Something I saved some time back (not written by me):
ffracer said:
Interesting. What would be a bad case of jitter (amount of jitter in picoseconds) on a CD transport vs. the graphs? That would really help answer the question.
That's the squillion dollar question! What we do know is that it can be shown mathematically that you need less than 121 picoseconds of random sampling clock jitter at 20 kHz to ensure FULL 16-bit performance. That requirement drops to 0.474 ps for a 24-bit system. Remember, we are talking about random jitter, and an input frequency of 20 kHz, so there is not a degradation over the entire audio band. The first question becomes: is this even audible? Published papers tend to suggest that the answer is: NO. Using random clock jitter, the jitter level has to be MUCH higher before degradation is audible -- this random jitter is at the nanosecond and higher level level according to these publications. However, many ADC/DAC designers believe that sinusoidal jitter is (a) more likely in practice, and (b) more likely to be audible. Of course, one still has to take the frequency and level of the tonal jitter into account along with masking effects to determine whether it will be audible, or not. The level of sinusoidal sampling clock jitter that audibly degrades a sustained piano note, say, may be very different to what is needed to audibly degrade an orchestra/band playing more complex sounds. (It should be obvious that you need your best [lowest phase noise] clock at the ADC, since any sampling jitter due to the clock gets 'baked into the cake' during A-to-D conversion.)
Now, what does this have to do with the audible differences between various CD pressings? IMO, not a lot! Jitter often gets mentioned with regard to differences between pressings. However, the jitter due to inaccuracies in the pit/land structure on a CD is part of what is called transmission jitter. This is quite different to jitter on the ADC/DAC clocks. When CD was designed, all aspects were fully characterized, which led to its forward error correction schemes, channel coding, etc., etc. One of the key design goals was that it should be cheap/easy to duplicate. This is not possible if you have a very strict tolerance on the length of the pit/land structures (and other disc parameters). So, the channel code and decoding system were chosen to allow some 'slop'. In fact, the Red Book specifies a maximum disc jitter level of 35 nanoseconds! This is orders of magnitude larger than the levels we were talking about above so it must be audible, yes? Well, no!
As I'm sure you and Dr. crap... know, NONE of the data on a CD makes it to the DAC! Let me repeat: NONE of the data on a CD makes it to the DAC. Due to sample offsetting, left/right channel interleaving, convolutional data interleaving, Eight-to-Fourteen-Modulation (EFM), and the use of DC-free run-length limited coding, a CD player has to read a large amount of serial data, deserialize it, correct it (if needed), unmap the EFM, and undo the interleaving/offsetting to arrive at the parallel 16-bit left/right audio sample values that need to be presented to a DAC. All this is done in DSP and buffer memory. Now, what about the pit/land jitter? It is ALL removed in the data slicing process. The light reflected back from the 3T to 11T pit/land run-lengths gives rise to an AC signal at the output of the photosensor. Due to the properties of the coding used, this AC signal is sliced, and a determination made about the bit values. At this point, one of two things can happen: the values are determined correctly, or not. However, their value is 'set' and they are loaded into DSP/buffers until you have enough data to do error detection/correction, de-interleaving, etc. That is, the audio sample values are computed anew within the player. The transmission jitter is completely removed. If it is large enough, it will cause bit errors, of course. If there are only a few, they will be corrected, if they are constant, playback will fail.
At this point someone will make the case for coupling the DAC clock to the jittery clock recovered from the disc, or noise from the laser servos finding a way into the output stage, etc., etc. Of course, there are bad ways to implement things, but the presence of these issues is now well known, and the better players mitigate against them. As an example, consider the Sony SCD-1 from 1999. I chose this player because Sony produced a fairly detailed document on its design, and their methods of dealing with these issues, which can be found here:
http://www.docs.sony.com/release/SCD1_TWP.pdf
The player has also been analyzed in depth here:
http://www.stereophile.com/hirezplayers/180/index.html
and if one looks at its jitter spectrum:
http://www.stereophile.com/content/sony-scd-1-super-audio-cdcd-player-measurements-cd-player-3
one can see that ALL the test signal data-related sidebands are below -120 dB. This level of performance is not unique to the SCD-1, but shows what is achievable in a well-designed system.
In the early days of SACD development, one of the Philips disc technologists who was busy making the hybrid disc a reality came to see me (I was at Philips Research at the time). He had read about audible differences between CD pressings, and wondered if there was something in it. His idea was if we could determine which physical disc parameters influenced sound quality, then we could determine some optimum combination, and use this as a carrot to the large disc replicators to get on board: not only would they get SACD technology, but they would also get the best sounding CDs possible (which would also be a feather in SACD's cap if we could show that the CD layer of a hybrid was the best). Since I had links to professional, golden-eared listeners, some of whom had mentioned pressing differences to us, I agreed.
We obtained digital masters directly from some key recording/mastering engineers, and also bought UK, US, European and Japanese pressings of the titles (all known to be sourced from the same digital master files). One of the titles, by a major band, was on both EMI and Columbia in various territories at the time, so we were sure the pressings in that case were made on very different equipment. We had a separate mastering facility make CD-Rs from the digital masters, and my colleague also made hybrid SACDs. Discs were then sent to the golden ears, and all they were asked was to rank them.
My colleague then did an extensive series of measurements to determine disc jitter, pit/land lengths, track pitch, flatness, eccentricity, tilt, reflectivity, etc., etc. We checked C1 and C2 error rates, looked at the EFM eye diagrams, considered the AC signal level of the critical 3T code value, etc., etc. We then tried to correlate the measurements with the rankings. There was no consistency to the data. A parameter that might be close in value on several highly rated discs would then be vastly different on another highly rated disc. No matter how we re-arranged the data, we could not find a magic subset of parameters (and their values).
Then we thought more about it. The Red Book does not dictate exact values for disc parameters, nor does it stipulate which servos, laser pick-ups, decoding ICs, etc. MUST be used, instead it defines all the physical parameters in terms of [typical value +/- tolerance]. Disc replicators are free to vary the parameters as they see they see fit, provided ALL parameters fall within their tolerances. Similarly, player manufacturers are free to choose optical pick-ups, clamping schemes, decoding schemes, etc. such that they can cope with any disc having any combination of parameter values (within the tolerances). Think about optical pick-ups, there are those that move along a radius, there are those that move along an arc, and then there are those that are fixed and move the disc instead. The signal from the photosensor can contain HF EQ in order to make reading easier, but this is not mandatory. There are various schemes to deal with jitter, there are different DAC chips, different crystal oscillators and ways of shielding them, different power supply strategies, etc.
Now, given the variety in player designs, why would one set of disc parameters provide best sound quality in ALL cases? Even if we assume that the phenomenon is real, we should not expect that a 'grail' pressing can be found that is optimal for all players. So, at the very least, stick with what you like!
For myself, I was given a complete set of the discs for a certain title. I had identical players for fast comparisons, I had Stax headphones, I had two listening rooms, one of them fully floating with an NR = 5 dB (so very quiet), I had no end of high-end monitoring equipment. At times, I could convince myself that I heard differences, but every time I went back to the other versions, I would hear the exact same things, at the exact same relative level, with the same placement in the image, etc. (For the record, I do not consider myself a golden-eared listener!)
ffracer said:
One question that rarely gets mention is what is the effect of jitter on DVDs and Blu-Rays?
If there is an effect, it is the same as for CD, since all optical media work in the same fashion. Moreover, hard-disks work in essentially the same way too. So, if there is a problem it should affect all systems, and yet it is common here to see claims that digital music played from hard-disk somehow magically works, and yet CD is subject to all kinds of issues. (Let's ignore known bad implementation issues here, and consider only well-designed CD players/transports.)
- Black Elk, forums.stevehoffman.tv, Jan 11th, 2013