Digital Audio Overview
Digital audio is constantly changing and this page is intended to provide a brief overview of some of the key concepts and terms in the digital audio world as well as debunk some of the myths that exist in audio.
A CD Player is basically a CD Transport with a stereo DAC and a clock generator built-in. Players commonly have a basic digital audio output (so that an external DAC may be used instead of the built-in DAC if required), but may not have a full set of digital interfaces as featured on a similar Transport.
The advantage of a Player over a separate Transport and DAC is that it is a more affordable, compact, 1-box source with minimal set-up required. The disadvantage is that interaction between the 3 functions (Transport, DAC, Clock) adversely affects the sound, despite multiple levels of power supply regulation and acoustic damping. The main problem in CD Players appears to be the motors in the mechanism interfering with the stability of the clock generator.
An audio DAC is a device which converts streams of digital audio data to analogue audio signals, which can drive a preamplifier / power amplifier / loudspeaker chain and so recreate the original music.
It is common practice these days to use a sigma-delta DAC architecture, in which the digital data is oversampled to increase the sample rate by perhaps 4, 64 or even 128 times, then reducing the number of bits and driving a very simple DAC circuit.
This approach gives good linearity while avoiding the need to manufacture an array of extremely accurate (and temperature stable) binary-weighted resistors or capacitors. The DAC is commonly 1, 2, 3 or 4-bit, switching at high speed. However, very fast clocks are more difficult to control and tend to exhibit more jitter, so higher switching speed is not necessarily an advantage.
Good quality sigma-delta audio DAC chips are inexpensive, and readily available from a number of semiconductor manufacturers. It is common practice for a high-end DAC to use 2 pairs of these DAC chips in a balanced arrangement.
Now ask yourself.....if everyone follows the same approach and uses the same chips then how can their products sound better or even different?
Our approach to this vitally important part of the digital audio playback chain is totally unique as all dCS DACs and Players use our patented dCS Ring DAC technologyto deliver the ultimate digital signal conversion - making your music sound more like the original analogue signal than any other product.
In a traditional system, the system clock is inside the CD Transport (or Player), an external DAC is slaved to the Transport. Listening tests indicate that this arrangement has plenty of scope for improvement.
More sophisticated Transports and DACs feature a Word Clock input / output on each unit and the ability to use the DAC as the master clock. This allows the Transport to be slaved to the DAC, where the DAC uses its internal clock to synchronise the entire system. This approach results in a better sonic performance as the clock in the DAC is less prone to interference from motors or external sources. The DAC is the point in the chain where dynamic errors in clock timing (known as jitter) are directly translated into errors in the analogue audio, so this should be the best place to site the master clock. Listening tests indicate that this mode gives a very worthwhile improvement in sound quality, which will obviously vary from system to system.
The best option is to use a dedicated Master Clock unit to run the whole system. A Master Clock has a very simple job to do: generate an accurate, stable clock signal with minimal jitter at a frequency either equal to the sample rate or an exact multiple of the sample rate. Listening tests indicate that using a good Master Clock improves the sound quality beyond the improvement gained by using the DAC in Master Mode.
An Upsampler is a digital-to-digital converter (DDC) capable of converting digital audio data at one sample rate to a higher sample rate. The term "Upsampler" was first used by dCS in 1998 during development of the dCS Purcell. Some other manufacturers appear to have confused the term "upsampling" with "oversampling", which is a similar process used in good audio DACs for many years. They have missed the point that upsampling is an intermediate step intended to provide enhanced data to drive an oversampling DAC.
Some positive observations that help to characterise the effects of upsampling in a dCS system:
- Upsampling to progressively higher sample rates makes progressive improvements to fine detail, sound stage depth and image separation. So, the sound quality increases as you upsample CD data first to 24/88.2, then 24/176.4, then 24/352.8 kS/s.
- Converting 16/44.1 to 24/44.1 kS/s makes a worthwhile improvement to the fine detail, so the resolution is important also.
- If data with a higher information capacity is presented to the Ring DAC, sonic improvements are reported.
One present view is that upsampling works by breaking the DAC's oversampling function down into 2 steps, presenting the DAC with finer, smoother data, making the DAC's job much easier and more accurate.
Jitter is the uncertainty in detecting the clock edges in a digital system. It is caused by a variety of effects, such as electrical noise, cable reflections, mains hum, radio interference or crosstalk from other signals on the board. All digital systems have some jitter - this is another fact of life. Jitter causes small timing errors in the system. Severe jitter causes errors in the digital data.
The jitter problem generally shows itself in the DAC, where digital data is converted into analogue data. The small timing errors cause the contribution of each sample to be slightly different to the contribution of the others, resulting in distortion. If there is a pattern to the jitter (for example, power line hum or an audio signal), this can appear as intermodulation distortion in the analogue output.
The type of PLLs used in digital audio act to lock the two signals together so that they are in phase with each other. This is the ideal lock point - but it also something of a "dead band" where the PLL is least sensitive to jitter - and so least able to filter out jitter.
The result is analogous to cross-over distortion in a class B power amplifier. When the amplifier is not being driven, the output sits at zero volts. At this point, both of the output transistors are switched off, so they cannot correct for external interference or soak up the current produced by pushing a loudspeaker cone. The usual way to correct this is turn both transistors on slightly, even when there is no signal. This "quiescent current" swamps out the dead band and ensures the amplifier is always in control.
You can exercise a PLL in a similar way by dithering the clock frequency. Dithering the clock involves changing the width of the clock pulse in a well-controlled way, to ensure the PLL always has something to do. This sounds rather like jitter, doesn't it? Unlike jitter, clock dither is deliberately made as random as possible (there is no pattern to it) and is limited to a frequency band that the PLL can easily filter out.
Samples per second (S/s) is technically the correct unit of measure for a data sample rate. As an example, each wire in a Dual AES stream runs at 48kHz, but the combined transmission rate is 96kS/s. Also, for packetised audio (e.g. USB, FireWire), the physical rate of the interface doesn't usually relate to the number of samples transmitted.
Master clocks generate a standard frequency (measured in Hz or kHz or MHz), not a sample rate.
You mean 5.644 MS/s!
It is easier to oversample to 2xDSD (1 bit data at 5.644MS/s) - the data stream is much more like the old bitstream DACs. The disadvantages are that "native" SACD playback is now impossible, and the system becomes very sensitive to clock jitter.
I2S is the data format produced by CD mechanisms, it consists of 3 wires: a Frame (or Word Clock), a Bit Clock and a data line carrying a stereo pair of data streams. I2S is intended for connections between chips a few centimeters apart, not between separate boxes. The 3 cables MUST be exactly the same length, otherwise the data clocked into the DAC may be corrupted due to cable delays.
I2S is used in this way to avoid the need for a Phase-Locked-Loop (PLL) in the DAC, in the belief that this will reduce the level of jitter. In reality, the clock from the CD mechanism is typically very jittery, but there is no PLL in the system to filter out this jitter!
SPDIF is well established and gives excellent results in a well-designed system.
Reed-Solomon encoding gives CDs effective protection against scratches, dust and other sources of reading errors.
There is a notion circulating to suggest that Reed-Solomon error correction causes jitter, and that it is best to ignore the error correction. The proposed alternative is that the disc is re-read several times until the data is stable. This could be effective if the error is caused by a moving piece of dust or hair. It is hard to see how this can work reliably with a scratched disc, as re-reading several times is likely to return data that is stable, but corrupted.
We do not agree that error correction is a source of jitter in a well-designed system.
The idea of optional filtering in DAC's relates to non-oversampling DACs, which run at the same sample rate as the incoming data (e.g. 44.1kS/s for CD data). One of the penalties of such a design is that the Nyquist images of the wanted signal produced during conversion to analogue appear at frequencies from half the sample rate upwards - typically above 23kHz for CD data. A conventional design would include an analogue filter to roll-off image frequencies above 20kHz, but this results in a frequency response that droops as the frequency approaches 20kHz. So, to avoid this, the analogue filter is left out.
These images are all outside the audio band, where conventional theory suggests they are inaudible, but unlike random noise, the images are directly related to the frequency of the signal components and appear at similar levels to the wanted signal. Many power amplifiers and speakers do not respond well to such a huge level of ultrasonic energy, producing intermodulation products in the audio band and other effects. In addition, there is evidence that the ear does respond to frequencies outside the audio band if they are related in frequency to the in-band signal.
As an example, if a 14.7kHz signal on a CD is processed by such a DAC, a high-level image will appear at 29.4kHz - the second harmonic of 14.7kHz.
In oversampling DACs, the images are filtered out by the oversampling process with minimal effect on the audio band.
There are some people whio believe that jitter is the only cause of performance difference in DAC's. This is definitely not the case.
The performances of different DAC designs are influenced by several factors, such as:
- Linearity of the converter
- Oversampling rate and resolution
- Accuracy, cut-off frequency and phase response of the digital filter(s)
- Random noise
- Truncation method - either dither (which deliberately adds noise), or noise-shaping (which moves quantization noise out of the audio band instead)