Originally Posted by nick_charles
Er, you have it wrong again ! Benjamin and Gannon (AES paper) used correlated jitter (the worst kind) at different frequencies (2k, 5K and 17K) and thus had lower thresholds ~ 20ns while it was Ashihara et al "Detection threshold for distortions due to jitter on digital audio" (Acoust. Sci. & Tech. 26, 1 (2005) not AES) who used random jitter the far less noticeable sort and found the ~250ns thresholds. Please get your facts right
There is something very wrong here. I did not read those papers, but 250nsec? Are you really talking about 250nsec? or is it 250psec?
250nsec is 1000 times bigger then 250psec.
The SPDIF data rate for a 44.1KHz sample rate is 5.6448Mhz which is 177nsec. If the jitter is 250nsec, you will be missing a lot of data (or repeating data) to the point of such breakdown of the sound, that jitter is no longer the issue. Something is screwed up here.
When you transmit digital audio, the transmitter has a clock and you send one data state at each clock.
Inside the receiver, there is also clock circuit. The clock is locked to the AVERAGE rate of the data, so in fact it is at the same rate as the transmitter clock. The receiver clock happens once for each data bit. If the reciever happens every 177nsec, but you allow a jitter to be greater then 177nsec you may will miss the data, or have a "double clock" causing the data to be repeated. All the other bits will be out of sequence. And such a violation is not subtle. A single violation is going to make a 6dB sudden step, a terrible distortion.
In fact, jitter figure is in r.m.s. and random jitter has a bell shape curve - normal distribution. Put it together and you find that the peak to peak of the jitter (peak to peak deviation) for a 250nsec is 1.5usec!!! That is for 3 sigma so that only covers some 99.9% of the times. In reality you want to have than 25nsec to just make for reliable data transfer, before any consideration for sonics.
So even data transfer of 44.1KHz spdif requires much less then 177nsec timing error (less than 25nsec or so). And at 25 nsec it the jitter is huge. For good audio one needs to clean it way down. The issue is than about how much jitter was removed before it gets to the critical circuit where it matters.
I wrote about the differences between data transfer jitter and conversion jitter. Maybe my posts are too complicated. I tray to keep it simple, but it seems to me that what I say is not sticking too well. In this case, 250nsec (that is 1/4usec) does not make sense to me at all. I assume we are talking about digital audio and spdif...