CD: Jitter, Errors & Magic More Measurements
After the above article had been set in type, I saw a fax that Robert Harley had received from AudioPrism. It concerned measurements they had made that seemed to indicate that painting the edge of the CD green with CD Stoplight reduced the amount of noise in the final analog signal. Along with the idea that I gleaned from the Montreux AES Convention—that jitter noise appears as sidebands in the analog signal recovered from CD—this prompted a simple measurement that I thought might reveal such a difference. Using DRA Lab's MLSSA system, I would analyze the noise spectrum of a single tone signal with the CD first played au naturel, then with the edges painted green.
Upon reflection, the experiment would not be as simple as I had at first thought. An FFT-type spectrum analyzer, such as MLSSA, trades off bandwidth against resolution. With the maximum frequency resolution offered by the system, I would be restricted to a 1kHz bandwidth. Conversely, trying to analyze noise over the 20kHz audio bandwidth would restrict frequency discrimination to 37Hz or so, which might not be sufficient to show up any differences. In addition, the 12-bit A/D converter used by MLSSA might not feature a sufficiently wide dynamic range to reveal such differences. However, if I looked at -20dB tones, the noise floor would be just in reach, with meaningful results resulting.
The experiment would be twofold. I would first use the -20dB 920Hz tone from the initial pressing of the Stereophile Test CD with the analyzer bandwidth set to 1kHz (4kHz sampling). This would give a 2Hz frequency resolution, which I hoped would show the fine structure of the noise around the analog tone. Second, I would use the -20dB 1001Hz tone from the first HFN/RR Test CD with the analyzer bandwidth set to 25kHz. This would give me an idea of broad changes in the noise floor up to CD's 22.05kHz Nyquist frequency that might be placed at the feet of CD Stoplight. Regarding hardware, it seemed pointless to use the best transport possible; I wanted something that would misbehave if it could. I therefore played the discs on my Philips D6800 portable CD player, powered from an AC adaptor.
Initially, I got meaningless results. With the 920Hz test, painting the edges of the CD green produced differences in the levels of noise. However, so did removing the disc from the player, then replacing it and repeating the measurement. While the levels of the fundamental and of the AC mains harmonics were repeatable to within a fraction of a dB, the intrinsic error in the measurement system at the very low levels of background noise seemed higher than what I was trying to find. However, if the CD Stoplight did produce any difference, carrying out a number of measurements and averaging the results might reveal them. (Random up-and-down fluctuations in the levels stored in the frequency "bins" due to errors and quantizing problems will tend to cancel out, while any systematic change will be left unchanged.)
Time to start again. Using fresh discs, I carried out the noise analysis five times with untreated discs for each of the two tests, then averaged the results. This procedure was then repeated with the edges of the discs painted green (footnote 1). I then wrote a short program that would input the averaged data files produced for the two tests and output a graphic representation of the difference between them, as well as calculating the mean difference per frequency bin and the standard deviation. (I'm a great believer in letting computers do what they do best—boring calculations!) This program also gave me a good idea of the kind of differences that were produced by identical repeat measurements of the same signal, so that I could easily see what differences were of the order of the experimental error and what were not, and thus could be attributed to CD Stoplight.
The results were interesting, not the least because they were apparently positive. There appeared to be consistent, if tiny, changes in the nature of the noise floor that presumably correlated with the application of CD Stoplight to the disc edge. Looking first at the narrow-band analysis of the 920Hz tone, there was no overall discernible difference in the nature of the noise between 10Hz and 1000Hz. However, both times I carried out the test the nature of the noise floor either side of the fundamental tone seemed to change. With application of Stoplight, the noise floor between 900Hz and 910Hz went up slightly, by between 1dB and 5dB; that between 925Hz and 935Hz went down slightly. This might seem trivial, and it could never be audible, but it did happen with both samples of the Stereophile disc with which I performed the experiment.
With the wideband analysis of the noise floor and the 1001Hz tone, I didn't expect to find any differences around the fundamental, the frequency discrimination being too poor. However, I wondered if any systematic differences could be observed at higher frequencies. At first sight, there appeared not to be, differences in the levels stored in the frequency bins between the two conditions appearing random. (The levels of the fundamental tone were identical.) But eyeballing the difference data file, it appeared that there were more negative differences than positive.
This indeed was the case, the mean level per frequency bin decreasing by 0.28dB. Though this is very small, particularly if you remember that this difference is occurring at -100dB referred to a CD's maximum level, it still seems to me to be statistically significant. (For all you statisticians out there, there were 512 bins or samples, with a standard deviation of the differences between the value with and without CD Stoplight of 1.67dB.) For comparison, comparing the data files from supposedly identical tests—ie, capturing the waveform two successive times and performing the analysis—gave a change in the mean level per bin on the order of ±0.1dB, with a standard deviation of 3.54dB.
Admittedly, this was a very hurried experiment, with the measurable effect due to CD Stoplight being at the very limit of the system's resolution, around the level of the experimental error. It would therefore be foolhardy to argue that these minuscule but apparently real changes would directly correlate with the unsubtle improvement in sound quality noted when CD Stoplight is applied to the CD edge. I offer these results, however, as food for thought for those who persuasively argue that nothing that can be done to improve the character of the optical signal recovered from the disc can have any effect on the analog signal ultimately output from a CD player's DAC. Also, I would welcome hearing from anyone out there who repeats this kind of experiment with test equipment with higher resolution—the 16-bit DSP module for the ubiquitous Audio Precision System One, for example.—John Atkinson
Footnote 1: Five was an arbitrary number—five measurements were carried out because that was the number I had chosen for the program I had written to average quasi-anechoic loudspeaker frequency responses across a 30 degrees lateral window for Stereophile's loudspeaker reviews. A modified version of that program was used here.—John Atkinson