Optimize music files to reach your device’s sound quality limit, get compatible audio format or save disk space. Audio file converter software for music production and home hi-end Read more...
Probably, you already read articles, that explain DSD audio (Direct Stream Digital). But it maybe is not as simple to understand why 1-bit DSD may successfully compete with multi-bit PCM.
Read an easy explanation of what is DSD, how it works, why 1-bit may have hi-fi sound quality, what is better DSD or PCM, myth debunking and other essential issues.
Author: Yuri Korzunov,
Audiophile Inventory's developer with 25+ year experience in digital signal processing
March 2, 2023 updated | since December 25, 2017
Watch and share: What is DSD audio? Video explanation
DSD (Direct Stream Digital) is an audiophile high-resolution audio format. It is a representation method of acoustic waveform as digital data. The format is based on 1-bit sigma-delta modulation. Each sample of this signal type contains 1 bit per channel.
Initially, DSD format was designed as an alternative to PCM (Pulse Code Modulaion), which is used in CD-audio, WAV, FLAC, mp3 and other formats. Direct Stream Digital is distributed in SACD optical disk, .dsf, .dff and SACD ISO stereo and multichannel computer files.
DSD was originally intended for archiving of analog recordings. Now, this audio format is popular among music lovers and can bring master studio quality.
An audio signal (the electrical oscillations) may be presented as DSD or PCM.
It’s like we can assemble the same mosaic from small squares or circles.
Below we’ll return to the metaphor.
What is the difference between DSD and PCM?
Both DSD and PCM are a sequence of measured values (samples) of an audio signal's oscillation.
Sound/air oscillation each moment has a certain position (instant pressure level). The pressure level after the microphone is converted to an electrical oscillation.
The electrical oscillation at each moment has a certain position that refers to the air pressure level.
PCM consists of samples. The sample is a measurement (number value) of an instant position of the electrical oscillation (voltage) at the moment of the measurement.
1 sample = 1 instant measurement of the voltage.
The voltage level at the microphone input is measured over equal time.
In sigma-delta modulation (a.k.a. DSD) each 1-bit sample shows a positive [1] or negative [0] altering of the pressure level relative to its previous value.
And, as rule, DSD is explained as a saw (level grows or falls) between samples. But, it does not give an easy understanding of the basic DSD principle.
Below we'll consider the easier explanation. Keep reading.
Audio CD optical disk (PCM modulation / PCM format) was one of the first digital formats. It was a source with low distortion and noise level comparing analog ones. I think, now many people forgot what is "real analog noise".
From CD (PCM) to SACD (DSD)
After CD, demands to sound quality improvements could be implemented by increasing sample rate or bit depth.
Audio quality (sound quality) is a level of noise and distortions, that cause format/software/equipment. The level may be normalized by psychoacoustic criteria.
However, Sony and Phillips decided to use another kind of modulation - 1-bit sigma-delta modulation. At the first glance, using 1-bit is impossible, because such bit depth (resolution) causes a huge noise level. But, if high sample rates are used, noise energy may be pushed out of the audible frequency range (to ultrasound). That action is called "noise shaping".
Practically, 1-bit conception may be easier, than multi-bit, in a DAC implementation (read below).
When DSD music is converted to analog, ultrasound noise is filtered.
Medium, which contains 1-bit DSD audio stream was called SACD (Super Audio CD).
Multibit signal is like to a computer display. It consists of several pixels along both X and Y axisses.
XXXXX
XXXXX
XXXXX
DSD looks like 1-pixel screen.
X
And we can't understand: how 1-pixel display can show qualitative picture?
Imagine that, this one pixel is put on a truck. And the truck moves the pixel sequentially through all locations of multiple pixels of the first display. The movement is cyclic.
X . . . .
. . . . .
. . . . .
. X . . .
. . . . .
. . . . .
. . X . .
. . . . .
. . . . .
And it moves very-very quickly. And so rapid movement allows us to see high qualitative picture.
As rule, DSD has been explained in the time domain: positive samples increases level, and, negative sample decreases. After it, analog filter smooth "saw" form of signal.
In DSD analog-to-digital converter (ADC), measuring the "saw" signal is compared with the input analog signal.
DSD and PCM difference
At the ADC output "1" is present. And measuring-signal level grows until it exceeds the current level of the input analog signal.
It causes the output signal to switch to "0". And measuring signal begins to reduce level.
When the measuring signal level is a lesser input signal of the ADC, its output is switched to "1". And measuring signal grows until the exceeding the input signal.
Above we discussed DSD vs PCM like a mosaic from squares and circles.
These figures may be significantly smaller than the assembled image. If our eye can't resolve the figures, there is no difference between squares or circles we use.
So, qualitative DSD and PCM systems can bring us the same final result.
Such an explanation causes difficulties in understanding what is an actual difference between PCM and DSD and how to process DSD. Probably, it creates the myth about "native" DSD processing. About the processing read below in "DSD editing software" part.
Instead of the explanation, in the author's opinion, the easiest way to understand how DSD works is its spectrum consideration.
To estimate DSD and PCM sound quality, the quantization-noise level is the main criterion.
There is an explanation, that DSD has a super high sample rate and, thus, super low bit depth 1-bit does not matter for noise.
But DSD 64 sampling rate (2,8 MHz = 44.1 kHz * 64) is not enough to be better CD audio (see details).
To decrease noise level, noise shaping is used.
1-bit quantization-error (noise) energy has a significant level.
However, there is, so-called noise shaping of the spectrum is used. And quantization-error energy is re-distributed.
Noise shaping (NS) of 1-bit signal (spectrum)
Noise shaping in the simple words
Let's imagine noise energy as water in a pool.
If we blow strongly to the water surface at one poolside, water level of this side is decreased (noise is decreased in audible range).
It is the same adding more bits of audio resolution.
The water will be displaced to the other poolside (inaudible ultrasound range). And level there grow.
The quantization-noise spectrum (at the left of the picture) has a level comparable with a musical signal. The sigma-delta modulator pushes a significant part of the energy from the low- to high-frequency range, out of the audible band (0 ... 20 kHz).
Also, 48000-kHz base is possible. There are no technical limitations to the sample rate value. But hardware/software compatibility issues are very probable.
For the Nyquist theorem no difference between DSD and PCM:
the spectrum above [sample rate]/2 is
the same as the spectrum below [sample rate]/2
that flipped along the frequency axis.
Noise, Maximal Level and Overload Stability
When a sigma-delta modulator is designed, the engineers pay attention to two main parameters:
the noise level in the audible range and
tolerance to overload.
To solve these issues the engineers have:
bit-depth,
sample rate,
noise shaping.
These parameters should be considered comprehensively.
Bit depth
Bit depth reduces the quantization error level itself.
Noise shaping
NS "pushes" a significant part of the quantization error energy out of the audible range.
We can suggest that more energy may be pushed out of the range in the sigma-delta modulator. But it requires a steeper noise shaper.
Steeper NS can cause a higher probability of glitch (broken stability) of a sigma-delta modulator when input overload happens.
After the glitch, silence or some oscillations are generated at the output of the sigma-delta modulator.
After broken stability, the modulator should be forcibly reset.
Overload tolerance of sigma-delta modulator
Sample rate
We remember that quantization error is noise spectrum level.
A higher sample rate distributes the same noise energy into a wider band.
It reduces noise level or "quantization error" of DSD.
Also, a wider band allows using sloper noise shaping.
Energy is a square of the figure, concluded between the spectrum line and horizontal axis into band 0 ... [sample rate]/2.
Sample rate and quantization-error spectrum level
In the left and right pictures, the squares of the noise energy figures are the same. But the figure, which is more expanded in the horizontal axis, gives a lesser noise level.
A higher sample rate allows for reducing the noise level in the audible frequency range. It allows for reducing noise shaping's steepness, which increases the overload tolerance of the modulator.
Resume
We can see that lower noise and higher stability to overload can be achieved in different ways.
As an example, better quality is noise shaping implementation matter for the same bit depth and sample rate. But, on the other hand, we can increase the sample rate and/or bit depth to decrease the error level for the same NS method. It allows for the improvement of the audio quality.
DSD in figures
Pro audio modulators have noise levels in the audible frequency range for sample rates (read details):
D64 about -125 ... -145 dB (comparable with PCM 24-bit)
D128 about -165 dB (better than pulse code modulation 24-bit)
D256 and higher about -170 ... -200 dB (comparable with PCM 32-bit)
Noise level in the audible frequency band almost doesn't depend on demodulator. But the noise should be maximally suppressed out of the band. Ultrasound noise can cause audible intermodulation distortions.
Bit depth defines audible noise. More bit depth is lesser noise.
In the picture, we can see that bit-depth decreasing is achieved by using only part of the full signal frequency band. In other words, if we want to reduce bit depth, we must increase the sample rate to keep the audible noise level.
Read details of the format comparison in the table: DSD vs WAV vs PCM vs FLAC vs DXD
Direct Stream Digital (sigma-delta modulation) is like to pulse code modulation, but quantization error spectrum is shaped for decreasing noise into the audible range.
We can apply NS for usual PCM. But the difference here is band reserve for pushed noise out of audible frequency range.
DSD versus PCM
PCM has a lesser band reserve (above audible range) than sigma-delta modulation format, but pulse code modulation has higher bit depth.
Noise shaping for pulse-code modulation is an implementation matter too, like sigma-delta modulation.
Therefore, no format advantages as itself. But implementation makes a difference.
Sigma-delta-modulation decoder (demodulator) is 2-position (1 / -1) voltage generator and low-frequency filter. It is simpler than pulse-code-modulation hardware demodulator. Because pulse-code demodulator contains multi-voltage matrix or 1-bit decoder(s). So we have more abilities to make cheaper 1-bit DAC better than multi-bit one.
There is no univocal answer. There are several reasons:
Quality of a record
Mixing/re-mixing of the record (including quality of the processing)
Playback software/hardware implementation
DAC implementation.
These reasons trigger infinite discussions about "what is better PCM or DSD".
Technically, DSD and PCM are incomparable. Even as single DSD/PCM compatible DAC is used for the test, the difference may be caused by different mixing/post-production of records, and the DAC uses different circuits (see here).
General answer: "Each case is individual and depends on playback system implementation and listened to record".
DSD sound quality and other [brief F.A.Q.]
Are DSD downloads worth it? Are DSD files better?
There is no unambiguous answer to the question. It depends on record quality, your software and equipment.
Is DSD better than FLAC?
There is no answer. Theoretically, DSD has better potential than FLAC and other PCM formats. Recording, certain equipment makes sound quality. Read details...
Both kinds of DAC contains sigma-delta modulator. It converts the analog voltage to digital samples in a format like Direct Stream Digital.
Before modulator analog filter ["DSD" sample rate]/2 is applied. Looks details here >
PCM ADC is more sophisticated and contains multibit modulator.
To reduce sample rate, "multibit-DSD" signal should be filtered (low-frequency filter) and decimated (removing part of samples). The filter also eliminates shaped noise of "DSD signal".
To easier providing of PCM DAC precision/linearity into this kind of DAC, a digital sigma-delta modulator is used. After it, sigma-delta de-modulator converts digital "DSD" to an analog signal.
In the general case, DSD DAC is simpler. It contains sigma-delta de-modulator only.
It is technically impossible to compare PCM and DSD DACs, because they use different processing / electrical circuits.
DSD editing is cutting, merging, gain altering, equalizing, etc. of file(s).
"Native editing" (without conversion to multibit and DSD re-modulation) is possible for cutting and merging DSD records only.
All other kinds of editing demand conversion to multi-bit values and re-modulation. In this case, losses of the editing are comparable with PCM resampling.
1-bit audio files (DSF, DFF, SACD ISO) and disks may be either uncompressed or compressed with DST method.
DoP is an open protocol that allows packing 1-bit audio as multibit for compatibility with software and hardware [4]. DoP cannot be played as usual pulse-code modulation.
Read details on how to open, play DSD on Mac, Windows, iOS, Android, Linux: player list >
Read how to download DSD music and how to get free DSD files here >
DSD audio players (hardware and software) can playback all or some of the 1-bit file formats.
1-bit files may be played back directly on DSD DAC/player or converted to PCM "on fly" to playback to PCM DAC/player. About SACD conversion read more here
1-bit playback may be performed via special ASIO-driver under Windows, including DoP (DSD over PCM) digital audio packing format (example).
SACD optical disks may be played back at hardware player. The author knows nothing about available SACD-drives for consumer computers to playback/record SACD optical disks.
A stereo player may downmix multichannel to stereo. As alternative multichannel files may be pre-converted to stereo. It allows for saving space on the hard disk of an audio player. Downmix is lossy audio processing. Its quality is defined by an implementation.
Convert DSF/DFF/SACD ISO to *m4a (ALAC, Apple Lossless). Read also (conversion DSF/DFF/SACD ISO to ALAC(m4a) is same) DSF to FLAC, DFF to FLAC, ISO to FLAC.
Conversion DSD to PCM degrade sound?
Conversion with processing audio (including DSD to PCM) adds distortions. But final distortions is a matter of used playback distortions. Read details here
Conversion DSF to DFF and back is lossless?
It may be lossless when audio converter doesn't process audio stuff (resampling and other) in conversion.
Do various DSD converters have different sound quality?
Yes. There is noise level, frequency/phase distortions and tolerance to overload are estimated. Read details here.
DSD editing is sophisticated issue due modulation noise. Non-linear processing can cause audible intermodulation distortions by ultrasound noise.
Currently no information, that is known for the author, about "native" 1-bit processing (example: gain altering, resampling, etc.) without conversion 1-bit to multibit and back. Except, merging/dividing the audio file.
PCM here may be considered as "multibit DSD". Pulse code modulation is not obligatory to mean "24 bit / 352 kHz" or so on. Author recommends use 32- or 64-bit float point bit depth. This PCM contains high frequency modulation noise. But for conversion, this "multibit DSD" to 1-bit need re-modulation with NS.
Losses of editing with 1-bit/multibit conversion are comparable with resampling.
Recording studios may distribute 1-bit records without editing.
DXD is PCM (as rule "24 bit / 352 kHz" or so on) with high sample rate and bit depth and legacy DSD noise. However, the noise can cause audible intermodulation distortions. Before non-linear processing, the high frequency noise cutting is recommended.
Watch and share: Experiment with cutting ultrasound noise
Is DXD better than DSD?
DXD is PCM with noise hump in high frequencies. DSD DAC is simpler than PCM one in theory. From DAC point of view, DSD is ideal format. But, actual result is refer to digital-to-analog converter model. Read details here...
There are compatibility and other issues of tranfer 1-bit and multibit audio data from computer to PCM or DSD DAC.
All audio interfaces (USB, in instance) are designed for multibit data. And 1-bit data packing into ordinary multibit value was suggested.
DoP (DSD over PCM) is a format providing compatibility with some DSD hardware (USB DAC, for example). The format allows packing 1-bit DSD data in usual multi-bit words (like PCM).
These multi-bit words are transferred as usual PCM. Upper bits of the multi-bit word contain special "DSD over PCM" marker code.
DoP receiver should recognize the marker code and interpret lower part of the multi-bit word as raw DSD bits.
DoP format may be stored into usual lossless PCM file: WAV, FLAC, AIFF.
DoP coding may be chosen in audio player settings.
Both DSD and PCM have advantages and disadvantages. Its discussing is theoretical rather. Read details here...
Is DSD the best audio format?
No. DSD is alternative audio format to PCM one. Read more information here...
What is the highest quality audio file?
DSD and PCM lossless files are the highest-quality audio files. Read details here...
Is DSD better than 24 bit?
DSD and PCM bit depth may be compared by quantization noise level in the audible band. However, equipment and software play big role there. See approximate comparison DSD and PCM bit depth...
Is CD better than FLAC?
FLAC format supports high-resolution audio. It give technical abilities to better sound quality.
![What is DSD [Quick explanation]](/sites/default/files/u1/what-is-dsd-audio-explanation.png)
