When is digital going to get the soul of music?

Mapman - I looked into DCs Ring Dacs to see how they get more detail thru dithering and found out that they don't. Addition of noise in not intentional - it's just byproduct of their scheme. If I understand it correctly now, they use number of current sources at lower bits and rotate them constantly to even out bit-weight. Extra resolution they try to preserve comes from digital oversampling filter. I found this description of differences between Multibit, Onebit and Ring Dacs :

"Multi-bit Nonlinearity — In multi-bit DACs there is a resistor associated with a current source for each bit. Each resistor is half the value of the one before it. So the ratio goes something like this 1 : 0.5 : 0.25 : 0.125 : 0.0625 etc. By the time we divide by two 24 times, the theoretically correct value of the last resistor is 0.000000119209289550781 of the first. Because it is physically impossible to achieve this type of accuracy, all multi-bit DACs suffer from some non-linearity (they distort the signal). This distortion becomes greater as you move from more significant bits to less significant (loud stuff to background detail). Typically, somewhere around the 20th bit the ability to resolve any additional detail is lost.

One-Bit Noise — In Bitstream (1-bit) DACs the resistor matching problem is eliminated and linearity is very good. However, the signal to noise ratio is terrible (6dB). A technique called oversampling is used to improve the signal to noise ratio to acceptable levels. However, the high oversampling frequencies result in narrow pulse widths. Timing errors now become significant, jitter increases, and the end result is thesame. The signal is distorted and the resolution of low level detail is degraded.

dCS Elgar Ring DAC — The dCS Ring DAC uses a patented 5-bit unitary weighted design (i.e., all the resistor values are the same). Oversampling frequencies are low (i.e., it’s less vulnerable to clock frequency errors). But, even this design isn’t perfect. Small variations in resistor value could still have an adverse effect on performance. Even with the carefully matched resistors used in the Elgar their resistance can change with age or temperature. To address this the Ring DAC, instead of using one resistor per bit, uses a large array of resistors. By using a proprietary algorithm (or is it Elgar-ithm) to continuously vary the number and positions of the selected resistors from sample to sample, as though around a circle (hence the name "Ring DAC"), the inevitable slight variations in the values of the resistors are randomly distributed throughout the quantizing range. This effectively turns any tolerance errors into random white noise, which is far more benign than the distortion that would otherwise have occurred. Finally, sophisticated noise shaping is used to move the bulk of the random noise into the high frequency spectrum above 100 kHz, where it is easily removed with analog filtering."

So, previous description I read (from Arcam if I remember correctly) was claiming extra resolution by random switching of current sources and dithering (adding noise). Now I found that they only try to preserve resolution coming from low order oversampler by rotating resistors in multibit converter (that follows) to keep necessary linearity - that would make more sense.

Kijanki,

I'm not an electrical engineer, but what you are describing does make sense to me based on my experience. It is a very sophisticated approach from the description.

I would expect that it helps deliver some of the distinctive "smoothness" I heard with this particular unit. Other aspects of the d/a conversion performed may account for the good resolution I believe I heard in conjunction.

It was a short but enjoyable audition with just a couple different source recordings. I would like to a/b compare it against various other designs to tell for sure how different it was. The technology applied is very unique and sophisticated for certain.

Kijanki, thanks for the info, and Albertporter, it was a feeble attempt at humor. I went to the live versus recorded event hosted by BAAS on Saturday. It was held at Cookie's studio(She formerly worked for Windham Hill.). A wonderful musician played acoustic guitar. Unfortunately, we didn't compare digital versus analog recording as I had hoped for. Cookie records to 2-inch analog tape. 5 microphones were used, and it was pointed out that using just 2 microphones has some problems(room, other instruments, etc.). The highs were the area that really stood out, as far as losses are concerned. The complexity of the highs was lost even on the best speaker we had in reproducing the highs( A Lowther cone with a ribbon tweeter. I still don't care for most ribbon tweeters, even after this demonstration.). It was also pointed out that many people in the recording chain may change the sound of the final product. It has me wondering if this is why Linn says it's the beat(foot tapping) we should look for when evaluating audio equipment. For those that are looking for the absolute sound, I would suggest that you only use perfectionist recordings, or ones where you were there, to determine the "absolute sound". After you've done this, let the chips fall where they may. In other words, don't try to optimize your system based on other recordings, because those other recordings may be wrong. I don't know if this technique will work, but theoretically that is what should be done.

It's the best you can do if that is the sound you want, but I would agree with Albert that it will never completely equal or surpass the detail possible with analog source, at least technically on paper.

On purely "technical grounds" or "on paper" - the CD is extremely good - far superior - perhaps it just doesn't sound as pleasant or as detailed.

FWIW - Dither is used to reduce/randomise "quantization errors" - it is especialy important when taking a 20 or 24 bit master and converting it to 16 bit. It is most important for the least significant bits where quantization error becomes important. Quantization error is due to the fact that the least significant bit (LSB) is only known to an accuracy of half of the LSB (the maximum digital resolution). When these errors are correlated with an input signal you can get some unwanted harmonics which dither eliminates by "randomizing" this resolution error to become white noise (raises the noise floor slightly rather than create an unwanted harmonic which might be audible).

For sure - if a studio makes some errors in the mixing and mastering they can create these unwanted harmonics and it can get onto your CD. A possible explanation for bad CD sound is that "sound engineers" are anything but "engineers" (most often they have a musical background rather than math and science) - it is very rare that they have a degree in time series analysis and signal processing. They may not fully understand what they are doing and generally learn by trial and error (sound engineers often start out in the tape room as a "gopher" and eventually work their way up to the mixing console).

Kijanki,

The generalizations of 1 bit versus multi-bit are kind of correct - but they make it sound awful - remember most of these DACS are achieving very low distortion levels (way way way lower than your speakers) - even the old multi-bits (and dynamic range way way beyond Vinyl, which is limited to about 60 db SPL on a good day with an ideal setup).

Initially, high clock speeds were difficult to achieve - so the resitor network DAC's were popular. These have been mostly replaced by delta sigma one bit DAC designs which became possible with higher clock speeds. (eventually higher speeds led to the concept of DSD and SACD technology being possible - essentially SACD is like a one bit DAC in a mathematical sense) The bleeding edge is now pushing the limits of clock speeds/circuit design and there is once again interest in a resitor network type DAC solutions (or a combination of both by a reduced rsitor network AND a high delta sigma clock speed) to improve S/N ratios above 110 db SPL (bear in mind that 110 db SPL S/N is stupendous already)

The ring DAC does sound like a form of variation on the latest DAC designs (astounding 120 db SPL S/N ratios are now becoming possible). AKM makes chips like this but they don't call them "ring DAC's", but they do use a "random" selection from a resistor network in order to solve the issues of non-linearities in resitor network DAC designs.

One thing to bear in mind is that digital technology is so extremely accurate that it is pushing the limits of both clock speeds and circuit design. The nice thing is that designers are now able to use clever mathematics to overcome even the limitations of both analog resitor network accuracy AND clock speeds to create extremely linear devices through a "random selection" which eliminates 'systematic errors' from real world devices by employing mathematical solutions.