DSD Audio – An Introduction
Image Credit: Sony
Let’s first take a quick look at the most well established and widely used of all audio formats, Pulse Code Modulation (PCM).
Pulse-Code Modulation (PCM)
An analog signal may be represented digitally by sampling it at regular intervals and converting that samples voltage level into the nearest corresponding digital value, a process known as quantization. Frequently there isn’t a digital value that exactly matches the analog value, this rounding error is referred to as a quantization error and causes noise.
The digital values selected are referred to as words and are comprised of a series of binary bits. As an individual digital bit only has a binary value of 0(off) or 1(on) these words are therefore comprised of a series of 0’s and 1’s, E.G. 11011001. The number of 0’s and 1’s, or bit depth, represents the maximum number of discrete analog levels that the word can represent and is calculated using the formula 2n where (n) is the number of word bits as listed below:
- 1 bit – 2 analog levels
- 2 bits – 4 analog levels
- 4 bits – 16 analog levels
- 8 bits – 256 analog levels
- 16 bits- 65,536 analog levels
- 24 bits- 16,777,216 analog levels
- 32 bits- 4,294,967,296 analog levels
Typically these digital words represent the same step change for each analog level, but non-linear step sizes can be found in some PCM systems. For most common, high-fidelity digital audio media, like CD’s, DVD’s etc, the PCM system is referred to as Linear PCM (LPCM) due to the same step size per digital level. PCM is a lossless digital encoding system and is defined in detail in the Compact disc Red Book standard introduced in 1982.
LPCM Sampling
Bit depth together with sample rate and other technical parameters like jitter have profound effects on the digitized signals fidelity. Sample rate is directly related to the highest frequency that can be converted to a digital signal and impacts the required filter designs. For CD’s the sample rate is 44,100 times per second but other higher rates of; 48kHz, 88.2 kHz, 96 kHz, 176.4 kHz, and 192 kHz are also quite common in high quality audio digital systems. 44.1kHz is the lowest sample rate that can ensure that the full frequency range of human hearing (20Hz to 20kHz) is covered.
NOTE: PCM can only accurately capture and reconstruct frequencies up to half the sample rate (Nyquist rate/frequency) or interfering signals are created from a process called Aliasing. Steep cut-off filters are used at both the input and output of a PCM chain in order to minimize this issue.
PCM Digital word creation (4 bit)
The resulting analog sample digital words are then used to create an organized digital bit stream through a process called Pulse-Code Modulation (PCM).
Direct Stream Digital (DSD)
DSD is the trademark used by Sony and Philips for their systems that digitally encode audio signals for the Super Audio CD (SACD) and was originally used in the 1990’s as a way of archiving old analogue recordings.
DSD works in a very different manner to PCM, using only a single digital bit to represent a CHANGE in the analog signals level, so it only has two values 1 or 0. A 1 representing and increase in analog signal level, a 0 representing a decrease in analog signal level, both being relative to the previous samples level. The DENSITY of the 1 and 0 data is then encoded into the bit stream.
Bit Creation (Concept)
At first sight a digital word with only two levels may seem very limiting, but DSD has an extremely high sample speed from 2.8 MHz all the way up to 22.5 MHz. This huge increase in sample speed together with the sampling technique ‘make up’ for the lower bit depth. The sample speeds are commonly displayed as:
- DSD64 – 2.8224 MHz (a sample rate 64 times that of a CD)
- DSD128 – 5.6448MHz
- DSD256 – 11.2896MHz
- DSD512 – 22.5792 MHz
DSD is a 1 bit variant of Pulse Density Modulation (PDM). It is the process of controlling one of the bit clock frequencies shown above with the analog signal, such that the resulting signal is a continuous stream of 1s and 0s of varying DENSITY. The density of the bits is proportional to the original analog signals amplitudes. Typically a large cluster of 1s correspond to a high (positive) amplitude value, a large cluster of 0s would correspond to a low (negative) amplitude value, and alternating 1s and 0s would correspond to a low or zero amplitude value. In reality it is more complex than this but the end result is a continuously varying bitstream of 1 and 0 bits as shown below.
A Typical DSD Bit Stream
A DSD stream inherently has significant inaccuracy in its quantization of the audio signal to a single bit stream and various electronic processes are employed to reduce noise and distortions produced by the technique. Fortunately the very high sampling rate makes the processing techniques quite straightforward.
There are two competing DSD file formats. The formats differ in compression and metadata options. DSD Interchange File Format (DSDIFF), developed by Philips, uses a .dff file suffix. It supports compression of the DSD payload but does not support storing of metadata such as artist names or song titles. DSD Stream File (DSF), developed by Sony, uses a .dsf file suffix. It has support for storing meta data but does not support compression of the DSD payload.
Both the PCM and DSD formats support a dynamic range that exceeds that required for music (20-20KHz) with DSD64 supporting greater than 120dB (rising to >150dB at lower frequencies) and 24bit/192KHz PCM having a theoretical maximum of around 144dB. In comparison CD’s at 16bit/44.1KHz support about 96dB dynamic range which is generally considered to be sufficient for most music types.
DSD/LPCM Dynamic Range/Bandwidth Comparison (Approximate)
DSD signals do not require sophisticated “brick-wall” filters like PCM, using digital noise shaping filters and simpler lowpass designs. Also every time you double the DSD sample rate the bandwidth doubles, pushing the noise generated by the DSD process (shown above) further out of the audible range, thereby requiring very little filtering. Audio bandwidths of 50-100KHz are easy to create with DSD.
Many claim that DSD is as close to analog audio as digital gets. There are and will continue to be hot debates as to whether a DSD stream is more “audiophile” than a PCM stream once both are converted back to the analog domain, with DSD64 being claimed to be roughly “equivalent” to 24bit/88.2KHz PCM.
I am in the DSD camp regarding its audiophile performance. While great as a storage/archiving format and despite the less expensive DAC’s and simple filters it has the disadvantage of being very difficult to handle for editing and post production, which probably accounts for its’ lack of popularity.
With Mobile Fidelity now using DSD256 for master capture any noise and distortions are pushed well out of the audio 20-20KHz bandwidth and filtering demands will be light.
It’s such a pity that you cannot purchase the DSD master files…..Mobile Fidelity??
Technical Footnote:
It must be realized that ALL digital formats are subject to various artifacts which include Jitter, Quantization Errors, Aliasing and intersample peak clipping. The first three issues can generally be easily controlled in a state-of-the-art DAC system to keep the resulting noise and distortions between 110-140dB below the music peak levels, becoming totally inaudible. Unfortunately peak clipping can produce distortion levels that are only 30-40dB below the music peak levels and can be clearly heard.
These intersample peaks occur when two adjacent samples that are at 0dB full scale (dBFS) cause the DACs analog output to try to go above 0dBFS. If the peak digital levels are restricted to approximately -3dBFS this issue is unlikely to occur.
It has been said that when todays comparable state-of-the-art DAC systems do not introduce intersample peak clipping then they are often indistinguishable from each other AND the original analog signal.
For more information on audio formats check out my Introduction to Audio Formats on CD, SACD, DVD, DVDA and Blu-Ray.
Also check out my post on Hi-Res Audio Formats.
Hi,
I am wondering: do you think that the sound quality of DSD256 will be 4 times better than the usual SACD disk, or even better at all?
Hi Dori,
No, DSD256 is not four times the quality of DSD64. DSD256 certainly is better, but it’s really about the original master quality and the alignment/capture process. DSD256 has a BW of about 80KHz whereas DSD64 only has about 20KHz bandwidth. The increased bandwidth not only allows for less impact from filters but provides more ‘breathing’ room for the middle and top end and potentially lowers distortion levels. As for dynamic range there is no real difference, that is just a function of the master source material.
I can hear the difference between DSD64 and DSD128 so the improvement from using DSD256 or even DSD512 should be audible IF the analog master source material is up to it and it is well mastered with high quality encoding hardware during the transfer process.
No matter how good DSD or PCM are, they are NEVER as good as the original analog source masters and a well encoded DSD64 can still sound very good with good master material and technical expertise.
UPDATE: DSD can sound ‘better’ than analog tape masters IF the mix is going directly to a DSD encoding system. Unless of course you want that ‘tape’ sound.
Well that’s my 2 cents!
Paul
Thanks. And what about albums like ‘Brothers In Arms’ or ‘The Nightfly’? Do you think DSD256 can sound as good as the digital master source?
Hi Dori,
I have both the original MoFi vinyl Brothers In Arms and the SACD. See my review here: https://fromvinyltoplastic.com/dire-straits-brothers-in-arms-45rpm-vinyl-review/
Unfortunately I only have an original release pressing of The Nightfly, but it sounds great for a mass produced album, see my review here: https://fromvinyltoplastic.com/donald-fagen-the-nightfly-yesteryear-vinyl/
In short my answer would be guarded yes.
The issue here is that at these audiophile levels its more about what is actually on these old masters and what can be retrieved, as some are in poor condition. True to say that digital doesn’t degrade in the same way that analog tapes do, so over time digital might “sound” better. No digital process can really recapture lost analog information. On top of that it depends what PCM standard was used to create the PCM masters.
Given the same material, if DSD256 and 24bit192KHz PCM were the original master formats for mastering vinyl, I very much doubt you would hear any differences that would translate to being audible in the home. Just remember all the distortions created in the vinyl cutting and replay chain, let alone DAC digital filter issues.
Just think of all those other areas that have a big impact on your home sound like; room acoustics, speakers, amplifiers, vinyl playback components and CD DAC’s, or even cables and power. I would be far more worried about those.
Regards
Paul
I read something by one guy that said that DACs are of 2 different types and each of these types converts DSD or PCM better than the other so it depends on the type of DAC you are using as to which format will sound truest. If you like DSD, then use a DAC that is the type that converts DSD better than other DACs.
Best,
John
Hi John,
Yes you are correct in that some DAC’s can convert DSD streams directly while others need it to be converted to PCM first. Handling DSD as a native stream is clearly the better way of decoding it, rather than having to convert it to PCM first. ESS produces a state of the art range of audiophile SABRE DAC’s that can handle DSD natively or as PCM. Check out the following link for more details on their range of audiophile DAC’s.
https://www.esstech.com/products-overview/digital-to-analog-converters/sabre-audiophile-dacs/
Thank you for popping by and for your comment.
Regards
Paul