Up to Computers and related stuff Home (komarix.org)

In the tables below, I show an averaged error measurement for the digital outputs of several Redbook CD transports. The individual errors (which I summarize with average and standard deviation) are the horizontal distances between times at which the waveform actually crosses zero volts (according to my oscilloscope), and the nearest multiple of 177 nanoseconds since the waveform's first zero-crossing. This is related to the jitter measurements you mind find in old Stereophile jitter articles, but my test equipment isn't nearly as nice as theirs. There are many good digital audio articles about the S/PDIF and AES/EBU standards, which you should read if you want a better understanding of the digital audio waveforms I am sampling.

All measurements are in nanoseconds, and come from four 600 nanosecond samples each, from four diverse music tracks (see the list of tracks further below), for a total of 16 600ns samples. Plots of one 600ns waveform from the Shostakovich preludes and fugues are linked from the tables, as well as a 120ns sample. Each horizontal tick on the graph is 500ns, and the vertical scale is in volts. The raw data for these graphs is also linked, though due to mime-type difficulties, you might not see any newlines if you view the data through your browser. Instead, try downloading the data files, and reading them with your favorite text editor.

Subjective "a/b" listening experiments, by one to three people, ordered the transports from best-sounding to worst-sounding as follows:

That is the same order as seen in the averaged error measurements, as well as the player's prices. We did not audition the Lexicon RT-10 using it's S/PDIF interface, a configuration which had larger average error than the Pioneer Elite 79avi. The Lexicon S/PDIF error measurement is most likely artifically inflated by my primitive error-calculating scripts, which do not handle spurious zero-crossings after digital state transitions (see the accompanying plot for the Lexicon's S/PDIF interface). My guess is that an S/PDIF receiver (e.g. my Anthem AVM-20) is likely to ignore these spurious off-clock zero-crossings, and hence the S/PDIF interface will have actual average error close to the AES/EBU interface's average error. My evidence is that the random bit errors caused by wrongful-interpretation of spurious zero-crossings are likely to sound like popping noises or static, and none were heard during a quick listen to the Lexicon's S/PDIF interface. Further support comes from the digital output of my Anthem AVM-20, which cleans up the spurious zero-crossings (though it introduces significant time errors).

The bigest surprises for me were

  1. the variety of waveforms between players
  2. the spurious zero-crossings in the Lexicon S/PDIF data
  3. the additional error created by the Anthem AVM-20's digital output
  4. the correspondance between my primitive error measurements, and the subjective sound quality
  5. the clear difference in sound between digital transports

The best-sounding player, the Lexicon RT-10, has fabulous detail. In a Shostakovich Preludes and Fugues recording with Nikolayeva playing piano, you can hear the piano's hammers, and the sustain and damping of the piano strings is obvious. After a blind a/b test against our beloved Rotel RCD-02, my (ex-)wife said "I know which is which, and I don't appreciate how bad the Lexicon makes our Rotel sound." Another surprise: I can finally understand what Michael Stipe is singing on R.E.M.'s "Murmur" album.

The recordings used in the measurements below are

I didn't get fancy and try to guarantee that the exact same sample from each track was used in each experiment, which seems nearly impossible with my test equipment (small buffer on the scope and one-second resolution on the transports' displays). Instead, I took several samples from starting at same indicated time on the transport's display, and computed average error across all zero-crossings in all the samples. The same test equipment (scope with 1Mohm impedance, low impedance probe, and 75 ohm S/PDIF cable) was used in every experiment. The S/PDIF output was not presented a 75 ohm terminating load, and this is probably the biggest weakness of my measurements.

I tried one additional experiment, for a quick assesment of how CD read errors affect sound. My original hypothesis was that read errors would be the main factor differentiating transports. I did an overlapping sector rip of the Shostakovich prelude no. 24 to a wav file, and accessed it through the Momitsu v880n's ethernet connection. The result was underwhelming, with no memorable difference in sound quality compared to the Momitsu v880n reading directly from the CD. I later learned that read errors often cause popping noises (imagine changing the MSB for one audio sample from 0 to 1, while the surrounding samples maintain an MSB of 0), not a subtle difference in quality. I am convinced that CD read errors are not an important distinction in CD playback sound quality.

S/PDIF Waveform Samples


6000ns plot 6000ns data 1200ns plot 1200ns data


6000ns plot 6000ns data 1200ns plot 1200ns data


6000ns plot 6000ns data 1200ns plot 1200ns data


6000ns plot 6000ns data 1200ns plot 1200ns data

Other Waveform Samples


6000ns plot 6000ns data 1200ns plot 1200ns data


6000ns plot 6000ns data 1200ns plot 1200ns data


6000ns plot 6000ns data 1200ns plot 1200ns data

Up to Computers and related stuff Home (komarix.org)
Created by Paul Komarek, komarek.paul@gmail.com