View Single Post
Old 2008-02-11, 06:30 PM
Re: Vinyl records vs. Cds.

Here's a good article about multibit DACs vs. single bit sigma delta DACs. The SACD system is single bit sigma delta @ high sample rate by design. What that experiment shows, or attempts to show, is that the 1 bit sigma delta DAC is the best digital sound will ever have to offer. I don't believe this is the case, and I also don't believe it is just a case of better mastering making all the difference in sound quality, although this is a big factor. Perhaps 24 bits is too little to really make a huge difference? Perhaps all the source components in the test had 1 bit converters which might have been the equalizing factor (bringing sound quality of all sources down) no matter if the source was 1 bit SACD, 16 bit CD, or 24 bit DVD-A? 1 bit DACs are severely lacking IMO and a particularly bad 5 disc Sony CD changer made me loathe digital sound with a passion for the longest time. I much preferred my LPs, there was no contest, and I only had an entry level Denon turntable w/stock cartridge. There are other ways to describe the sound of analog other than warmth. Presence or realism come to mind. Theoretically, this can be matched eventually with digital if enough bits per sample(esp.)/samples per second are used.

It is ironic that 25 years after the introduction of the first generation CD player by the Sony/Philips consortium a groundswell of designs has emerged that challenge one of the medium’s fundamental engineering assumptions. It had been considered a given that an anti-imaging filter is necessary to remove the ultrasonics generated during the digital to analog conversion process. Recall that the earliest players used multi-bit DAC chips followed by analog "brick wall" filters designed to steeply attenuate the image spectra above 22 kHz. While the specs looked good on paper, the promise of "perfect sound forever" turned sour quickly as a chorus of complains echoed a common theme: bright, fatiguing sound that ultimately resulted in digititis – an allergic reaction to digital sound. The establishment’s initial reaction was to blame the messenger; namely, digital masters were said to be the culprit. I recall listening to J. Gordon Holt’s Sony CDP-101, the first kid on the block with a CD player. This one being a seminal first-generation player, and a gift from the marketing folks at Sony. Out of a stack of some 30 CDs, only a couple managed to sound decent. Slowly, the real problem was recognized to be the brick wall filter and sure enough second-generation players took advantage of evolving digital signal processing technology and incorporated digital oversampling filters positioned prior to the multi-bit DAC. I have no preconceived bias against digital filters; they neither generate new information nor improve resolution, but they do allow the use of much gentler analog filters, which are audibly benign.

Now, just when it appeared that the digital ship had righted itself, a new "dark age" was spawned with the advent of the single-bit sigma-delta converter. Author Ken Pohlmann (The Compact Disc Handbook) gives an excellent analogy of how such a chip works. He likens it to a light switch operating at a high frequency. The two extreme amplitude states are off and on, but also any intermediate level can be achieved by toggling the switch off and on rapidly at a given frequency. The single-bit DAC was seen as offering the potential of increased linearity relative to the older R2R type at lower cost while obviating the need for factory calibration. In contrast, the R2R DAC uses an onboard voltage-divider resistor network or ladder capable of generating 65,536 voltage values. Each bit in a 16-bit data word enters the ladder through a switch. The most significant bit (MSB) enters the ladder at the top, while the least significant bit (LSB) is assigned the last section in the resistor network. A data zero keep the corresponding switch open, while each 1 closes a switch and allows that bit to contribute to the overall output voltage. Granted, the LSB is difficult to maintain in calibration, but the output is free of the RF switching noise that afflicts the sigma-delta type, which requires sophisticated digital noise shaping techniques to work at all. Since noise cannot be destroyed, noise shaping merely shifts the noise: reducing noise within the audible bandwidth while increasing it at higher frequencies. More recent low-bit sigma-delta designs improve the situation somewhat but fail to change the bottom line: improving linearity at the cost of increasing high-frequency noise is a poor tradeoff.

In my experience, designs based on the sigma-delta chip tend to sound bright and/or lack convincing timbre accuracy. Altmann Micro Machines’ progenitor, Charles Altmann, is much more empathic about this issue. In his opinion, there is no music possible with sigma-delta DACs. In his experience R2R chips are the only way to achieve a listenable sound quality. However, the type of R2R chip used, he feels, is of far less importance than the skill of the designer. While from a technical standpoint he appreciates R2R type DACs such as the Burr-Brown PCM1704 (true 24-bit noise-free resolution), they are not necessarily well suited to zero oversampling applications. He believes that he has been able to push the Philips TDA1543 dual 16-bit DAC to incredible results. Incidentally, by virtue of its single +5V voltage rail, it is compatible with battery power supplies. Some might consider the TDA1543 as a relic from a bygone era, but its low-cost, potential for good sound and compatibility with battery power have made it a favorite with many designers. In fact, three of the four DACs reviewed herein use this particular chip.

So Are Anti-Imaging Filters Really Needed?

Well, yes, if using a sigma-delta DAC; the noise shaping employed makes anti-imaging filtration mandatory. However, the answer would appear to be a resounding No in the case of an R2R DAC. The definitive means of settling such questions is, to my mind, the time-honored listening test. It was the inimitable Harry Olson, who elucidated this principle many ears ago in the form "the ear is the final arbiter in audio matters." There is not necessarily a direct relationship between what is measured and what is perceived. Al Bregman, author of Auditory Scene Analysis: The Perceptual Organization of Sound (MIT Press, 1994) relates that in about 1969, a few years after he had arrived at McGill University as an assistant professor in cognitive psychology, he became involved in an experiment on auditory perception in which the signal was a rapid sequence of unrelated sounds. The realization that the perceived sequence was not the actual sequence of sounds launched him into a life-long study of auditory perception. Unfortunately, the scope and spectrum analyzer have displaced the auditory system in the mistaken belief that perceptual pleasure is an inevitable consequence of engineering excellence.

It was the golden-eared Peter Qvortrup at Audio Note UK who first dared to listen directly to an R2R’s DAC output – minus analog filters and what he refers to as "digital trickery." What Peter discovered is a sound quality that is much more closely akin to the vinyl experience. In my experience, zero oversampling gives the impression of a more believable soundstage. The spatial impression in terms of depth and width perspectives is typically better defined relative to oversampling designs. It is as though the auditory system is presented with a better set of cues with which to synthesize a 3-D impression of the auditory stream. Perhaps the Gestalt psychology principle of closure is at work here. This is a mechanism for dealing with missing information, a means, if you will of connecting the dots (to use a visual analogy). An example given by Al Bregman is that of a soft sound being masked or drowned out by a louder one. If the softer sound can be heard both before and after a burst of the louder sound, it can be perceived to continue behind the louder sound – even if it is physically removed while the loud sound is being played.

Welcome to the natural voicing of zero oversampling. What a breath of fresh air! Gone is the endemic brightness of early CD players and sigma-delta converters and its associated sensory overload. Yet, the relative paucity of zero oversampling designs makes you wonder whether most DAC designers spend more time measuring than listening.

What about the ultrasonic energy present at the output of a zero oversampling DAC? Is it likely to overload or generate distortion products in the associated analog chain? While that is always a possibility with high-frequency test signals, there is no evidence that this is at all an issue with music program material. The proof is in the listening.

Last edited by Tubular; 2008-02-11 at 06:36 PM.
Reply With Quote Reply with Nested Quotes