Sunday, February 26, 2012

Why the performance of high-priced disc transports isn't measured

Rev A (see Notes)

There are many claims that high-priced transports produce cleaner data, but no measurements to support these claims.  Robert Harley of the Absolute Sound might have inadvertently provided an explanation for why so many audio reviewers, including himself, resort to mumbo-jumbo and hand-waving when it comes to "explaining" the supposed differences in transport sound quality:
The Memory Player (see p. 120) might, indeed, produce a sonic improvement, but that improvement is certainly not attributable to the mechanism claimed by the manufacturer. First, uncorrected errors on CD playback are rare. The data stream read from a disc is rife with bit errors, but those errors are instantly corrected and replaced with the identical original data. Corrected bit errors result in absolutely no loss of information or change in sound quality. The CD’s CIRC (Cross-Interleaved Reed-Solomon Code) error-correction scheme is remarkably robust, allowing burst errors of up to 4000 consecutive bits (equivalent to 2.5mm of track length on a CD) to be reconstructed with the identical missing data. This is not interpolation or error concealment, but perfect bit-for-bit correction. This powerful error correction is what allows CD-ROM to work so well. Missing or interpolated data in audio data could introduce a momentary glitch; errors in computer code could cause a program to crash. Granted, CD-ROM has an extra layer of error detection and correction above and beyond that of CD, but that extra safeguard is rarely invoked. Second, if uncorrectable errors do occur (primarily because of disc scratches or dirt), error-concealment circuits replace the missing data with a best-guess interpolation based on the waveform’s values before and after the missing data. Error concealment is rarely audible. In some instances, concealment can be heard as a low-level tick sound. The manufacturer’s assertion that error concealment is commonplace, and results in “a congestion and harsh odd (order) harmonics borne of a synthetic tone” is simply false. Uncorrected errors, at worst, introduce momentary ticks, not a global change in sound quality. If the error is too great for concealment (12,300 consecutive bits, or 7.7mm of track length), the player momentarily mutes the signal. Furthermore, these interpolated data are not “called ECC (error correction codes)” in any of the technical literature. I must repeat that an uncorrectable error is a rare event on a CD that has not been abused. In fact, the uncorrected error rate for a reasonably well-made CD that has not been abused is roughly 10[exp]–10, or one error in every 10 billion bits. This isn’t to say that the Memory Player doesn’t improve the sound. There are many instances in high-end audio where an audible change exists that cannot be explained by our limited understanding of the science. Indeed, other mechanisms might be at work. But “a congestion and harsh odd (order) harmonics borne of a synthetic tone” is not one of them. I have some familiarity with this subject: I worked in a CD mastering lab where my job duties included modifying CD players to extract the various error-correction flags from the decoding chips for input to custom error-analyzers, performing bit-for-bit comparisons between the data on CD master tapes and CDs replicated from those masters, examining CD pit structures with a scanning electron microscope, and correlating error rates on replicated CDs with mastering parameters (see my Audio Engineering Society paper, co-written with Ray Keating, “CD-V Signal Optimization,” presented at the 1989 Hamburg convention). 
from Robert Harley comments on the Memory player
The Absolute Sound, December 2006

In other words, assuming that an optical-disc reader is functioning properly, and reading a disc that's in decent condition, its output will contain less than 1 error for every ten billion bits!  The cheapest CD player therefore produces essentially the same data as the most expensive transport.

While searching for reviews of high-priced transports which include performance measurements, I found an article written by someone else who wanted to get to the bottom of this particular issue, and did so by performing a test.  He found that the above claim by Robert Harley is correct.  So, the notion that there is any advantage to spending enormous amounts on CD transports is pretty thoroughly discredited, and additional refinements have been made in optical disc technology since then to make DVDs and Blu-ray discs feasible.

However, I noticed that, of all of the reviews of expensive optical-disc transports I perused, none include measurements, probably because measurements would prove that they ultimately aren't any more accurate than any cheap optical drive.  Instead, they say that the supposed audible superiority is due to mysterious factors.  This has been the standard line for years, and there has been no organized attempt to identify these mysterious factors, although the old bugaboo, jitter, is usually mentioned, but without showing how it effects the data.  As long as jitter isn't so bad that it causes data errors, then it doesn't matter. However, it is critical to minimize jitter at the DAC's clock inputs, because at that point, it becomes part of the analog output.

Blinded by audio hucksters

Now that I've examined the issue of errors from optical-disc drives more closely, and realized that I was wrong on the issue for quite a while, mainly because I fell prey to BS intended to sell something, such as audio streaming systems or Bedini Clarifiers (which help on some machines by reducing the amount of work done by the laser servo, which can affect the audio stage via the power supply unless isolated).  I realized that I was wrong when I stumbled onto an alternative explanation that makes sense, so that my mind was opened to evidence that CD error correction is indeed amazingly good.

Rev A:  Added last paragraph