Digital audio quality: is jitter a factor?

Factual Questions
1

Recently, I’ve seen a number of people spending upwards of $100 on “high performance” optical cables for their digital audio setups, raving about the improved sound quality they get over run-of-the-mill optical cables. My natural response was a cackle, justified by my understanding that digital audio signals will sound perfect, no matter what the cable they travel over, as long as no bits are corrupted along the way. “Ah, but what about jitter?!” the audiophiles exclaimed.

Their belief is that the timing of digital signals is subtly altered as it flows over the cable, causing some samples to be longer or shorter than others. They think that this causes the Digital-to-Analog-Converter to modulate the analog output at slightly wrong times, resulting in reduced quality.

So, what’s the straight dope? Does this kind of jitter exist? Assuming that the jitter isn’t bad enough to cause any bit-errors, will it have any audible effects? Wouldn’t this also require you to believe that there was not a SINGLE buffer anywhere between the cable and the output of the DAC?

Here’s a link on this subject: http://www.digido.com/portal/pmodule_id=11/pmdmode=fullscreen/pageadder_page_id=28/

2

The problem with so many “articles” published by audio manufacturers and magazines is that they tend to be biased and unscientific.

Is jitter a problem in audio systems? Probably not. First of all, audio signals are very low in frequency, even when represented by a digital serial data stream. Secondly, just about every serial format incorporates some form of error correction. And then thirdly… do you really think optical cable & connector manufacturers would be producing products that have this problem?

I’m not saying jitter does not exist. I’m just saying that it is very likely a non-issue for audio applications.

Take the money you’d spend on fancy optical cables and put it in a Better Speaker Fund[sup]TM[/sup].

[IMHO]
I can’t think of a hobby or endeavor that has more B.S. than high-end audio.
[/IMHO]

3

My teo-cent version is that digital signals do degrade (after all, on a fundamental level, they are analog signals coding digital data) BUT it takes a huge amount of degradation to turn a 1 to a 0 or vice versa – basically, about half the strength of the overall signal.

That’s a little bit of an overstatement, but you get the idea: if ordinary optical cables were capable of inducing anywhere near that level of distortion, they’d be somewhere on the level of smoke signals, orders of magnitude worse than the cheapest wire. Why would we bother with them> We wouldn’t and they aren’t. They are orders of magnitude better.

There is some smoke being blown around, and I don’t mean inside those “premium” audio cables.

4

In a perfect world, jitter wouldn’t be a concern in this context. But the world is far from perfect. Here’s an article that addresses some of the issues of jitter and DtoA.

And cheap digital cables can cause enough things to go “off” timewise that it might be a problem for some. Of couse, that’s no excuse to go out and buy $1000 cables either.

Sending a high speed stream of real time bits a few feet isn’t so easy when the bozos who developed the standard didn’t know diddly about digital signals.

5

[Hijack]
What type of modulation is used for those signals?
Are certain methods of modulation more prone to jitter than others?
[/Hijack]

6

I generally trust the Digital Domain (the OP’s link, and ftg’s as well), because it sounds like the author isn’t just spewing high-end audio BS. One important point from the link (bolding his):

A digital cable is a D-D device; digital in, digital out. Based on that link, the digital cable shouldn’t add a jitter problem. The overall timing would still be that of the data entering the digital cable.

One thing to keep in mind is that the digital data from a CD isn’t converted one bit at a time to audio output. For error correction purposes, any individual time sample is mixed with a whole bunch of others, and all that data is spread out over a portion of the CD data (about 2K bits, IIRC). All of that 2K has to be read before any of it is available to the D/A to be converted to audio.

7

Thanks for the responses, guys. Zenbeam, we’re mainly considering what happens to the audio after it’s been decoded from the CD bitstream into LPCM and sent out along a TOSLINK optical connection to a DAC. Am I right in assuming that even a single buffer within the DAC would eliminate the effect of any jitter induced by the optical cable?

Also, how is it that an optical cable can affect some bits more or less than others? Shouldn’t every bit be subjected to identical conditions in the cable, since we don’t have to worry about EMI with optical cables?

8

I think ZenBeam’s point was that anything before the DAC wasn’t going to be affected that much at all.

Not only do systems have buffers, but there are error-correcting codes. KP, I’d think the degradation would have to be even more than half.

9

From what I’ve heard about jitter, the only time it matters is when it isn’t a random change. Random differences in the time between samples will come out in the wash, if those differences are cyclic, it’s another story. For instance, if sample gaps get longer and shorter over a 1000Hz cycle, it could be noticable.

It’s hard to believe that cables are going to make a difference, though. Considering that the data being sent is all identical just different patterns of 1 and 0, how is it that the gaps between data change? They’re not different frequencies, reflection effects shouldn’t disrupt the data coming through. I can’t see where the mechanism for change is.

Analog signals are much more prone to subtle alteration than digital, and it’s tough to prove changes in them based on the cable used.

10

At the receiving end of an optical cable, there is a circuit that reads the light levels and reconstructs the pattern of ones and zeros. The only way that a cable could affect this process is if the cable attenuates the signal too much. In that case, the signal strength will be low, with respect to the noise in the system, so some ones will be misinterpreted as zeros and vice-versa. Also, the clock recovery circuit will have a harder time determining where the boundaries between them are, so you could see some errors due to sampling at the wrong instants.

This pattern of recovered ones and zeros then goes to the DAC (digital to analog converter), which uses its own crystal-based clock to output the data.

So the jitter in the cable transmission system won’t be seen as jitter in the output, but it could be seen as wrong bits. However, for this to be a problem, it would have to be so bad that making it just slightly worse would cause the audio to go away completely. That’s because of the nature of sending digital bits for one, plus the error-correction that’s built into the system. If you take the cable and make a small loop, you’ll cause it to have more loss. If this doesn’t cause the audio to go away, then your system isn’t near that marginal point where signal loss and jitter cause perceivable audio problems.

11

The 1’s and 0’s are sent as pulses along the digital cable. Even if they start out as square pulses, by the end of the cable, they are distorted. For the signal …010…, what used to be a square pulse of duration T might look bell-shaped, and maybe only the middle T/2 of it is high enough to register as a 1, so the transition to 1 is only T/2 before the transition back to 0.

Even if the CD player does all the error correction, and just sends audio samples along the cable, the cable still won’t add appreciable jitter. It has to send at least 16 bits times two channels, 44100 times per second, or 1.4 million bits per second. The jitter introduced by the cable can only be about +/- 1/2 the period of each bit, or about 0.35 millionths of a second, or about 1/65th of the audio sample period. That’s simply not going to be audible. The jitter you might hear would be from the D/A if it has its own clock, or else from the CD decoder if the D/A uses that. Either of these presumably has a clock that’s supposed to run at 44100 audio samples per second. If that varies up and down between, say, 44090 and 44110 samples / second (numbers admittedly pulled out of the air. I couldn’t find real numbers.), you’re looking at jitter at 1/4410 of a second, or about 10 audio samples, which could plausibly be audible.

12

What kind of bandwidth are we talking about here? I’ve built numerous PWM and V-to-F circuits using fiber optic cable. (This was for galvanic isolation or noise immunity.) A couple were in the 10 MBit range, and I never used exotic hardware… just COTS components from Digi-Key and Newark. And the receive signal on the scope always looked perfect.

I’m not saying a fiber optical can’t degrade a signal; it certainly can if you’re operating outside its rated specifications, or if the coupling sucks. But this is rarely a problem. My advice? Just use run-of-the-mill optical cable and invest in better speakers.

13

Really cheap cables tend to have poor quality control and are more likely to get crimps in the plastic tubing that acts as the light path. More expensive cables have thicker shields and are more likely to have glass light paths instead of plastic. These can still break, but they don’t crimp like a cheap plastic cable. On the other hand, there is no reason to spend a crazy amount on a cable unless you need a long run >25ft or so. $25 for a decent 3ft cable isn’t unreasonable.

I agree with the posters above. A really expensive cable is not going to give you the bang for the buck you will get from updating your speakers. Put the extra money in the new speaker fund.

14

Since you raised the issue, I’ll explain my reasoning for saying “half” is an overstatement, not an understatement. Please excuse me if this is too simplistic for you. Since this is a general interest board, I figure it’s best to explain on the level of the Teeming Millions, whose expertise may lie elsewhere.

In any analog modulation encoding a digital signal, from a simple square wave on up, noise and errors are stochastic, probablistic phenomena, described as an average over time. (When we model the effects of noise make certain assumptions [with varying degrees of justification] about the statistical properties of the effective interfering signal, but noise is -er- noisy, and variable in the short term.)

One error per million would probably be imperceptible in audio: even without error correction, it would create, at worst, an isolated ultrasonic click every 3 seconds or so. A dog whistle or the bats who fly, unheard, in the summer night of every temerpate or tropical clime are like foghorns or stadium cheers by comparison.

The same error rate would quickly crash a computer or trash a hard drive, because the bits in a program or datafile aren’t independent. Audio bits aren’t independent, either, but it’s hard to select a threshold where the effects are preceptible, esp. when the errors are likely to be systematic (e.g. occurring more in certain bit combos) due to physical effects; and therefore produce overall nonrandom bias.

In fact, digital audio EC schemes begin to be overwhelmed at surprisingly low error rates: most errors are fixed, but some are not, and some new errors are created by incorrect on-the-fly correction. The algorithms are written to err on the side of “unobtrusiveness”, but that in itself is a systematic bias.

EC isn’t magic. It can’t restore what is gone. It can only guess the correct bit, or fill in something unobjectionable. Error correction capabilities aren’t thresholds. If a bitstream has 10 ppm errors, and an EC is rated for 100ppm correction; some errors will still get through, if only because sometimes two of the 10ppm errors will occur within, say, 1 kbit of each other, for a transient error rate of 2000ppm

Since perception can’t be objectively quantified, I looked at the threshold of objective digital measurement. which is ridiculously over-sensitive in the time domain. [Who cares about X bit errors per 70 minute CD?] I apologized for my overstatement because it’s possible for noise that is under 1% of the signal power to flip bits, because noise often comes in transient spikes. Those errors should usually be inaudible even without EC and processing, but they’re there.

Raise an incredibly good post EC error rate to the 650 millionth power, and you’ll several errors. That’s why mission critical servers use EC RAM even though normal RAM may only generate one error in a million million million. are these few errors meaningful in audio? No, but it’s all I can objectively calculate.

A Modest Rant
Here’s a paradox for you: a lot of very knowledgable people, even professional audio engineers, present very technically compelling arguments for the significance of jitter and other mechanical end analog effects in CD reading. Yet if you rip a CD ten times on the cheapest desktop CD drive, and then do a binary compare of the resulting files, you’ll find remarkably few bits of difference.

This easy test seems universally ignored. I did it almost 10 years ago on a $20 drive in a rattletrap benchtop system built of spare parts – my first pair of files were identical! Was it my raw manly McGuyver-with-better-hair tech wizardry, or had I come into possession of cleverly disguised alien technology?

CD playback is far more accurate than many audiophiles like to admit. It’s no fun. If I record a CD-full of data, I can reasonably expect a bit-perfect playback, time and again, until the CD-R begiins to age (admittedly CD-ROM is a different disk format). When I start getting bit errors, the CD isn’t far from dead.

When I hear bout mechanical jitter (and the like) ]perceptibly affecting CD-audio playback, I have to wonder what the heck these guys are playing their CDs on. A Flintstone foot-driven player with “shiny rock” optics? Cheap CD/DVD players actually use commodity-market CD/DVD computer drives for playback.

I’m not the first to raise this argument, and audiophiles have plenty of rejoinders involving the differences between CD-audio and CD-ROM - obscuring the fact that when you rip a CD, it is CD-audio, not CD-ROM. I’ve definitely seen bit errors between successive rips, but if those few bits affect your injoyment, real life must be intolerable. There must be ADA disability for that.

The Unbearable Sharpness of hearing
Science isn’t ready to draw firm thresholds in human audio perception yet, and audiophile perception… well, money can’t buy love or happiness, but apparently you can pick up Steve Austin’s bionic ears for $1395 (plus accessories) at your neighborhood audio ship.

In a mysterious quirk of statistics, up to 3/4 of owners of expensive audio gear can detect differences with the audiological sensitivity of the top 0.1% of humans, even if years of riock concerts have left them unable to respond to their own names. Just ask them - or read the reports/opinions in prosumer mags that quibble over effects at levels that even recording studios don’t worry about.

[What are they listening to, if not studio productions? Are artists being kidnapped to secret governemt labs to record über-CDs for the poor unfortunates whose sensitivities force them to fork over five figures to listen to 1970’s rock that they once loved on AM radios and scratchy 45s? (Five figures would be admittedly cheap for a machine that could make 80s Mmm-bop tolerable) Is it truly possible that any live jazz or classical performance meets their rigorous standards? It’s hard for me to understand how classical music demands a reproduction quality that exceeds what the original artists and audiences had available. That’s not fidelity, that’s alteration, as surely as anything a rapper or remixer does.]

15

Ahhh, hah hah hah ha!!! There is nothing more fun or interesting that trashing audiophiles, and I belive that we may have just arrived at the crux of the argument right there.

It’s no god-damned fun.

16

Jitter doesn’t have anything to do with bit errors or error correction. It has only to do with the timing of the individual samples that go through the D/A converter. Samples that are too closely spaced will result in a steeper curve and samples too widely spaced will have a more shallow curve. If you take a 1V sine wave and make the section around the 0V point steeper and the parts around the ±1V point shallower, you wind up with something more like a square wave.

If jitter occurs with a fixed frequency, it can alter the shape of the analog waves that come out of your D/A converter. Every bit can be correct, but if the bits aren’t timed properly, the resulting analog wave is wrong. Computer CD-ROM is concerned only with each bit being right and in the right order, it doesn’t care when the bits show up.

I personally think the “perfect music forever” slogan in the early days of CD was laughable. CDs aren’t perfect, they are good, consistent, easy to use, but far from perfect.

Jitter from a cable? silly. Jitter from a CD reader that is perhaps imperfectly isolated from a wide variety of electronic gear running at a variety of speeds? perhaps.

17

One of my previous jobs was to transfer customer-supplied media to 8mm data tape prior to mastering in a CD plant. In the old days (although I understand it’s still being done at some facilities), the customer would supply a 1630 master (3/4" video tape format), and this was used to input directly to the Laser Beam Recorder’s encoder, and cut at 1X.
Later, the supplied media - 1630, CDR, DAT, etc. - would be transfered to 8mm data tape (performing error checking during the transfer), and the data tape would be used to input the LBR, cutting at 2-4X.
Still later, CDRs would be used to input to the LBR for a few customers.

One particular customer (a largish label) complained that there was a difference in the sound between the supplied CDR, and the test press we provided.
We had always been satisfied if the bit-to-bit check between the supplied master and the replica was exact, and previously pooh-poohed any contention that there could be a difference in sound quality. “Bits are bits”, we said, scornful of those loopy green-magic-marker-around-the-edge-of-the-disc audiophiles. “It’s a digital transfer and the two compare exactly.”

This time, I performed a jitter analysis of both the supplied CDR (using standard CD test equipment - Koch at that time), and the replica. What I found was that jitter on the CDR was way high - 80 to 100ns from nominal, whereas the replica was very nominal - average jitter being 10-20ns from nominal.
This made sense as the CDR was made on a device that (at the time) probably cost less than $500. The replica was made using a device that cost close to a million.
What was happening was that the incoming (screwed up) bit stream was being reclocked by the LBRs encoder, and the variability of the pit lengths (or the CDR’s equivalent) was being corrected by the much more accurate LBR.

Hmm, said I. We then performed a listening test using our best equipment. I tried to make the test as honest as possible by rigging a way to directly AB to two sources - CDR and replica. We spent quite some time doing this (double blind, identical CD players, blind switching between the two), and our conclusion was there was a difference in sound between the replica and the CDR. The difference was subtle to be sure, but not so subtle that all who listened could not tell that there was some difference between the two. The sound of the replica was better, IMHO - the high end was glassier and had more sheen, the bottom was tighter.

But the upshot was that there was a difference in sound between the CDR and the replica, and we attributed the difference to jitter. After this episode we, and others, investigated ways to ensure that even the jitter was replicated faithfully by screwing around with the reclocking of the LBRs encoder.

With all of that in mind, I would be very suprised if cables would cause nanosecond skewing of the bit stream. The last time that I looked, the high end cable suppliers were touting picosecond accuracy. Picosecond? Rubish. There’s more value in investing in better speakers.

Now excuse me, I have to go immerse my CDs in liquid nitrogen. Improves the sound, you see…