see it. A rise beginning just above 20 kHz is what a healthy DSD64 modulation looks like.

Each doubling of the rate shifts the noise envelope one octave higher. Where DSD64 reaches -60 dB around 50 kHz, DSD128 reaches it around 100 kHz, and DSD256 around 200 kHz. The total unfiltered noise energy stays the same at every rate. What a higher rate buys is distance: each doubling adds an octave of clean spectrum between the music and the noise, which is precisely the room a DAC's reconstruction filter needs. With more of it, the filter can be gentler, and the noise is easier to remove in the analog stages after conversion.

Many albums at NativeDSD are offered at rates above the one they were recorded or edited in. Those editions are conversions, not remasters.

Every edition we sell is produced directly from the master the label supplies, and never from another edition we made. Each rate is a separate pass from that same master, so no edition carries another of ours stacked on top of its own.

What varies is the master itself. When a label supplies its DXD edited master, every DSD edition made from it is a first-generation modulation: the only ultrasonic noise in the file is the new modulator's own, positioned where that rate places it. When a label supplies a DSD master instead, much of its noise profile carries through into the higher-rate editions. The conversion's filtering attenuates part of the source's ultrasonic content, but we don't apply corrective filtering aimed at removing it, so the source's signature remains measurable in the result.

The practical consequence: a DSD256 made from a DXD master and a DSD256 converted from a DSD64 master are the same rate with measurably different ultrasonic signatures.

Filtering is not free. Every low-pass filter that removes ultrasonic content also disturbs phase behaviour inside the audio band, and the gentler the filter, the less it disturbs. So our conversions filter as little as the process requires rather than scrubbing the source clean, because aggressive cleanup would cost something audible in exchange for removing something that mostly isn't. Beyond that, filtering is a mastering decision, and we treat it as belonging to the label who made the record rather than to us. It's why we ask labels for the edited master in the first place.

The filter designs inside professional conversion tools are proprietary to their makers, which is why we describe what our process does rather than publish specifications we don't have.

Within the audio band, no. The noise there sits below -120 dB, is uncorrelated with the music, and does not interact with it. It also does not fold back into the audio band: in a linear system, ultrasonic energy stays ultrasonic, and your ears low-pass it in any case.

Above the band, it depends on your equipment. Most modern DACs and amplifiers handle ultrasonic content without complaint. Some systems are sensitive to it: amplifiers with limited stability margins, and tweeters with low power handling driven at high listening levels, can react to energy they were not designed for. If your system is in that category, the practical options are choosing a rate whose noise onset sits beyond your equipment's sensitivity, choosing the DXD edition where one exists, or applying a low-pass filter in your playback software. This is system matching, not a fault in the file or in your equipment.

DSD trades a flat noise floor for a shaped one, and the trade buys an audio band quieter than the formats it replaced. Higher rates move the noise further away but do not remove it, and our conversions preserve the label's master as delivered, ultrasonic behaviour included. The rate that sounds best is the one that fits your DAC and your system.

For more on how higher-rate editions are produced, see DSD 512: how it's made, and what it does.