First, let’s gather some background information. ‘Direct Stream Digital’, invented by Philips and Sony was launched as part of the introduction of the Super Audio CD (SACD) in 1999. The SACD is a conventional CD with regard to sound quality, space and multi-channel capability and was originally designed as the CD’s successor for the broad market. During the encoding for the CD as the existing PCM method was performed, the underlying storage principle of the SACD of the so-called is based on the ‘Delta-sigma modulation’. DSM could never really assert itself as a true opposition to PCM, instead it’s usually just dealt with through discourse on DSD, although we look at the storage methods based on the “Delta-sigma modulation” a bit closer. Ultimately the DSD and SACD only occupy a small niche of the hi-fi and recording market. Nevertheless, the technology is in space and is always drawn as an alternate storage format to PCM into consideration. So it is not surprising that the current trend towards high-resolution music files for the end user calls the proponents of DSD encoding back to the plan.
Sampling and Storing
Although there are two different procedures and formats for digital storage of music information: PCM and DSD, there is one – or really two – common denominators. We can discuss the difference between DSD to PCM method at length as well as in a nutshell. As before, our goal is to analyze an analog signal that varies in frequency spectrum and amplitude over time whilst storing a digital signals. For this purpose we defined in the previous item to PCM method a steady beat and a raster values with which we were able to express changes in the amplitude of our signal at each sampling time in a certain number of possible states. Even with the DSD encoding is all about these two terms. Actually, the same thing happens almost exactly, but only almost.
This time let’s choose an extremely high sampling rate of 2.8224 MHz, of which 2.8224.000 from the signal are analysised per second. Remember that with our PCM-based digitization in “CD quality”, there were only 44,100 samples per second, which corresponds to a mere 0.0441 MHz. This is not just small difference, so let’s look at it some the bit depth. In contrast to the PCM method in which changes of the signal states possible in 65,536 (16-bit) and in 16,777,216 possible states (24 bits) have been shown, it is to DSD suddenly only two states (one bit), which is why is often referred to the “1-bit technology”. While you might not even would skeptical at a high sampling rate, you must be with our prior knowledge and in view of the low bit-depth inevitably ask: How are two conditions are sufficient to reproduce the complex dynamic course of a music signal?
The key to this question lies in the task, which belongs to one bit in this case. Thus, the amplitude curve is not shown as as the PCM method approximately in the possible states, whose number varies according to the selected word length, but by the frequency with which the one available to us bit changes state. In this case, these frequency changes according to the following rule: A high amplitude corresponds to a few state changes, while a low amplitude frequent state changes result. For the two extreme cases of a permanently maximum or a durable non-existent the amplitude accordingly no or constant change would mean our two states in the timing of our sampling rate. On the basis of this current state changes, which includes even the smallest changes in the signal due to the high clock rate can thus also a digital copy of our original signal to create and save.
Consequences – Dynamics, Mistakes, Filters.
All aspects of the next section are touched on in the article for PCM encoding. If you’ve informed yourself from that, the following statements on DSD should be understood relatively easily. On the input side, we need no aliasing filter, since the sampling frequency is so high that even after halving (Nyquist-Shannon sampling theorem) no audio signal will exceed the upper limit frequency of 1.4112 MHz. A possible influence of the signal through a filter is thus ruled out at this point, this is definitely an advantage. When reconverting the story is different, here’s a filter is necessary to remove high-frequency signal components even when DSD data. We shall come back to it, but contact us before further properties of our signal to. There is for example the standard one for DSD and – compared to a PCM signal with 16 or 24 bit – very high noise level. Largely due to quantization error resulting from the low bit-depth, this looks natural negative effect on the signal-to-noise ratio, thus the dynamic range of our signal. The remedy here is noise-shaping algorithms that banish the problem of human hearing threshold out into the ultrasonic range. This method is already known to us and, in the case of DSD essential. Only the result is a dynamic range of about 120 dB over a range of 20 Hz to 20 KHz. Above the noise level then rises rapidly and the dynamic range for frequencies in the range of psychoacoustics, which theoretically can be present in the digital signal due to the high sampling rate well, is significantly less than 120 dB. At 35 kHz, the dynamic range corresponds still about 96 dB, is on CD-level. Frequencies it would have been very prominent sound to still her – mind psychoacoustic – to make influence. The like to run into the field displayable limit of DSD of impressive-sounding 100 KHz we forget in view of this fact simply times.
PCM vs. DSD
Once we have compiled the most important information and facts about the DSD method, we can now venture into comparison. Let’s start with the physical disks, a CD (PCM, 44.1 KHz, 16 bit) and a SACD (DSD, 2.8224 MHz, 1 bit). Without much more consideration we find that the SACD is superior to the frequency representation both in terms of dynamic range, and with respect.
35 KHz and more // DSD
22,05 KHz // PCM
120 dB (above 20KHz lower, see above) // DSD
96 dB // PCM
Let’s switch now a different realm other than that of physics: let’s look at a DSD file compared to a PCM file with 88.2 kHz 24 bit. Now things start to look a bit different.
35 KHz and more // DSD
44,1 KHz // PCM
120 dB (above 20KHz lower, see above) // DSD
144 dB // PCM
Of course, it is also possible with DSD to double the sampling frequency or even quadruple it. The problematic quantization noise can then be moved to an even higher frequency ranges and benefit dynamic range and frequency representation. This process can, in principle, continue with both formats to the point of absurdity, because the technical options make that possible. But instead we want to seek another more preferable example to remind ourselves that we could someday hear best frequencies up to 20 kHz in our lives. And even the vastness of psychoacoustics come to an abrupt end when the highest quality microphones encounter while recording no later than 30-40 KHz to their limits. Physically understandable give high resolutions so only up to a certain extent, in particular during the production of sense. Not be avoided to the extent possible and resulting errors as possible – your benefit to be stressed again here is, to a considerable extent that the digitization process inherent problems – be they aliasing effects, influences through a filter or quantization the musical frequency range of influence, but can be shifted to higher frequency ranges in doubt. This applies to DSD and PCM alike and so the aforementioned output-side filter in both cases is only logical, since the high-frequency noise components are undesirable in the signal at the gain or when imaging through the speakers.
In addition to the comparison of raw numbers, the eventually the format that wins is just doubled in the resolution, there is another aspect that we do not want to ignore. This refers in this case is not about subjective listening tests, here anyone interested must form his own opinion as I said. Rather, it is about the technical reality and since DSD occupied just actually a niche. Although there are converters which operate exactly as the DSD-standard and consistently convert to the appropriate format, but these are few and far between. Accordingly, “pure” DSD is extremely rare from the A / D to D / A conversion. It would be conceivable an old master tape of a fully analog recording, the digitized with such a converter, stored and then is converted and reproduced according to head back. In general, however, one can assume that at some time between recording, production and reproduction of one or more conversions into PCM format have taken place, even if in the end there is an SACD or DSD file. How is that important whether a final DSD conversion makes the difference or whether at the end maybe not the format, but the choice of components and their technical finesse really matters, will probably continue to eagerly discussed.