Feb 2001
Review Dec 2000 Meeting
|
|
Feb 2001 Review Dec 2000 Meeting |
The Problem With the Sony/Philips DSD Scheme and Super Audio CD: Why Professional 1-Bit Sigma-Delta Conversion is a Bad Idea Review On Tuesday 12 Dec 2000, the Toronto Section of the AES saw out the old millennium with their annual Christmas social in the scenic 10th floor Artist's Lounge at the CBC Broadcasting Centre. But the first order of business was Stanley Lipshitz and John Vanderkooy's exposé on the Sony/Philips Direct Stream Digital (DSD) recording system and their Super Audio CD.
For years Lipshitz and Vanderkooy have been educating the audio industry on the importance of properly dithering digital audio systems. This must be why the Drs. find the DSD system particularly galling. DSD is based on 1-bit sigma-delta conversion. The problem, in a nutshell, is that it is not possible to properly dither a 1-bit system.
Dr. Lipshitz began by reviewing the fundamentals of ADC and DAC converter technology and the principles of dither and noise-shaping. In order to quantize an audio signal properly, +/-1 least-significant-bit (LSB) of triangular probability density function (TPDF) dither must be added to the signal. Without this dither there will be distortion and noise modulation in the stored signal. (For more detail on how dither works and why it is necessary, read the review of a Toronto Section meeting from February 1994 at http://www.vex.net/TorontoAES/93-94/apr_94.htm#Digital
In the mid-1980's one of the difficulties of designing converters was in eliminating most-significant-bit (MSB) glitches. One trick was to add some DC offset to the converter so as to avoid the zero crossing. Another one was to use 1-bit converters. Since there was never a signal at 0, there was no MSB glitch. As a bonus these circuits were much cheaper to make and drew significantly less power than their multi-bit brethren. Here was an ideal technology for cheap and portable players.
But there are many problems with single bit converter technology which professional audio practitioners should find alarming. In a single bit system there are two levels. Recall that a properly dithered digital system requires +/-1 LSB of dither. The truth is that you can apply exactly the correct amount of dither to a 1 bit system. But, if you then add any audio to the input you've overloaded the system! Whoops.
The extreme non-linearity of a 1-bit system means that a great deal of feedback must be employed in the circuit. The noise-shaper for DSD creates a huge amount of noise and at very high frequencies. This noise cannot be feed to loudspeakers. The filter used to remove the extreme amount of noise must be at least a seventh order analogue filter.
Another problem for the industry is the lack of DSP available for DSD. Rare is the recording which does not require some sort of processing after the recording. How do you perform an edit? Adjust the gain? Add reverb? Each step requires a conversion to PCM for processing and then back to DSD. Each conversion step means that more artefacts and high-frequency noise are added to the signal and because of the nature of DSD, these artefacts are significantly greater than those introduced in properly dithered multi-bit systems.
It is possible to partially dither the signal and improve things. But the distortions and noise modulations are there and have become a permanent feature of the recording.
In a world which sometimes seems to be driven almost solely by business it is reassuring to know that the halls of academia can still offer some respite. With no commercial interests at stake Lipshitz and Vanderkooy are free to criticize and expose. "To use 1-bit sigma-delta conversion as a professional archival format would be stupidity of the highest order."
With the presentation of the paper in LA Sony has admitted that their professional systems in fact use 8-bit converters and not 1-bit as their literature had previously stated. Questions remain. Is the output of the converters stored as 1 bit? What happens in postproduction? How many 1-bit conversions is the signal subjected to before it reaches the consumer? Would Sony have admitted to this without prodding from Lipshitz and Vanderkooy? Further reading Read a précis and abstract of the paper at our website at http://www.vex.net/TorontoAES/00-01/dec.html
Tomlinson Holman features the paper presented at the L.A. AES Convention in the October 2000 issue of Surround Sound Professional (see page 11).
More fun can be had by going to Google and entering "Lipshitz Vanderkooy DSD" as search terms.
Return to the Front Page