December 2008


Music Playback From A Computer – The Audiophile's Concerns
Article Submitted By Daniel Weiss Of Weiss Engineering Ltd.

  With the rising popularity of computers as a music source, more and more audiophiles are asking themselves whether a computer can be a decent playback system, similar to e.g. a CD Transport. To answer this question we have to identify the potential sources of disturbances. In digital audio there are two main factors responsible for the quality of a playback system:

1.)  The  proper transmission of the bits, i.e. a 0 must stay a 0 and a 1 must stay a 1.

2.)  The proper timing during sampling of the analog signal (A/D Conversion) and during D/A conversion.

Number 1 can be achieved relatively simple. The CD uses a powerful error correction mechanism which can correct, i.e. do a perfect bit reconstruction of error bursts of about 4000 bits. 4000 bits corresponds to a scratch on the CD of about one tenth of an inch (2.5mm) length. In addition, using the redundancy in music signals, it can conceal error bursts of up to 12300 bits or a scratch length of three tenths of an inch (7.5mm). "Conceal" is the term meaning interpolating the music signal out of error free data. Of course this is not bit perfect anymore, it is estimation.

Hard disks used in computers are even better in terms of data recovery, as a single bit which is wrong can render a file on a hard disk useless. So one can safely assume that it is easily possible to get the bits unchanged from the data carrier to the final destination, the D/A Converter.

Number 2 is more difficult. The "proper timing" means that the samples which are taken of the analog signal during the A/D conversion process, ideally are taken at absolute equidistant times, i.e. in the case of the CD at exact multiplies of 1/44100^th of a second. Deviations from those ideal sampling points are called jitter. In real world systems this jitter can not be zero, as it is an analog process which generates the sampling clock and thus is subject to noise, which results in sample timing jitter. As music consumers we trust that during the A/D Conversion care has been taken to minimize the sampling jitter, as we can’t (easily) do anything about it once the signal has been digitized. But during playback, we can take care that the D/A Converter we are using has low intrinsic jitter and in addition is not sensitive to jitter of the sampling clock fed from the source, e.g. the CD Transport or Computer.

Consequently the essence is that the D/A Converter used should be (virtually) immune to jitter. If this is the case, then the computer can be a playback system as decent as a CD Transport. Potentially it is even better than a CD Transport, as it is capable to play back high resolution files with e.g. 24 bit wordlength and a 192 kHz sampling rate!

Jitter usually is suppressed with a so called PLL (Phase Locked Loop) circuit which can be viewed as an electrical equivalent to a flywheel. The PLL is fed the sampling clock (e.g. 44.1 kHz) and generates the same clock "phase and frequency locked" to the incoming clock. I.e. it follows the incoming clock on a long term basis, but short term fluctuations (jitter) are suppressed. This is similar to a flywheel which does not react much to changes in the driving force. Our D/A converters (Medea, Minerva) use a scheme with two cascaded PLLs for even larger jitter reduction.

Alternatively the D/A converter can be made the master clock for the whole system. That means that the audio source is slaved to the D/A converter. This is not easily achieved with a CD transport as it has to have an appropriate clock input. With computer playback via FireWire on the other hand, such a scheme can be easily implemented. Our Minerva FireWire D/A converter for instance can be the clock source and slave the computer.

One thing which has to be kept in mind though is the fact that the computer can easily change the audio data during playback. With today’s player programs (e.g. iTunes, Windows Media Player, Foobar2000, MediaMonkey, etc.) one has to be careful that no unwanted signal processing is applied to the signal. Potential harm can come from:

-   A level control in the digital domain, if it is one without dithering applied

-   A sampling rate conversion going on in the background because the sampling rate the D/A Converter is running at does not correspond to the sampling rate of  the file played. This rate conversion can go "unnoticed" as the operating system engages the rate conversion as required.

-   Any other processes like equalization or other sound effects.

The player programs currently available are not yet very user friendly when it comes to issues of concern to audiophiles. One has to know how to make sure that e.g. the file is not sampling rate converted. Choosing the right player and configuring it properly thus is a topic for itself. Currently preferred players are MediaMonkey or Foobar2000 for Windows plus iTunes and the soon to be released Amarra for OSX. If the player is capable of feeding the audio files unaltered to the D/A Converter it is called to be "bit transparent", a term used in pro audio circles.

ASIO drivers make it simpler to have bit transparent transfers, i.e. one therefore would prefer D/A converters which can work with ASIO drivers. Both Foobar and MediaMonkey support ASIO drivers.

A nice thing would be to have a means of determining whether the player software is bit transparent. One way to check for this is with a D/A converter which supports HDCD decoding. If the HDCD LED comes on upon playing a HDCD encoded file, the player is bit transparent for sure. For our Firewire devices we plan to implement a software which lets the user test whether the player is doing bit transparent playback.

Another question is whether the digital level control in those player programs can be used for a "high-end" level control and thus a preamplifier down the chain could be omitted. The answer is "it depends". With today’s 24 bit D/A converters there is a huge dynamic range available which gives room for a digital level control. Usually level controls are not required for huge level changes, but mostly to adjust the volume between different tracks. I.e. a range of e.g. 0 to 24 dB of attenuation is sufficient in most cases. A 24 dB attenuation means that the bits are shifted four places to the right (6dB per bit). E.g. a 16 bit source which occupies bits 1 through 16 in a 24 bit word, will occupy bits 5 through 20 after the attenuation. So still plenty of room in a 24 bit word.

Another example: A 24 bit source which occupies bits 1 through 24 in a 24 bit word will occupy bits 5 through 24 after the attenuation, the 4 least significant bits are shifted out of the 24 bit format. This is where the dithering process comes in. If there wasn’t any dither applied, the 4 least significant bits in our example are simply cut off (truncated). This quantization process generates a quantization error which exhibits itself in the so called quantization distortion. On the other hand if dithering is applied, the quantization error gets decorrelated from the music signal, i.e. it is wideband noise. So the music does not get distorted it rather can be heard completely undistorted down to levels way below the 24 bit limit. There is only noise added – much more pleasant to the ear than distortion.

In-depth treatments of those topics can be found at Dithering and Jitter. The advantage of a level control in the digital domain is that a preamplifier can be omitted for a truly minimalist setup. Many D/A converters can easily drive a power amplifier and thus a typical setup would consist of a player, the D/A converter, power amplifier and speakers. It can’t get much simpler.

Another issue is the ripping of CDs to the computer’s hard disk. A very popular program to do this in a decent manner is the Exact Audio Copy program. This program makes sure that the bits are properly read off the CD. For audiophile quality ripping, never use any of the lossy compression schemes like MP3, WMA, AAC etc. Use either uncompressed data (e.g. WAV files) or lossless compressing formats like FLAC, WMA lossless etc.

Besides all the technical issues mentioned above, a computer based player has some other nice features, like:

-   The capability to convey audio data via networks, be it wire based or wireless. I.e. the music collection can be accessed from any room within the house.

-   The precious music collection can be backed up on a separate harddisk and put away, even to another place than the house in case there is a fire or a flood.

-   The access to the music is so much simplified that one all of a sudden rediscovers those "never heard" tracks.

To summarize, there is no reason why a computer, combined with a proper D/A Converter can’t be a decent playback system. There are pitfalls, but those will become less and less with the emergence of computer playback.

Daniel Weiss
Weiss Engineering Ltd

Voice: www.weiss.ch