CD: Jitter, Errors & Magic Page 3
The second prevalent misconception about the CD is that if the ones and zeros are the same, the sound must be identical. This tenet is widely held, especially among engineers and computer users. A cornerstone of digital audio theory is that sound quality is independent of the recording or transmission medium. The collective attitude among many engineers is succinctly expressed by the phrase "bits is bits." (footnote 4). This adage expresses the idea that if the ones and zeros in a digital audio signal are identical, no other digital-domain phenomenon (in a properly designed digital audio system) can influence sound quality. This attitude explicitly rules out sonic differences between CD transports, optical vs coaxial cable, CD Soundrings, CD Stoplight, isolation feet, Armor All (footnote 5), and any other CD tweaks that incontrovertibly influence sound quality to sensitive listeners.
Those who subscribe to this theory tend to be scornful and contemptuous of anyone believing such differences exist. The following excerpt from a letter published in the Los Angeles Times, in response to an article about green marking pens on CDs, exemplifies the vehemence with which this belief is held:
"I got a good laugh out of Patrick Goldstein's March 4 Pop Eye column on the supposed salutary effects of green marking pens on CD sound quality. Yeah, and if somebody takes a picture of you, it robs you of your soul.
"What gullible and technically illiterate 'CD fanatics' fail to understand is that CDs are fundamentally different from vinyl LPs, where the precision with which the stylus tracks the squiggly grooves has a direct bearing on sound fidelity.
"The information on CDs is computer data, pure and simple—nothing but long strings of ones and zeros. And like all computer data, it's either there or it isn't; there are no 'in-between' gradations of playback quality.
"The chance that a marking pen can improve CD sound isn't minuscule. It's zero. People who claim to hear a difference are suffering from wishful thinking or hallucinations."
It is apparent that the author of the above letter (a computer engineer and president of a musical electronics company) formed his opinion based on prejudice, not experimentation. During my involvement in CD mastering, I too shared this view. However, listening experiences forced me to abandon these closely held beliefs. No amount of rhetoric will convince the unbelieving. I encourage anyone who asserts dogmatically and without a trace of doubt (a good description of my attitude at one time) that "bits is bits," to listen with an open mind through a decent playback system to the effects of, say, CD Stoplight. Critical listening is an essential tool for exploring these phenomena, yet is dismissed as "unscientific" or "subjective" by the "bits is bits" mentality. As we will see, variations in playback quality exist even when the digital code is unchanged.
Jitter: I was fairly confident that an examination of the various tweaks' effects on error correction would reveal no changes. Indeed, I had conducted a similar experiment with CD Soundrings a few years ago and found no reduction in error rates with the rings (footnote 6). In addition, it is possible to prove that a digital master tape and a multiple-generation digital copy have identical data (in a bit-for-bit comparison), even though they do not sound the same. Moreover, JVC's K-2 Interface seems to improve sound quality without changing the pattern of ones and zeros in the digital code.
Clearly, some other phenomenon is at work. The most likely source of audible changes in digital audio is jitter, a time-axis variation in the digital bitstream. Jitter is most often introduced by mechanical imperfections in digital audio-storage devices. A CD player's rotational servo, for example, can introduce time-base errors (jitter) in the recovered signal if its speed varies even by a tiny amount. Another example of a jitter-inducing mechanism is a digital tape transport. The speed variations which introduce jitter have an analog counterpart in wow and flutter. Jitter can also create bit errors if severe enough, although this is uncommon.
Fig.5 shows the effects of jitter on digital code. Jitter changes the position where a transition in the digital signal occurs. It has all the right information and in the right order, but skewed in time.
Fig.5 Jitter affects the time relationship between digital transitions. (Reproduced from Principles of Digital Audio, Second Edition (1989), by Kenneth C. Pohlmann, with the permission of the publisher, Howard W. Sams & Company.)
If you have ever watched a multiple-generation VHS video tape, you've seen the effects of time-base error, which is analogous to digital audio jitter. With each successive generation, the horizontal lines that make up the picture become increasingly skewed in relation to each other. As this time-base error accumulates, image outlines become slightly crooked, making the picture appear fuzzy.
If jitter reaches the digital-to-analog converter, severe errors in the output analog signal result. Fig.6 shows how timing variations can affect the shape of an analog waveform. In theory, however, jitter is never allowed to reach the DAC: jitter-laden data are input to a buffer and clocked out with quartz-crystal accuracy to the DAC. Many engineers, including those at JVC who designed the K-2 Interface and digital designers at Madrigal Audio Laboratories, maintain that jitter can still reach the DAC, despite the buffering.
Fig.6 Timing variations distort the shape of the reconstructed analog waveform.
In researching the article, I had a fascinating discussion with Steve Taylor, Vice President and Director of Engineering at Madrigal, about digital audio in general and jitter in particular. He shared the results of research conducted at Madrigal by himself and Kevin Burke, Vice President of Research. They found a correlation between sonic performance and the quality of the HF signal read from a CD. The amount of jitter in the signal, as well as the shape and asymmetry of the HF signal, caused sonic changes when all other variables were kept constant. The circuitry that processes the HF signal before decoding can also affect the sonic character. Consequently, they developed a circuit that cleans up the HF signal before decoding.
Their research also confirmed that even the smallest amounts of jitter in the word clock that synchronizes the D/A converter affect the musicality of digital playback. A shift of as little as 100 picoseconds (0.1 nanoseconds, or 1/10 of a billionth of a second) in the word clock causes audible conversion-timing errors. These errors are more noticeable on high-frequency signals, and, significantly, low-level signals. A slight time shift (see fig.6) causes amplitude distortion. If a sample uses only a few bits, such as when quantizing a low-level signal, the error in relation to the signal is much greater than if the same error were applied to a high-amplitude sample. Jitter thus has a greater deleterious effect on low-level signal components (footnote 7). It is probably no coincidence that JVC's K-2 interface and high-quality CD transports—both of which reduce jitter—improve soundstage depth and low-level resolution, contributing to a sense of spaciousness.
With some of the CD's technical aspects behind us, let's look at the measurement results.
First, none of the CD tweaks reduced data errors. Neither the Block Error Rate, nor the ability to correct, rather than conceal, errors was affected by any of the devices or treatments. In fact, BLER slightly increased after application of liquids to the disc, probably the result of tiny surface scratches introduced when buffing dry. The red pad supplied with Finyl left small fibers behind on the disc, also slightly increasing BLER. Uncorrectable (E23) errors occurred at the same place on the Pierre Verany test disc (track 31, with a 1mm dropout), with or without any device or treatment. The farther into the disc an E23 (uncorrectable) error occurs, the better the ability to correct large HF signal dropouts.
In addition, BLER remained the same when the CD player was in a quiet environment or when subjected to 90dB SPL, measured at the player. This experiment was performed when attempting to measure any effects of the Arcici LaserBase isolation platform. All measurements were performed five times in each condition (the untreated, or control run, and after application of the tweak). Figs.7 and 8 show the Design Science CD Analyzer printout of an untreated disc and the same disc treated simultaneously with Finyl, CD Stoplight, and a CD Soundring on top of the Mod Squad damper disc. As can be seen, no error-rate reduction is evident.
Fig.7 Design Science CD Analyzer error printout of an untreated disc.
Fig.8 Error printout of same disc but treated with Finyl, CD Stoplight, and a CD Soundring on top of the Mod Squad damper disc.
Although I heard an improvement with Finyl, it was most certainly not the result of error reduction, as claimed in their white paper. If the mechanisms by which these treatments work are unknown, the promoters should say so instead of blaming data errors.
I did, however, find huge variations in BLER between discs made at different manufacturing facilities. In fact, I measured several commercially available discs whose BLER exceeded Philips's "Red Book" specification of 220, all from Discovery Systems. Although a high BLER will not cause audible artifacts, it will significantly affect the CD player's ability to correct, rather than interpolate, errors. When a long burst error is encountered, the information used to reconstruct missing data comes from surrounding data. If these adjacent data contain errors, the ability to reconstruct missing sections is reduced. Even though a high BLER, by itself, will not be audible, it can cause the player to interpolate or mute more often in the presence of burst errors.
(Incidentally, I once bought a disc whose Table of Contents was not even readable. Subsequent measurement revealed that it did not meet the Philips "Red Book" specifications in five different parameters: BLER, HF signal level, track pitch variations, starting radius of Table of Contents, and starting radius of program. I returned the disc to the record company with my test data. Although this disc was an exception, many poor-quality discs do find their way into stores.)
The makers of CD Stoplight claim to have measured a difference in the recovered analog output signal with a treated disc. They played a pure tone from a test disc and measured the spectral content on either side of the tone. Reportedly, a CD Stoplight-treated disc produces lower-amplitude sidebands around the pure frequency. Just as this was going to press, AudioPrism faxed me graphs made on a Hewlett-Packard spectrum analyzer that support their findings. Without knowing all the measurement details, the graphs do appear to show a slightly lower noise floor after the addition of CD Stoplight. Watch for an update as more information becomes available and can be verified.
Footnote 4: This belief is held more by computer and electrical engineers who work in fields other than audio, or digital audio engineers who do not listen to and evaluate music as a part of their work. Most CD mastering engineers, who have daily "ears-on" experience with digital audio, tend not to hold this view.—Robert Harley
Footnote 5: There is evidence that Armor All applied to CDs can have serious side effects, including attacking the polycarbonate and fogging the laser lens. (See "Letters" and "The Audio Anarchist" in this issue.) I therefore discourage treating CDs with Armor All.—Robert Harley
Footnote 6: See "Industry Update" in Vol.12 No.8 and John Atkinson's interview with Monster Cable's Noel Lee and Rodney Herman in December 1989 (Vol.12 No.12).—Robert Harley
Footnote 7: An excellent discussion of the effects of data jitter appeared in an article by Malcolm Hawksford in the February 1990 issue of HFN/RR. In addition, at the 88th AES Convention, held in Montreux, Switzerland, in March 1990, a workshop was held on the audible effects of data jitter. A full report on the workshop's conclusions and on the convention in general will appear in the June issue, but briefly to summarize the workshop's findings, jitter of the digital data stream translates into sidebands either side of the spectral components of the finally recovered analog audio signal. For example, if jitter in the data stream presented to the DAC has a periodic nature equivalent to, say, 200Hz, then ±200Hz sidebands will appear around every spectral component in the analog music signal. As suggested by Madrigal's engineers, it appears that the degradation increases with rising signal frequency, implying that jitter has to be kept below a threshold very much lower than originally believed if audible degradation is not to result.—John Atkinson