Volume 9, Number 4
The Problem With the Sony/Philips DSD Scheme and Super Audio CD: Why Professional 1-Bit Sigma-Delta Conversion is a Bad Idea
The most eagerly anticipated paper of the LA AES Convention this fall was Drs. Lipshitz and Vanderkooy's exposé of the Sony/Philips DSD recording system. For several years Sony has trumpeted Direct Stream Digital as the ideal archival format for audio catalogues. More recently Philips joined with Sony in breaking away from the DVD group to propose their own Super Audio CD format, a format which uses DSD in place of PCM digital audio.
Lipshitz and Vanderkooy challenge the adoption of DSD. "To use 1-bit sigma-delta conversion as a professional archival format would be stupidity of the highest order."
The Toronto section will be treated to an expanded version of the presentation given in LA along with live demonstrations. From the people who taught the audio industry about dither and noise-shaping, don't miss this opportunity to hear a presentation which may well change the course of digital audio.
It's Stan and John versus Sony and Philips!
After the brouhaha, join us for a brew and a "ha ha" . . . or perhaps a "ho ho" . . . as we celebrate our Annual Christmas Social. Drinks and fine munchies will be provided. Enjoy the good company and the view from the tenth floor Artist's Lounge a CBC Broadcast Centre.
Presented by Stanley P. Lipshitz and John Vanderkooy, Audio Research Group, University of Waterloo Date Tues, 12 Dec 2000 Time 7:30 pm Place CBC Broadcasting Centre Address 205 John Street
Between Front and Wellington
St. Andrew or Union Subway stations
Please Pre-register Call National Mail Box at (416) 922-8122
1-Bit sigma-delta converters are in principle imperfectible. We demonstrate this fact. The reason, simply stated, is that, when properly dithered, any input signal drives them into overload. The consequence is that distortion, limit cycles, and noise modulation can never be totally avoided. Recording systems, and in particular professional recording systems, based upon 1-bit sigma-delta conversion are thus a bad idea. In contrast, multi-bit sigma-delta converters (and here, multi-bit refers to maybe three or more bits in the converter) are in principle infinitely perfectible. They can be properly dithered so as to guarantee the absence of all distortions, limit cycles, and noise modulation. The audio industry is making a tragic mistake if it adopts 1-bit sigma-delta conversion as an archival format to replace multi-bit PCM. We shall demonstrate these conclusions by means of live simulations during the talk. The Super Audio CD of Philips and Sony is a 1-bit format.
All the major manufacturers of high-quality analog-to-digital and digital-to-analog converters have, in the last few years, moved away from 1-bit sigma-delta structures, to architectures based upon oversampling converters using more than one bit. Why did they do this? Primarily because they have been unable, after a decade of trying, to make the 1-bit structures perform as well as they can make the multi-bit structures perform. In particular, they have been unable to completely prevent limit-cycle oscillations and noise modulation from occurring in the baseband under varying signal conditions. These disadvantages outweigh the one supposed benefit of 1-bit sigma-delta modulators -- their intrinsic linearity.
In spite of this, there is a push under way to persuade the consumer, and (most alarmingly) the professional audio industry, that 1-bit sigma-delta is the conversion format of the future. We argue that this is a retrograde step, and that encoding schemes like DSD [Direct Stream Digital (TM)] are misguided. We do this by showing that 1-bit sigma-delta conversion is inherently imperfectible, whereas the multi-bit architectures have no such theoretical limitation to their perfectibility. The essence of the argument is the fact that, in a 1-bit sigma-delta architecture, it is impossible to properly dither the converter without driving it into overload. A multi-bit sigma-delta converter (or one of another architecture), on the other hand, can be fully dithered without overload, and so can guarantee the absence of distortions, limit cycles, and noise modulation. Only a few least significant bits (LSBs) of converter dynamic range are needed to take care of this, and depending on the order of noise shaping used, a converter of three or more bits (i.e., at least 8 LSBs) is a suitable choice. The point is that this is unachievable with a 1-bit sigma-delta system. Moreover, the multi-bit system needs much less noise shaping to achieve the desired baseband signal-to-noise ratio. Indeed, when one compares 1-bit and multi-bit oversampling converters, it is not easy to find technical advantages to the 1-bit scheme. Almost every comparison favors the multi-bit converter.
In this talk we make comparisons of the two systems at a given overall data rate, and conclude that the multi-bit system can give the same audio bandwidth as DSD, with better signal-to-noise ratio and with no artifacts. In addition, every stage of further processing of a DSD signal involves a multi-bit PCM to 1-bit DSD re-conversion, with the attendant cumulative increase in the nonlinear artifacts. Multi-bit PCM processing does not have this Achilles' heel, and with proper dithering and optional noise shaping, it is free of all audible artifacts.
To use 1-bit sigma-delta conversion as a professional archival format would be stupidity of the highest order.