DSD support in Wax

Wax supports DSD (DSD64 only) by converting it to 24/96 HD PCM encoded to FLAC when you import it. Supporting DSD this way maximizes compatibility with industry standards. Some DACs accept DSD as an input, but all DACs accept PCM. Thus, conversion to PCM assures that users can connect any external DAC to the Wax Box. Moreover, the DAC in the Wax Box does not support native DSD, so the conversion also makes it possible to play DSD recordings without connecting an external device.

It is not possible to control the volume of a DSD recording in the digital domain. To support DSD recordings natively, the Wax Box would have to play them at full volume. When listeners switch from recordings in any other format to one in DSD, they would have to remember to turn down the downstream analog volume control before playing the DSD recording or they would risk blowing out their tweeters or their eardrums. They would have to restore the analog volume control when they finish playing the DSD recording so that other recordings would play at normal volume. Users would be responsible for determining before playing a recording whether or not it is DSD. Queueing recordings to play automatically would be impossible and the automatic volume equalization features would no longer work. Alternatively, we could require that users deactivate the volume control in the Wax Box if they include DSD recordings in their collection and rely exclusively on the downstream analog volume control, but most users appreciate being able to control the listening experience with one device. Internal volume control obviates the downstream analog volume control, so some users don't have one (they connect the Wax Box directly to a power amplifier). DSD complicates even mundane matters like volume control.

Almost all DACs that support native DSD use DAC chips that support both PCM and DSD. Such chips convert DSD to PCM so that they can use existing on-chip circuitry required to convert PCM (they do not incorporate a 1-bit DAC). Thus, DSD recordings are converted to PCM whether the conversion is performed by Wax or by the DAC. Because Wax performs the conversion ahead of time, there is no real-time constraint. If there is any sonic difference between conversion by Wax and so-called native support, the advantage probably lies with Wax because it has the time and resources to apply more sophisticated signal processing algorithms. Indeed, Berkeley Audio Design, manufacturer of some of the most highly regarded DACs in the world, advocates this architecture for exactly this reason.

3beez supports DSD for two reasons. Some Wax customers already owned DSD recordings. We want them to be able to listen to all of their recordings using our products. Second, if music lovers are interested in a recording that is not available in another format, we want them to be able to listen to it with Wax.

The trouble with DSD

DSD represents audio as one-bit samples with sample rate 2.8224 MHz. It is much more common to represent audio using PCM, which provides samples with many bits at a lower sample rate. For example, CDs represent audio as 16-bit samples at a sample rate of 44.1 kHz (64 times lower than the sample rate for DSD). PCM may be further encoded using a lossless codec such as FLAC or lossy codecs such as Ogg, AAC, or MP3 to reduce the bandwidth and data-storage requirements of the audio signal, but decoding such signals restores them to PCM. Thus, the one-bit representation of DSD is unique.

Uniqueness can be good. We have heard some very nice sounding DSD recordings. If the good sound quality is attributable to the digital representation, then DSD has a significant virtue. However, we have also heard PCM recordings that sound very good. Careful studies have failed to prove that DSD has a fundamental advantage in sound quality. Moreover, DSD has significant, known technical flaws and serious practical limitations. These disadvantages make DSD a strange choice for a distribution format.

Don't worry if you don't understand the more technical items in this list of disadvantages. The main point is that DSD is not the panacea it seems to be when you read about it in audiophile publications.
  • It is not possible to perform any signal processing on a DSD signal. We already mentioned that it is not possible to control the volume of a DSD signal during playback. The implications are worse in postproduction where common modifications include not only volume changes, but also mixing, equalization, dynamics, and reverberation. None of these operations are possible either.

    Any modification to digital audio is accomplished using mathematical operations. Performing any mathematical operation on a one-bit signal produces a result that cannot generally be represented in one bit. For example, the result of adding two one-bit numbers (the essence of mixing) requires two bits (1 + 1 = 2 in binary form is 1 + 1 = 10, which is two bits). Likewise, the result of multiplying two one-bit numbers (the essence of volume control) requires additional bits (1 x 0.5 = 0.5 in binary form is 1 x 0.1 = 0.1, which is two bits). It is possible to perform these operations on a PCM signal because that representation has many bits (16 or more).

    The impossibility of processing audio in DSD form requires that the audio be in a different form during postproduction with only the final result represented in DSD. One common solution is to record and process the signal in "DXD", a PCM representation employing 24-bit samples with a sample rate of 352.8 kHz (8x the sample rate of CDs). A second uses consoles that convert the DSD input to "DSD-wide", which is DSD with additional bits (a total of 8). The third is to convert the DSD signal back to analog and use an analog console for mixing and effects. All of these solutions have flaws. The high sample rate of DXD is antithetical to high fidelity. Faster may be better to a point, but the exceedingly high sample rate of DXD makes it hard to control noise and distortion. DSD-wide retains the noise-shaping strategy of DSD, so the noise level at frequencies above the audible range is dramatically higher. Mixing several such signals increases the noise level further and may exceed limits. The third solution requires a second conversion to DSD, with attendant problems that we will discuss next.

  • Converting an audio signal to DSD more than once is bad. DSD encoding pushes quantization noise to frequencies above the audible range. DSD-to-analog converters typically provide only moderate low-pass filtering (to minimize audible artifacts of the filter), so the restored analog signal contains unnaturally high levels of high-frequency energy. Such a signal can induce distortion when presented to any other analog circuitry, including a second analog-to-DSD converter. Also, the second conversion piles even more quantization noise on top of the quantization noise from the first.

    One scenario in which a second conversion is necessary is the third solution for postproduction mentioned above. At least one label records to DSD and then converts back to analog to use an analog console for all mixing and effects. The result of that processing must be converted back to DSD. Also note that the analog console must deal with analog signals containing an unnaturally high level of high-frequency energy. In a second scenario, engineers use an external device to accomplish a particular modification to the audio. A reverberator is an example of a common external device. Consoles that handle DXD or DSD-wide are not capable of performing every signal processing operation themselves. Moreover, engineers often favor a particular device because of its sound quality. External devices never use DSD, DSD-wide, or DXD, so the signal has to be converted to another format and then back to the first. The ideal for postproduction is to convert to a digital format once and then stick with it. In general, that solution is possible only with PCM.

    Note that DSD was invented for archiving existing analog recordings. In that application, there is no mixing or signal processing. The problems arise only when recording engineers try to apply DSD to recording, producing, and distributing, applications for which it was not originally intended.

  • There are problems with support for metadata. DSD has two file formats, DFF and DSF. DFF does not support any metadata useful to music lovers. (Note that this limitation of the DFF format does not affect Wax users because Wax stores metadata independently of the music files.) DSF supports ID3 tags, but they are affixed to the end of the file. Thus, it is not possible to stream DSF files as you will not receive any information about the music until it ends.
  • The data bandwidth of DSD is high – 5.65 Mb/s for stereo, which is 4x the data bandwidth of CDs. Thus, storage requirements are 4x greater and downloads take 4x longer. Double those multipliers for DSD128.
  • DSD wastes space. It is possible to compress DSD files using a lossless codec called "DST". DST reduces the size of DSD files by a factor of 2-3. It follows that 1/2 to 2/3 of a DSD file is wasted space. Nevertheless, commercial recordings do not use DST.
  • DSD suffers from limit cycles and idle tones (sometimes called "birdies"). These are both sounds that appear in the output unrelated to the input. Dithering can eliminate these problems, but it is not possible to dither DSD properly because of its 1-bit nature. The high-order filter (7-th order) used for noise shaping that is at the heart of DSD is the primary cause of these problems. PCM converters use lower-order filters and dithering to avoid them.


DSD is a strange choice for a distribution format. Your readings might lead you to believe that it is a recent innovation with signficant sonic advantages. It was actually invented two decades ago, and it was not originally meant for the applications to which it is being put today. As for sound quality, the good sound of some DSD recordings is more likely attributable to the expertise and care of recording engineers. Feel free to indulge your inclination to purchase a DSD recording if it satisfies you, but don't be seduced by the hype.