by R. A. GREINER
Bit By Bit (AES presentation, “Audio in Digital Times”, May 1989)
[P. A. Greiner is Professor of Electrical and Computer Engineering at the University of Wisconsin, Madison. He teaches and does research in electroacoustics, acoustic measurements, applications of digital signal processing, audio system design, and noise control. He holds over a dozen patents in electronic instrumentation and audio systems and was elected a Fellow of the Audio Engineering Society in 1984.]
During four days, May 14 to 17, 1989, about 400 persons listened to almost 40 hours of technical papers at the Audio Engineering Society's International Conference, "Audio in Digital Times." This event, held in Toronto, was the society's second major conference devoted entirely to digital audio and one of the most important conferences since the introduction of digital recordings to the general public some half-dozen years ago.
As one who has been deeply involved in digital signal processing for a number of years, I was impressed with the high quality of the papers presented and with the frankness and speed with which the attending audio engineers have approached the continued improvement of the technology and art of audio signal reproduction.
The range of the conference can be appreciated by going over a list of the sessions' topics. They included: State of the art in technology and basic trends, conversion techniques and performance evaluation, professional digital audio, consumer digital audio, digital studio design and practice, dig ital signal processing, digital audio in film and broadcast, and several sessions on theory of digital signal processing. It is impossible to describe all of this in a few pages, so I have selected some of the topics which I think will be of greatest interest to Audio readers. I will also give my overall impressions of the state of the art of digital audio and of future trends which the conference represented.
Digital audio is a new technology-a new art, if you like--and the professionals at the meeting were in general agreement that it is not at the present moment perfect. In recent years, there have been some very poor digital recordings made and produced on both vinyl and Compact Disc. There have been a great number of superior recordings done as well, and they clearly greatly outweigh the poor ones. The important issue, faced at this conference, is that the problems which caused the poor recordings have been, for the most part, tracked down.
This necessary step has been followed very rapidly with measures designed to solve the problems. As a result, most current and future recordings will continue to be better technically. The solutions, many of which were presented at this conference, are a combination of hardware advances, software design, and recording techniques.
One of the major problems has been, and is, the conversion of the signal from analog to digital and back.
In some cases, the analog-to-digital converters, the necessary filters, and the digital-to-analog converters may have introduced artifacts into the recordings because they have not been as precise as the technology requires; in other cases, they may not have come up to the standards which have been set for the CD. Some early master digital tape recorders have not had full 16-bit accuracy and may not have been linearized with the correct amount of dither. Some have simply not been maintained or adjusted with enough care. In most cases, these are human problems as well as technical difficulties.
The development of digital integrated circuits has more recently made possible the development of more re fined and theoretically better A/D and D/A conversion techniques. Oversampling during both the record and play back process and the application of digital filtering and noise-shaping techniques were the subjects of many of the technical papers presented. The results of these techniques, just now being implemented, are better S/N ratios, lower noise floors, much better linearity at low levels (probably one of the most important factors in perceived quality), and removal of anti-aliasing (or anti-imaging) filter artifacts that have made some CDs objectionable.
One of the major points of interest to the consumer is that these problems, which must be expected in a new technology, are becoming better and better understood and are being eliminated with advances in digital circuitry that are taking place on a daily basis. I believe that the critics of digital audio expected perfection to leap full grown from the bosom of technology. This technology, like any other, has had to grow and develop. It is doing so at a rapid pace.
Some comfort and some distress might be taken from several papers presented which discussed the technical adequacy of the CD format specifications. The format requires that the sampling rate be 44.1 kHz and the quantization be 16-bit linear. The question is, "Is this adequate?" This very important question was discussed in some detail at the conference, in the light of what we now know about digital systems and human perception. For regardless of the use of frequency oversampling and higher quantization accuracy at either or both the recording and playback ends of the system the stored data on the CD has only this accuracy. The answer, as I heard it, was that the sampling rate is entirely adequate and the quantization accuracy was adequate, if carried out with full precision for the final storage of the digitized information, but finer quantization (more bits) is required in the intermediate stages of production be cause of the digital computations that need to be made at these stages.
However, the standard is not much greater than that needed for the most critical applications. In fact, it may be just a bit on the marginal side for professional studio applications.
This is generally good news since it confirms the continued use of the cur rent CD format for the consumer market. It is not the best news for those CDs which have been produced using a large amount of intermediate digital processing or equipment that has not been in virtually perfect operating adjustment. Fortunately, professional dig ital equipment is becoming available for studio use. This equipment uses higher bit quantization and oversampling, which allow high-speed, high-accuracy digital filtering and much of the other digital processing that might be required in the production of CDs.
An overview of the conference, I believe, suggests that there is reason for great optimism. In the next few years, audio production will move well beyond its current status and the final CDs which the consumer hears will all be of the quality of today's very best.
Over 12 hours of technical papers were given dealing with signal conversion between the analog and digital domains. It was made quite clear that this must be done with great precision and care. An excellent lecture/demonstration was given by Professor Lipshitz, current president of the AES, and his colleague Professor Vanderkooy, both from the University of Waterloo in Ontario, Canada. This lecture demonstrated all that can go wrong if the analog signal is not converted to the digital domain accurately and with just the right amount of dither. Various kinds of squeaks, birdies, noise, and hash were demonstrated and shown to be indigenous to certain kinds of con version errors. That such errors exist in some currently available CDs was demonstrated by playing a number of examples. While these clearly audible defects are not common in commercial CDs, their mere existence is a bit shocking. In one of the worst cases, a Mozart piano concerto (CBS MK-37267), the music is modulated with noise and faded into dismal digital hash in a number of soft passages.
Several manufacturers presented papers describing their latest converters (IC chip sets), which provide 18- or 20-bit accuracy and a high degree of frequency oversampling. While only 16 of the bits are ultimately stored on the CD, high-accuracy conversion enables excellent digital filtering and assures that the rounded 16-bit words are, on the average, accurate to a small fraction of the lowest bit. It was shown that it is important to have accurate 16-bit conversion to actually get the full ad vantages of 16 bits in terms of S/N and linearity for very small musical signals.
(The reader might appreciate that Audio has tested and reported on this very important feature of CD players for some time.)
The new high-accuracy converters use a technique called delta-sigma modulation, oversampling conversion, or noise-shaping digital conversion.
Whatever term is applied, the converters use very high sampling rates, e.g. 6 MHz-and only a few bits, sometimes one. The data is then digitally filtered, decimated, and converted to 16- or 20-bit words at a normal sampling frequency. This process eliminates the need for anti-aliasing filters and can attain accuracies of 20 bits or more if required. The advent of [...]
(adapted from Audio magazine, Sept. 1989)
Also see:
A New CD Test Standard (Dec. 1985)
Crest Factors of CDs (Dec. 1988)
Thomson CD Recorder -- Exclusive U.S. Test!!! (Mar. 1990)
How Hot are CDs? Recording levels of CD format (July 1989)
= = = =