13th April 1999 - Super audio CD

The April lecture was presented by Philips and Sony, and covered their new jointly-proposed Super Audio CD (SACD) consumer format in fascinating depth. SACD is currently vying with the (also new) DVD-Audio disc, to replace CD for high-quality stereo and surround-sound music delivery.

Gerard Lokhoff, Philips' Technical Co-ordinator for SACD began with an introduction to the SACD format. The generic SACD is basically an 'ordinary' DVD, with about 12 times the data capacity if a CD (or 25 times for a 'double-sided' disc). This capacity, in both SACD and DVD-Audio formats, with the aid of costless data reduction, allows about 74 Minutes of high-quality multi-channel surround audio to be encoded on the disc, plus an assortment of 'extra data'.

But the SACD is different from a Conventional DVD in a number of respects; the most important being that it is a 'hybrid' disc - that is to say it has a second disc laminated onto it. The extra 'red-book' layer is a low-density, CD-compatible layer. It is this hybrid construction which Philips and Sony hope will prove to be the decisive factor in the format war: the red-book layer carries a conventional l6-bit stereo version of the program which can be played on a Conventional CD player. The high-density Layer, which is read by the new SACD player through the red-book layer, carries a multi-channel and/or higher quality stereo version for those consumers suitably equipped. This backward compatibility is unique to SACD, and will allow 'single inventory'; retailers will only need to stock one type of disc for all.

A point of debate between the SACD and DVD-Audio camps has been the cost led yield of hybrid disc production. According to Philips, only two additional manufacturing processes need to be added to an ordinary DVD-9 production facility to manufacture SACDs. Once the technology is mature, they project a low production cost with a high yield.

Peter Eastty, Chief Consultant Engineer of Sony Broadcast Europe, described the fundamentals of the Direct Stream Digital (DSD) audio coding scheme used on the high-density layer of the SACD. This is another fundamental difference between SACDs, and conventional CDs and DVD-Audio discs, which use PCM coded audio. In PCM coding, the sample rate is generally only about twice the required signal bandwidth (although DVD-Audio allows sample rates up to 192kHz) and each sample must be represented by a long binary number whose maximum linear error defines the dynamic range of the audio signal. The l6-bit wordlength of CD allowed an unshaded dynamic range of about 94dB, although DVD-Audio supports word-lengths up to 24-bits, or a dynamic range of about 142dB if this could be realised in the production chain.

A DSD coded audio signal consists of a data stream at 2.8224Mbit/s, and these bits literally represent single-bit samples at a very high sample rate. So how is the dynamic range of a DSD coded signal not, as one might expect, only a couple of dBs - albeit with a bandwidth of 1.4MHz!? It is diffcult for the mathematically-challenged among us to understand the principles of noise-shaping delta-sigma modulation, wherein an analogue or PCM signal can be converted to a l-bit representation whose dynamic range in the audio band is preserved by 'shaping' the inevitable degree of quantisation noise into the vast inaudible expanse above. Peter Eastty is unique in his ability to explain such 'impossible' mathematical feats so that your grandmother could grasp them in her sleep.

DSD coding brings certain advantages, the main quality benefit being that with a high enough data rate, a l-bit signal can simultaneously provide a wide bandwidth with a high dynamic range in the desired baseband. This trade-off of bandwidth versus dynamic range can even be chosen during the production process - a luxury which can't be achieved with a single PCM format.

Whether the 2.8224Mbit/s data rate of SACD is sufficient to provide a dynamic performance equivalent to the PCM possibilities of DVD-Audio, or whether the staggering bandwidth possibilities are superfluous have been much debated. However, Peter pointed out some advantages of DSD coding which are less well aired. For example, the processing delay for a DSD signal is negligible, since a sample period is only a fraction of a microsecond. Problems of overdue delays in mixing consoles and unequal signal path delays in broadcast installations have dogged digital systems since their earliest days. Even where PCM systems have 'adequate' bandwidth, the usual decimation and interpolation filters' employed in the converters have dispersive behaviour in the time-domain whose adverse audible effects we are only just beginning to understand. These filters are simply not necessary in a DSD system. The homogeneity of the DSD bitstream brings other advantages - interfacing is simpler since no framing is required, and since each bit is no more important than any other (unlike in a multi- bit PCM system) the tolerance to bit-errors is higher. Other proposed advantages include easy downsampling to all the mainstream PCM sample rates, and simpler D/A conversion. But on this latter point, Peter remarked that the theoretically- feasible dream of D/A conversion with a single resistor and capacitor will probably remain a dream, since designing high-quality l-bit D/A converters is actually quite challenging.

Much has been made of the difficulty - some say impossibility - of processing DSD signals. In a final tour-do-force, Peter confounded the critics with a remarkable demonstration of a complete two-channel processing chain such as might be used during pre-mastering. The chain included faders, multiband parametric EQs, fully- featured dynamics control and adjustable delays, as well as all the usual mastering bells and whistles. Not only did it all work, but so seamlessly. Peter was able to show us that the DSD processing was free from many of the usual DSP trouble-spots - no zipper noise at the bottom of the faders, nor undersampling artifacts in the dynamics peak-detectors, nor instabilities at extreme frequencies of the EQ. Such qualities are not yet routine in digital systems, even in traditional PCM systems, so the development of such a well-behaved prototype in an unexplored branch of signal processing stands out as a truly remarkable achievement.

Finally Paul Reynolds, responsible for Business Development for SACD, put the commercial case. Interestingly, Paul didn't start with the usual discussion of improved sound quality, but drew a picture of a music business in the doldrums, in need of revitalising. The affluent young no longer aspire to high-quality premium-priced sound systems and recordings - SACD must provide new excitement so that audio can compete with Nike and Nintendo. It seems that this is more likely to come from the surround capability than from any sound quality improvement over CD.

The SACD has been designed to meet the requirements of the music industry's International Steering Committee (ISC) in matters of copyright protection and anti- piracy measures, as well as providing the requisite video, graphics and text enhancements. Most importantly, Paul stressed again the need for compatibility of a new format with the world's 650 million CD Players. Our thanks are due to Gerard Lokhoff, Peter Eastty and Paul Reynolds for a comprehensive and compelling presentation.

Ian Dennis