Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
Audio codecs: the green magic marker of the '00s

By raph in Technology
Sat Jul 14, 2001 at 04:00:43 AM EST
Tags: Technology (all tags)
Technology

As a fan of urban legends, one of my favorites is that coloring the edges of a CD with a green magic marker will improve its sound quality. It sounds stupid now, but there were a lot of people who believed it, well after it had been thoroughly discredited. The same forces that propagated this urban legend are working their magic on beliefs about audio codecs.


Let's look at the five reasons cited in the FAQ referenced above, in turn:

The pyschological power of persuasion

In this case, there are a lot of claims made about audio codecs, mostly by the people promoting them, but some from free products. To me, one of the most interesting is that BladeEnc is more accurate in representing high frequencies than other MP3 encoders. While there is some technical basis to this claim (the fact that Blade does not by default use a lowpass filter prior to encoding), it says nothing about the fact that Blade has buggy psychoacoustics and far from optimum bit allocation, which is what really matters.

A more problematic belief is that Vorbis is just not quite as good as encoders for proprietary formats. While there is no technical basis for this belief, there are plenty of expressions of it, including this recent Washington Post article. This "study" was done by a reporter who, while he no doubt meant well, had no idea how to do a listening test. I won't go into the exact details of how he screwed up here. The interested reader is referred to the /. discussion, and pcabx.com for more insight on how to do it right.

The desire for control over technology

The problem is that audio codecs are truly rocket science. Only a handful of elite specialists really understand the detailed guts of psychoacoustic modeling, bit rate allocation, etc. By comparison, CD technology is accessible to most electrical engineers, or any curious person with a decent technical background, for that matter.

There are major differences between different audio codecs. Unfortunately, these differences are hard to understand, and even harder to quantify in any meaningful sense.

People understandably feel lost and confused when trying to evaluate these types of technologies, and are thus especially vulnerable to anyone who sounds authoritative.

The "something for nothing" syndrome

While the title of this certainly evokes the Napster-using hordes, in the sense of the original article, it's not particularly relevant here.

Snobbery

People really want to believe that they're superior to the people around them. Why else would you see so many 320k encoded MP3 files, when serious testing has proven that, with a good coder, 256k MP3's are indistinguishable from the CD in almost all cases?

Greed

This is, unfortunately, the killer. Most of the players in the "next generation" audio codec market stand to rake in considerable money through intellectual property licensing, and control over the music distribution process. Obviously, the exception here is Vorbis. Without such a strong incentive, there is no PR budget to make sure that the word gets out.

It remains to be seen whether the technical excellence of Vorbis, combined with its freedom from intellectual property problems, is enough to overcome all these other forces.

I'm not sure why this issue is pushing my buttons so hard. It might be that there's a sixth factor to add to the original list:

Indifference to scientific method

The quality of an audio codec can be determined by scientific testing. It's not necessarily trivial to do so, but it's not especially difficult or expensive, either. What bothers me is that the vast majority of people don't seem to care.

It is difficult to stand on the shoulders of giants when most people would just as soon push them over and trample them.

Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Related Links
o green magic marker
o discredite d
o BladeEnc
o Vorbis
o article
o /. discussion
o pcabx.com
o proven
o Also by raph


Display: Sort:
Audio codecs: the green magic marker of the '00s | 64 comments (64 topical, editorial, 0 hidden)
Green magic markers (4.62 / 8) (#1)
by fluffy grue on Fri Jul 13, 2001 at 09:55:05 PM EST

Some still believe in the green magic markers, as well as a lot of quackery-filled "acoustical treatments." It's one of the reasons I look forward to the monthly Audio Advisor catalog I get via snailmail - it's chock-full of laughs.

Other things which are getting very popular are "microfacet surface polishes" (to reduce "timing jitter" on your CD reads), "electrostatic discharge sprays" (also to reduce "timing jitter," since as we all know, static electricity refracts light, right?), and various expensive-as-hell little rubber feet which you can get for $1/dozen at Home Depot but $40/each at the likes of Audio Advisor.

Ooh, and eggshell foam as an audio dampening tool - cheap at hardware stores, expensive as hell (at about $100/sq.ft) at audiophile shops.

As long as there are people around who believe whatever bullshit marketing is shoveled at them, "green magic markers" will always be awarded "item of the year" in audiophile magazines.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]

Um (3.75 / 4) (#6)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:34:23 PM EST

Ooh, and eggshell foam as an audio dampening tool - cheap at hardware stores, expensive as hell (at about $100/sq.ft) at audiophile shops.

I see. By the way, care to show me the frequency absorption and diffraction data for Home Depot eggshell foam? What about for 'audiophile' foam? Have you done DBTs to prove they're the same, or have studies? Or are you just laughing even though you've no experience with these products?

[ Parent ]
I haven't done any studies... (3.83 / 6) (#7)
by fluffy grue on Fri Jul 13, 2001 at 10:43:42 PM EST

but these audiophile catalogs never, ever give specs, so as far as I care, neither have they.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

It can depend (3.80 / 5) (#10)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:50:15 PM EST

but yes, a lot of so called 'audiophile' catalogues can get pretty sketchy, and yes, many do overcharge, although I still don't agree that they're selling the same stuff as Home Depot for extremely inflated prices.

On the other hand, it is certainly true that you can get much better prices and value for certain kinds of equipment by looking at do-it-yourself designs, of which Jon Risch's are the most famous. Link here. Indeed, I am making silver cabling based off of a similar design, and braiding my own speaker cable from cat5 ethernet cable (Risch's designs).

All that said, the really reputable places will provide you with at the very least, general temperature vs. dampening data (stuff like Dynamat, used in cars, change significantly depending on the temperature variances of just a few degrees) for absorption panels. They should also be giving you, in many cases, frequency absorption data.

That said, it is true that accurate measurements are rather difficult to do, since they require, for the most part, not only expensive mikes and analysis equipment, but also an appropriate anechoic chamber, which is not only very expensive, but hard to come by period.

[ Parent ]
thanks (4.00 / 2) (#11)
by fluffy grue on Fri Jul 13, 2001 at 10:55:37 PM EST

I'll try to check out that link when I'm at work and can route to xoom/nbci.

Hadn't heard of using cat5 for speaker cable, but it seems like it'd be bad for stereo separation, since it's not shielded (ethernet's wire-level protocol is differential, so it's not nearly as susceptible to noise or crosstalk).
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

Hm (4.00 / 2) (#14)
by fluffy grue on Fri Jul 13, 2001 at 11:37:04 PM EST

After thinking about it a bit, I've realized that analog audio signals are effectively differential too (since there isn't supposed to be a shared ground between the L and R channels and it's the potential difference between the + and - lines which are supposed to make up the signal). Of course, in most stereo amps, there is a shared ground, so it's not really differential, but it might be Close Enough (and of course, this is why monoblock amps are supposed to be so much better than integrated amps).

I'd think (with my limited EE knowledge) that crimping all of the stripe wires together as one pole and the solids together as the other would make for very high-quality signal transmission. The cat5 I have doesn't give any sort of amperage rating, though, so I don't know if just one cat5 cable per speaker would be enough.

I'll have to try it out with whatever cat5 I have left over from wiring my new house after I move in. :) (It even has its own dedicated AV room, which is something I've never had the luxury of in the past. I'm excited about it.)
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

CAT5 current limit? (3.00 / 1) (#30)
by gordonjcp on Sat Jul 14, 2001 at 06:58:35 PM EST

I'd say that even if you only used 1 pair per speaker, normal CAT5 would be "adequate" as a certain famous car maker used to say.
If you really want to make sure that cable resistance isn't affecting your speakers, forget gold-plated, oxygen-free, aligned-grain, unleaded, semi-skimmed macrobiotic speaker cable. Get some cooker cable. The cores are about 1/4" thick, and are rated for 30A.
30A not enough? Go to welding cable...
Or just use 5A mains flex. It works just great. Sounds no different to the 2/metre stuff that $AUDIO_SHOP sells.

Give a man a fish, and he'll eat for a day. Teach a man to fish, and he'll bore you rigid with fishing stories for the rest of your life.


[ Parent ]
actually (3.00 / 1) (#31)
by jovlinger on Sat Jul 14, 2001 at 07:32:18 PM EST

I do remember some audiophile mag testing this. He went down to radio shack and picked up some heavy guage cable... the short of it was that it tested better than 90% of the cables he had lying around. IIRC, only the sooper 'spensive copper tuping cables (that require a special bender to go around corners...) did better. as for the differential signal, is the parent art. by any chance thinking abut balanced cables? IIRC they send the signal both normal and inverted. By inverting one and combinig at the dest, any common effect to the signal is cancelled out... nifty.

[ Parent ]
Lamp cord not sufficient (none / 0) (#43)
by jwb on Sun Jul 15, 2001 at 05:45:06 PM EST

Lamp cord or mains wiring isn't sufficient for audio, unless the wires are kept well apart. In lamp cord, the mutual inductance of the two wires can significantly attenuate higher audio frequencies. I've measured this with instruments, and also found it audible in a simple DBT with three participants. All three preferred the sound of the cheapest Kimber wire to the sound of a same length of home mains wire. None of the participants reliably prefered a very expensive Kimber over the cheap Kimber.

[ Parent ]
CAT-5 (4.00 / 2) (#15)
by sigwinch on Sat Jul 14, 2001 at 12:04:19 AM EST

Hadn't heard of using cat5 for speaker cable, but it seems like it'd be bad for stereo separation, since it's not shielded
It would actually be pretty good for speakers. You don't have to worry about electrical field noise for speakers since the impedance is so low (8 ohms), and the twists would make it reject external magnetic fields. The only issue is that CAT-5's resistance would be largish at the highest power levels. You'd want to use all four pairs in a standard CAT-5 to reduce the resistance: hook the solid colored wires up to audio black, and the striped wires to audio red (or vice versa).

If you wanted to get *really* fancy, you could use *twelve* CAT-5 twisted pairs (three standard cables). This would produce an 8.3 ohm transmission line, making negligible reflections even at the highest frequencies. Whatever you do, all the pairs have to be the same length.

(ethernet's wire-level protocol is differential, so it's not nearly as susceptible to noise or crosstalk).
A speaker is differential too: it only cares about the voltage difference at its two terminals.

--
I don't want the world, I just want your half.
[ Parent ]

Yes (3.00 / 2) (#16)
by fluffy grue on Sat Jul 14, 2001 at 12:27:23 AM EST

I know, thanks :)
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

Tell me (4.00 / 1) (#19)
by onyxruby on Sat Jul 14, 2001 at 02:46:13 AM EST

Have you ever seen a concert hall or a theater with Home Depot egg shell foam? Of course not, there is enormous difference in the products. Your statement is somewhat akin to saying a SCSI and IDE drive are the same thing since they both look the same and are the same size.

Now, being into audio (car stereo) I don't dispute at all that there is a lot of bunk marketing out there (I am certainly not an audiophile). The people involved in audio are far more like computer geeks than most of us (on either side) would care to admit. Half of being an audiophile is being able to seperate the green markers from the cables with extra shielding and grounding. One is well worth the money, one is without scientific basis. If you really want to take up this arguement of yours, the place to do it is Sound Domain. Think of it as being a cross between /. and K5 for the audio world.

The moon is covered with the results of astronomical odds.
[ Parent ]

yes, I know (3.50 / 2) (#20)
by fluffy grue on Sat Jul 14, 2001 at 04:52:07 AM EST

But the crap they peddle in the audiophile catalogs is probably crappy. without any real specs you can't be sure it's not just the stuff from home depot.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

Quibble with article discrediting green pens (4.50 / 8) (#2)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:00:10 PM EST

Anybody who understands how the CD system operates knows that if the data reaching a player's digital-to-analog converters (DACs) is fully and successfully corrected, the digital portion of the entire CD signal chain is operating perfectly -- the data reaching the player's DACs is truly, exactly, precisely, and genuinely identical to the data generated by the encoding analog-to-digital converters.

This, frankly, made me lose my confidence in the writer. I was pretty much nodding my head up until then, but after reading that quote, I simply read the rest with a few giant boulders of salt, as it was obvious that the man had no good idea of what he was talking about. Here's why:

according to the author, as long as the data is transferred to the DACs correctly (1's are 1's, 0's are 0's), the digital portion of the chain is "operating perfectly," to use his phrase. What the author has failed to take into account, is jitter: temporal inaccuracies in the pattern in which 1's and 0's are delivered to the DAC.

Thus, sending the correct bits to a DAC is by no means a measurement of the perfect CD player digital portion. There is a good reason why a Marantz 6000OSE cd player costs me $550, and a Mark Levinson 31.5 cd player sets a well-to-do audiophile $10,000 -- the latter has far less jitter (temporal error) than the Marantz, even though both may be able to output the exact same bits. This "jitter" causes temporal smearing of sound, and can cause loss of high-end detail, among other detrimental effects.

Mind you: this is a problem that has been documented, DBT-ed (successfully) and measured. Hi-end CD players now regard the reduction of jitter as one of the most pressing problems of extracting higher quality from a CD. The author, in this case, was making assumptions about performance based on his understanding of the factors. Sadly, he obviously did not realize all the factors, and his blanket statement made a red flag go up in my mind, at least, which told me that the guy does not know what he's talking about.

I hate to generalize this to his discussion of green marker pens in general, but the take-home lessons remains the same:

No measurement differences does not mean that no differences exist. It may mean that there are no differences, but just as well, it could mean that you're measuring the wrong things.

Also: the author admits that he performed uncontrolled listening tests, told us nothing about said listening tests, told us nothing about the upstream equipment (including his CDP) or his speakers (I'm sorry, but the fact that your Bose set can't articulate the differences tells us nothing). And uh, based on this, I am supposed to believe this guy why? Because he's a famous writer?

The author goes on to concede that any differences are not likely to be obvious (duh -- audiophiles don't spend multiple tens of thousands of dollars for obvious improvements, but incremental ones), and then says that proponents of green pens have performed no DBTs. (He completely misses the irony of his own method of testing, note.)

The author goes on to describe his testing (note that he does not tell us why his tests are relavent), and then further re-iterates that the data that reaches the DACs, if they are the same as what's on the CD, necessarily means that there are "no differences."

Unfortunately, the fact that David Ranada was a former Technical Editor of Stereo Review and High Fidelity really frightens me, for just the reasons I've explained above.

Lastly: links to jitter:

Here

and Here

as well as Here

Feel free to ask if you want more links about jitter...

Dude (4.25 / 4) (#3)
by fluffy grue on Fri Jul 13, 2001 at 10:13:17 PM EST

Did you even read the second link? It says, quite clearly, that timing jitter doesn't affect single units (such as a single CD player), and it only comes to play when you're synchronizing a whole bunch of serial-digital devices (such as DATs and other things which speak S/PDIF).

Also, if the bits were received out of order, you'd hear a lot more than just "faint differences" - you'd hear clicks and pops everywhere. And, contrary to what the guy is trying to say, bits are, in fact, bits.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

I think (3.00 / 2) (#4)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:27:21 PM EST

that the author was discussing cases when a CD player is outputting an analog signal, in which case (at least by his assertion) the units are coordinated inside in such a way that the DAC is coupled to the digital source well enough that jitter may not be a factor. (this is because if the DAC is internal, it runs off the same oscillator -- Bob Katz's article -- link @ bottom).

Obviously, this does not apply to the digital signal that a CDP outputs, which, if it's on its way to a speaker, must necessarily go to another device to be converted to analog before it reaches human ears.

But either way you read this, my point is still valid: that bits are not just bits, and that jitter matters. I will also note that this is not just a matter of hi-fi systems, where people have external DACs which are connected to CDPs digitally, but also for lots of other people, who have digital cables running from their CDP, to their receiver's digital in. The fact remains that digital signals go from one unit to another, and that they are time sensitive. It's too bad that author, who tries to refute green pens makes no acknowledgement of this is while maintaining "bits are bits" -- it makes him look silly, and uneducated.

[ Parent ]
Time sensitive (4.50 / 4) (#8)
by fluffy grue on Fri Jul 13, 2001 at 10:48:39 PM EST

I agree that jitter matters. However, this is why there's PLLs to latch onto the source timing signal. It's not like the DAC generates its own clock. Also, it takes a LOT of noise in order to make jitter signifigant even to relatively cheap hardware.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

confused (4.00 / 3) (#32)
by jovlinger on Sat Jul 14, 2001 at 08:04:34 PM EST

I must admit that I am somewhat confused as to why/how jitter can matter, in the general case. How can bits not be perfect and robust? That's what makes a digitial signal, well, digital.
Not that I'm claiming you're mistaken, but rather that I'm fundamentally confused.

So let me explain my confusion. There are three steps to the process of playing a CD:

1) extract data from CD to buffer
2) stream data from buffer to DAC.
3) analog cleaning up of the signal to make it suitable for amplification.

Now, we know that 1 is cheaply perfectable, in the 100% accurate no audiophile can deny sense of the word, as my computer is able to read CD ROMS and CDs. Any CDROM that can extract digital audio at at least 1X is able to keep the buffer full of PERFECT data. A super cheap player maybe has too small (or no) buffer, so hence may have buffer underruns. but that is not likely on non Wallgreens/CVS equipment.

that leaves 2. A cd has a 160KB/s data stream. So maybe keeping that rate is hard. But my $29 100baseT card deals with easily 100 times that rate, so I think consumer grade clocks are up to the task.

and 3 is analog, so that aint it.

So I am stumped. I can't figure out where jitter would live in the system, in all but the absolute cheapest equipment.



[ Parent ]
Before the bits (4.00 / 2) (#36)
by fluffy grue on Sat Jul 14, 2001 at 08:47:12 PM EST

Jitter occurs before the bits are read in. They're not stable and robust when they're pulses of reflected laser light whose rate is being controlled by a (very analog) motor. There's moving parts involved, as well as media with all sorts of other potential problems such as a slight lack of alignment (the center of the "grooves" on the CD being off from the axis of rotation by even a few microns) and dirt and the power draw caused by the other parts in the system (which leads to minute fluctuations in the motor speed).
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

ok, (5.00 / 1) (#40)
by jovlinger on Sun Jul 15, 2001 at 11:37:54 AM EST

thanks for the clarification, but I am still confused. I thought:

1) small scale timing errors are caught by circuitry that uses checksumming to re-read sectors/packets/chunks/whatever.

2) The checksum passed and extracted data is fed into a buffer.

3) the buffered data is dealt with somehow (in the case of a computer, sent down the bus; in the case of a player, sent to the DAC).

So I'll grant for jitter having an effect on the first stage. However, doesn't the second stage hides any low-level misreads by flutter or dust?

That leaves the only two scenarios that I can construct for jitter being noticable (in the "detectable from the outside" sense):

1) the buffer is so small that any trivial error in reading the disk results in buffer underruns

2) the checksumming algorithm fails to catch incorrectly read data.

Both of these cases have a very simple counter-proof: my computer's commodity cdrom drive can repeatably and accurately read discs. I strongly suspect it doesn't outperform a dedicated transport, leaving me to suspect it does just as well -- ie both perfectly extract data from CDs.

Let me reinforce argument with an absurd gedanken experiment:

I present to you Johan's incredible DIY PerfectTrasnport. It consists of a found computer with a CDRom drive, 700Mb of RAM, and a sound card with optical out. Total cost of about $500 dollars (mostly for the RAM). We know it can perfectly extract the raw data bits from an audio CD (easy test: do it twice and compare). It stores them all in RAM, and then streams them to the optical out at 44.1 Khz.

So how does this gedanken transport have any jitter? It effectively has NO MOVING PARTS!

Obviously, audiophiles don't pay 20 times that for a snazzy box that does the same thing, so how is my transport worse than theirs? (appart from my need to read-ahead a whole disc before playback begins)


TIA for explanations


[ Parent ]
Exactly (none / 0) (#42)
by fluffy grue on Sun Jul 15, 2001 at 05:44:19 PM EST

Jitter doesn't matter, for exactly the reasons you gave. :) I'm not saying that jitter matters, just acknowledging that it exists (I've never said that jitter doesn't exist, just that it doesn't matter except in extreme circumstances).

BTW, most newer CD-ROM drives have an S/PDIF port on them, and it shouldn't be too hard to make a minimal IDE controller which just sends playback commands. You could probably make a Pretty Good (and small!) coax-only CD player for about $50 (and optical is just S/PDIF with a 75ohm resistor and an LED).

And the reason audiophiles pay a lot of money is because audiophiles pay a lot of money.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

Oh (none / 0) (#44)
by fluffy grue on Sun Jul 15, 2001 at 06:48:50 PM EST

And before anyone jumps all over me for saying that jitter doesn't matter when I said that jitter matters above, I'd like to point out that above I said that jitter matters but the PLL takes care of it. That is, jitter does matter on the path between the sensor and the DAC, but there's already plenty of circuitry to compensate for that except in the most extreme cases.

That is, jitter matters to the circuitry but it doesn't matter to the listener, because the circuitry already accounts for most jitter.
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

CD-ROM vs CD-DA (none / 0) (#45)
by mbrubeck on Sun Jul 15, 2001 at 08:45:26 PM EST

Both of these cases have a very simple counter-proof: my computer's commodity cdrom drive can repeatably and accurately read discs. I strongly suspect it doesn't outperform a dedicated transport, leaving me to suspect it does just as well -- ie both perfectly extract data from CDs.
Be careful to note that CD-ROM data tracks have an extra layer of CRC error correction beyond the CIRC coding used on CD-DA (digital audio) discs. This takes extra hardware - often an onboard processor - which is one of the reasons that computer drive mechanisms are more expensive than their consumer electronics counterparts. The reason is that a single bit error once every dozen or hundred discs is perfectly acceptable for audio (the error will be detected and replaced with a linear interpolation), but it is unacceptable for data.

[ Parent ]
Ah, not quite... (5.00 / 3) (#48)
by MrSmithers on Mon Jul 16, 2001 at 12:22:16 PM EST

Now, we know that 1 is cheaply perfectable, in the 100% accurate no audiophile can deny sense of the word, as my computer is able to read CD ROMS and CDs. Any CDROM that can extract digital audio at at least 1X is able to keep the buffer full of PERFECT data. A super cheap player maybe has too small (or no) buffer, so hence may have buffer underruns. but that is not likely on non Wallgreens/CVS equipment.

It may seem that way, but in reality CDs are much more complicated than simply a stream of "perfect" bits. It depends on the format the disc is recorded in.

Physically, along with the pits that represent the 1s and 0s of the digital data, CDs (Red book) have some basic timing and error-correction information encoded. All drives, whether audio players or CD-ROM drives have built-in circuitry to decode and process this data. I am not sure of the exact format and encoding of the timing and CRC signals; perhaps someone more knowledgable in that area can jump in and fill in the blanks.

Once the timing and CRC information have been applied and removed, what is left is a series of 2352-byte frames. Each frame is raw PCM data (44.1khz, 16-bit, stereo) and is 1/75th of a second. A simple calculation can confirm this: 44100 *2 *2 / 75 = 2352. Most (except cheap-o discman knockoffs :) can stream these frames with little to no problem. Seeking, however, is another matter. The timing information is not accurate enough to seek to a particular frame. I am unsure of the exact resolution (perhaps nearest second). More advanced hardware can extrapolate the position of the disc with timing delays to get closer to the actual position (some are very good at this, but if the head gets slightly out of calibration it can impair it).

Data CD-ROMs are written using a special format known as Mode 1 (I forget which "color" book it's from). This format adds a 16-byte header containing positioning data, as well as a 288 byte trailer that holds additional error correction data. The positioning data helps to solve the seeking problem -- if a drive is asked to seek to a certain LBA and the laser misses, the drive can read the header data and figure out if it sould just keep reading until it finds it or try again. The extra CRC information is to help out since computer data is much more sensitive to single bit errors than audio. Thus, there are 2048 bytes per frame left for actual data.

This is the reason that a CD-ROM can only hold 650MB of data, but a full 74-minute CD ripped to WAV files will take up around 746.9MB of disk space. There is also Mode2-XA which has the extra positioning data but no extra error correction (a lot of Playstation games use this for background music and such). Mode2 allows 2336 bytes per frame for arbitrary data storage. Modern CD-ROM drives have hardware to take care of the Mode1 error correction, but most can be told to read the entire raw frame (this is used for digital audio extraction, among other things). CD Burners usually take mode1 data as input, but most can handle mode2 and raw data (usually only in disk-at-once mode).

Anyway, I digress. The cause of jitter is actually the lack of precision timing information on Red book audio discs. This isn't as much of an issue in older drives which simply read continuously (streaming), but many players with skip-protection use multi-speed CD-ROM style drives and memory buffers, so they have to seek some. Jitter can also become an issue even in continuous reading drives if the head gets out of alignment. Raw mode (2352 frame size) data access on CD-ROM drives does not guarantee that you will get exactly the data you asked for -- it may be offset by a few frames in either direction.

Most CD audio ripping programs (the good ones at least :) implement jitter correction in software by means of read overlap. They will request as many sectors as they can from the drive (limited by the read buffer of the hardware), but will overlap the reads some and compare the data to make sure it is aligned. If the overlapped data doesn't match up, it will attempt to figure out which direction the jitter was in and compensate, or re-read the sectors as a last resort (usually referred to as jitter errors). I assume that high-end CD players do something similar in hardware, and that is one of the many reasons they are more expensive.

I hope this helps clear up some of the jitter issues. As usual, take anything you read with a grain of salt -- I think I know what I'm talking about but there's a possiblity I'm dead wrong :)  If anybody has more information on the subject, please contribute and help enlighten all of us.



[ Parent ]
very helpfull, thanks! (3.00 / 1) (#51)
by jovlinger on Mon Jul 16, 2001 at 09:33:31 PM EST

Thanks, that cleared up much of my confusion, or at least lets me speak about it using precise terms. On the whole, tho, it seems that your definition of jitter is on macroscopic/intermittent level (frames arriving at the wrong offset: you would hear a chirp/tick/scratch on playback), as opposed to the microscopic/recurring, where you might not notice it without golden ears.

So basically we all (I'm including Fluffy, who kindly replied above) agree that jitter may exist internally, but even a hobbyist can build a transport from surplus computer parts that elimiates it to an ARBITRARILY high degree (by repeatedly ripping the track until it gets it right, for some statistical definition of right; bit for bit equivalent n times in a row?).

Makes you wonder about the people who buy multi $K transports. Not that I'm against nice audio stuff (the B&W Matrix nautius 801 speakers WILL be mine, but until then I'm saving for the Gallo round sounds, as my Opus 3's (anyone else even HEARD of those? Made of concrete, a bitch to move, but GREAT bass definition) are getting long in the tooth), but it's just VULGAR to spend that much on a pretty box and a name.

also, I wouldn't mind one of VanAlstine's dynaco rebuilds; the price is right, and they just look so cool!

[ Parent ]
Jitter (5.00 / 2) (#52)
by sigwinch on Tue Jul 17, 2001 at 04:07:51 AM EST

So basically we all agree that jitter may exist internally, but even a hobbyist can build a transport from surplus computer parts that elimiates it to an ARBITRARILY high degree (by repeatedly ripping the track until it gets it right, for some statistical definition of right; bit for bit equivalent n times in a row?).
That's usually *not* what is being talked about when people refer to digital audio jitter. I would call the missing data problem, well, the missing data problem (and I would expect it to produce horrifically bad sound, although maybe it can be interpolated well enough to not suck totally).

What is usually worried about is timing jitter on the digital values. The CD reader produces a continuous stream of numbers. Half the numbers are for the left speaker, half are for the right. For simplicity, I'll just consider one of the speakers.

Each one of those numbers represents a voltage that is to be driven across the speaker wires for a short period of time. A particular number is turned into a voltage by an analog-to-digital converter, that voltage goes to an amplifier that makes it much bigger, and the big voltage goes to the speaker. The speaker position depends on the voltage, and so is determined by the number. Successive numbers are applied to the system one after the other, causing the speaker to move and causing sound to be generated.

Ideally, the numbers will come out of the CD reader at a constant rate (if I remember right, CDs are 44 kHz, which is a number every 22.73 microseconds). Unfortunately, the real world is not ideal. CD readers are mechanical devices and are subject to the usual vagaries of gadgets with moving parts. The disk speed fluctuates, the circuitry has to do varying amounts of error correction, the crystal timebase varies, etc.

The net effect is that the numbers coming out of the CD reader come out at varying rates. Sometimes they come a little faster, sometimes a little slower. Now, this variation isn't enough to screw up the digital interpretation. If it's supposed to be the sequence 4, 5, 6, the analog-to-digital converter will still see 4, 5, 6. It will just see them at slightly wrong times. The effect of this jitter is to cause frequency modulation of the audio signal, which is undesirable. (I don't know how audible it is in practice, but nevermind that.)

The naive (and cheap) way to build the analog-to-digital converter is to feed the numbers directly from the CD reader to the analog conversion circuitry. If you do that, whatever timing jitter is present in the mechanical parts will carry directly into the sound. You can improve the jitter with flywheels, fancier motors, and shock absorbers, but it will still exist. Unfortunately, mechanical solutions are not totally effective and are rather expensive.

The correct solution is to put a FIFO (first-in first-out) buffer in front of the analog-to-digital converter. The input port of the FIFO accepts data from the CD reader as it arrives, and stores it. The output port of the FIFO sends data to the analog-to-digital converter at a constant rate.

Which leads to the next question: how do you clock the FIFO output? If you clock it in sync with the CD reader, you won't have done anything for the jitter. If you clock it with an independent quartz crystal, the crystal will gradually fall behind or work faster than the CD reader; the FIFO will run out of data or overflow. Either way, distortion will have been introduced into the sound.

The clocking solution, as fluffy grue mentioned, is a PLL (phase-locked loop). A PLL is a type of oscillator that produces a clock with the same average frequency as another clock, but with the short-term variations filtered out. With a FIFO + PLL, the analog-to-digital converter gets samples at a constant rate (in the short-term), and the timing jitter is eliminated.

I honestly don't understand why the listening-only audiophiles are so freaked out by jitter. It takes a trivial amount of circuitry to eliminate. Every portable CD player that has 'skip protection' has very good jitter elimination built in. Building a receiver/amplifier with digital inputs that has jitter elimination is dead simple. I mean, if they can put it in a $50 portable player that runs 6 hours on batteries, an $600 receiver ought to be trivial, right? I guess I can understand why artists who are merging different audio streams that are at slightly different clock rates might be nervous, but even that is a fairly simple problem.

I guess I can understand, though. If I had a choice between selling $0.75 worth of PLL/FIFO, or a $6000 rig that floats on air, I'd probably pick the latter. Profit margin, profit margin, profit margin.

--
I don't want the world, I just want your half.
[ Parent ]

Good info (3.00 / 1) (#58)
by MrSmithers on Tue Jul 17, 2001 at 10:08:01 AM EST

Thanks for the info, I didn't realize that there were multiple definitions of jitter. I had thought about possible timing slips if the clocks within the player weren't synchronized properly, but didn't know what to call such a phenomenon.

Just a nitpick, CD players use DACs (digital-to-analog converters). ADCs are used for recording.

[ Parent ]
Good grief! (3.00 / 1) (#60)
by sigwinch on Tue Jul 17, 2001 at 11:28:23 AM EST

Just a nitpick, CD players use DACs (digital-to-analog converters). ADCs are used for recording.
Boy do I feel silly. Here I am a EE who has designed both DACs and ADCs onto boards, and I go and reverse them. **shakes head**

--
I don't want the world, I just want your half.
[ Parent ]

Obsolete issue (5.00 / 3) (#63)
by wnight on Wed Jul 18, 2001 at 02:45:06 PM EST

In the CD -> DSP, jitter is a solved problem. In the DSP -> DAC, jitter is a solved problem.

Any semi-modern CD player will have a buffer of at least a frame, this is usually enough time at 4x or higher (which most modern players run at) to re-read a frame in case of error, before having to send bad data.

Because of this oversampling (which is to reduce errors while reading, not jitter) the CD player buffers data, and this buffer seperates all the mechanics from the data output.

Once a signal reaches this buffer, it IS just 1s and 0s, regardless of the jitter in the reading process, the stream of digital data is identical.

This data is passed from the player's buffer into either the DSP (is the device supports any sort of processing, which almost 100% do, even if it's just pop-removal in case of a missed frame) and the DSP then buffers the data to send to the DAC at least a frame ahead of time (in order to be able to examine frame 1 and 3, in case 2 is missing, to create a best-fit replacement for 2).

The DAC then reads this in a buffer, probably at least eight bytes at a time.

All of these data transfers are bursty, you get a frame, wait, get a frame, wait, etc. As such, if a device falls a few cycles behind another device it doesn't result in a missed bit, merely in a slightly shorter wait between data transfers.


Only the very oldest players ('80s) would try to pass data directly from the reader to the DAC, and even then they did it from a buffer (needed, to use the RS ECC codes). If the clocks got out of sync, a bit could be missed, but that wouldn't be a common thing.

The company I work at makes a device that talks to it's other half over a (at max) 100M cat-5 cable. The device syncs clocks at start and the clocks drift enough (on average) for the non-reliable protocol (ie, no error checking or ACK messages) to slip one bit and drop a 5ms audio frame aproximately every 7 hours. (Depends on the actual rate of the two clocks, could be between 4 and 10, or so) That's how controllable jitter is between widely-seperated devices, both completely unshielded, over unshielded cable.

Between a DSP and DAC on the same board, running from the same clock?

You might drop a bit every year or so.

Jitter is just a boogeyman for audiophiles. It's a way that companies can sell $100/30cm cables for a digital signal. It's not in the best interest of companies to try to disprove it, and quite frankly, I doubt most audiophiles have a clue about running unbiased testing to detect any theoretical differences.


In modern audio, it's 99.5% in the analog stages. The amplifier and the speakers are 90% of that. Cables and decent filtered power (run the stereo on a UPS) are the remaining 10%, IMNSHO.

(Disregarding, of course, any recording or mastering issues, and the limitations of 16b/44kz audio.)


[ Parent ]
Jitter (none / 0) (#57)
by MrSmithers on Tue Jul 17, 2001 at 09:58:13 AM EST

Yes, you're right -- it would seem that there are multiple definitions of the word "jitter". One for the seek problem (on multi-speed drives that use digital buffering), and another to describe timing slips between the read rate of the laser and the clock on the DAC...

The first is easy enough to fix. There is a reply below your message that describes a method to get around the second, though IANEE (I am not an Electrical Engineer :), so I don't know how much it would cost to build such a system.

[ Parent ]
Other question: (3.25 / 4) (#5)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:31:46 PM EST

Did you actually read all the articles? Wow, fast reader!

Also, if the bits were received out of order, you'd hear a lot more than just "faint differences" - you'd hear clicks and pops everywhere. And, contrary to what the guy is trying to say, bits are, in fact, bits.

Have something to back that up, other than 'fluffy grue says so'? By the way, who said out of order, anyway? I know for a fact that I made no statements to this effect. I merely spoke about timing inaccuracies, not bits arriving out of order. In any case, you *do* get clicks and pops if jitter is bad. See: this article.

[ Parent ]
Yes (3.66 / 3) (#9)
by fluffy grue on Fri Jul 13, 2001 at 10:49:40 PM EST

I did actually read all the articles. Well, not the one on xoom, since I can't route to xoom, and pages on free webspace providers aren't exactly the bastion of accurate reporting anyway.

It's fairly easy to show that you'd hear huge clicks and pops if bits arrive out of order. Say you have a sample which looks like this: 01000000 00000000. If the first bit arrives early, you get this: 10000000 00000000. The resulting voltage is twice as much.

For a more extreme example, let's say you have two bits near 0. 00000000 00000001 followed by 00000000 00000000. Let's say the 1 arrives a little late. Then the next sample appears as 10000000 00000000. i.e. you get a value of -32768 instead of 0. A slight difference, that.

You didn't say that the bits arrive out of order, but all of the jitter freaks say that they do.

What do you mean by "timing inaccuracies" if you don't mean bits arriving at the wrong times? And typically, the DAC has to wait for the entire sample to be loaded into its memory before it can actually convert from digital to analog, and typically the player has a large enough of a buffer that a couple of samples being delayed won't make a difference since it can just catch up later.

The link you just gave shows an extreme case of jitter, yes, and it's not the case which audiophiles are always talking about eliminating. And regardless, a green magic marker or a static cling discharge won't help with that - if you're getting that kind of jitter, either your disc is dirty/scratched or your player is dying.


--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

Agree (3.00 / 2) (#12)
by Anonymous 7324 on Fri Jul 13, 2001 at 10:56:49 PM EST

yes, extreme jitter causes bits to arrive out of order. But, just simple timing inaccuracies can still cause degredation.

If the first byte comes at time 1.0 (, and the player expects a byte at time 2.0, and the digital source waits until 3.0, the receiving end sees no voltage swing for time 2.0, and decides that the byte "time 2" is still whatever the last bit for "time 1" was. Thus a problem, even though the bits continue to arrive in order, but late.



[ Parent ]
Okay (4.00 / 2) (#13)
by fluffy grue on Fri Jul 13, 2001 at 11:12:59 PM EST

That's still a pretty extreme case though, and if that sort of thing is happening, then chances are you'd be hearing other things too. Also, again, the buffer in the transport should take care of that sort of thing as long as it has at least a few samples' worth of memory.

BTW, I just discovered that although nbci.com and xoom.com are the same site, they're on different IPs, and I can actually route to nbci.com. So I'll go back and read the sites on xoom by a simple URL substitution. :)
--
"Is not a quine" is not a quine.
I have a master's degree in science!

[ Hug Your Trikuare ]
[ Parent ]

The clock isnt really part of the digital chain... (4.00 / 4) (#17)
by MfA on Sat Jul 14, 2001 at 12:40:04 AM EST

The clock jitters, and that jitter is determined by analog electronic effects. If the right bits reach the DAC then the digital part has done all it can (the mechanical parts too BTW, its entirely arbitrary as long as they read the CD correctly... stabilizing CD players is another one of those things in which you can recognise somoene who has gone to the wacky side of high end).

Its a pity consumer audio doesnt support source clocking, if external DAC's could just provide a clock to CD players they would become fairly inconsequential for highend... any old CD player would do. The quality would be concentrated where its needed, with the DAC.

[ Parent ]
Temproal Jitter (3.75 / 4) (#18)
by delmoi on Sat Jul 14, 2001 at 01:26:49 AM EST

Hrm. I'm a little confused here. If the data were put into a small buffer, say a couple milliseconds, and then fed into a DAC at exactly 44.1khz, wouldn't this solve the problem?

What in gods name did they put into that box to make it cost $10,000?
--
"'argumentation' is not a word, idiot." -- thelizman
[ Parent ]
He just THINKS it sounds better :) (3.50 / 2) (#23)
by dgwatson on Sat Jul 14, 2001 at 11:54:20 AM EST

After all, no sane person would pay $10K for a CD player unless it really does sound better, right?

And even if it doesn't sound any better, he wants it to sound better (since he paid so much), so he hears it as sounding better.

I tend to think all this audiophile stuff is just a scam...

Now, I'm sure that the more expensive equipment does sound slightly better, but there's NO way it sounds 100 times better.

[ Parent ]
Read the post you're indirectly replying to (4.00 / 2) (#33)
by reeses on Sat Jul 14, 2001 at 08:16:13 PM EST

Now, I'm sure that the more expensive equipment does sound slightly better, but there's NO way it sounds 100 times better.
I guess you missed where the grandparent poster said:
duh -- audiophiles don't spend multiple tens of thousands of dollars for obvious improvements, but incremental ones
This is called the law of diminishing returns, felt most acutely in the audiophile arena.

Almost any consumer good can be used as an analogy. Going from the very low end in cars (say, a <$10k car, if they exist) to an upper-low-end car (say, $20k), will get you such a huge suite of improvements, that you could easily argue that the more expensive car is possibly twice as good. The things that make it better are usually obvious, with some notable exceptions. However, going from a $20k car to a $40k car, there will still be a large number of improvements, but it will be hard to argue that the more expensive car is twice as good. This quality or performance increase is even smaller going from $40k->80k, $80k->160k, etc. For similar segments of automobiles, it takes increasing discretion to appreciate the gains. Unless you're a fabulous driver, your M3 is going to perform as well as a Ferrari, possibly better, because the Ferrari is less forgiving of inexperienced manipulation.

The same can be said for something as simple as speakers, a technology that has been produced for several decades. Cheap $100 Radio Shack speakers sound ok, but $200 speakers generally sound twice as good. $400 speakers sound better still, and are near the top end of the spectrum that can be discerned by inexperienced users. $800 speakers are probably around the bottom end of what is acceptable to a self-respecting audiophile, (apart from notable exceptions, like the Axiom and NHT products, for special applications) just as a performance car enthusiast isn't going to waste too much time below the M3, even though your Eclipse might make you perfectly happy zipping around the highway. $1600, $3200, $6400, and even $25.6k speakers push you into the high-performance range, where room geometry, wall construction, power supplies, suspension systems, and interconnects might actually matter, just as your M3 can handle midgrade gasoline, where your Ferrari would cry if you forced it to choke down anything less than premium.

I find myself right around the top of the inexperienced-listener category. I'm perfectly happy with my NAD equipment, because I have a fairly noisy home, and I just can't hear the difference to make it worth spending more for something like a Naim, let alone Mark Levinson, but I could hear the difference enough to keep myself from spending less. Besides, I like the humor potential in having "NAD" stenciled on my audio gear. :-)

I'm not sure what the equation for audiophile returns on "investment" is, but it's definitely logarithmic.

[ Parent ]

I understand diminishing returns (2.50 / 2) (#35)
by dgwatson on Sat Jul 14, 2001 at 08:33:47 PM EST

But what I was saying is that in my opinion a $10K CD player is probably just a rip-off. I mean, you can get CD-ROM drives that do jitter correction for just a few hundred, so there's no way that a CD player should cost $10K. With speakers and amps I can understand, but CD players? Come on...

[ Parent ]
I agree (none / 0) (#62)
by reeses on Tue Jul 17, 2001 at 03:58:16 PM EST

Not only is there no way I could justify paying more than $500 for a CD player (transport AND DAC :-)), I'm perfectly fine buying the $500 CD player now, and waiting five years, when the next $500 CD player I buy has most of the improvements that the previously-$10k CD player offered.

Most of the claims for high-end CD players fit the stereotypical audiophile nut mold: "The sound is warmer," "It's more expressive," "It retrieves more information," etc. You don't see a lot of verifiable quantitative analysis of the improvements, and my ear isn't trained enough to pick them out.

[ Parent ]

Buttering out jitter (4.25 / 4) (#29)
by gordonjcp on Sat Jul 14, 2001 at 06:50:18 PM EST

CD players do have a buffer, in the order of a few milliseconds. This allows the disc speed to fluctuate slightly without noticeable effects. CD has no measurable "wow and flutter", unlike analogue recordings because of this buffer.
The problem comes about when you try to read the buffer at a constant rate - the clock has to be extremely stable, and certainly in older players, the clock wasn't particularly good in this respect.
Personally, I suspect that even the cheapest, nastiest, 5-per-unit-in-bulk OEM CD mechanism is probably just as stable as a good semi-pro deck by now - the cost of good stable crystal oscillators has come down steadily over the past few years. It's the analogue bits after the converters that make the difference. The clock signal has to be pretty hairy before it makes an audible difference.
Of course, I speak as an electronic engineer, who spends all day listening to broadcast and recording equipment. I'm not an audiophile. Quite possibly, I'd hear something different if I'd spent 10,000 on a CD player, and fitted it with a gold-plated mains plug...

Give a man a fish, and he'll eat for a day. Teach a man to fish, and he'll bore you rigid with fishing stories for the rest of your life.


[ Parent ]
Wow, jitter is real (5.00 / 3) (#21)
by raph on Sat Jul 14, 2001 at 05:31:52 AM EST

When I first read your post, I was a bit skeptical, but after reading your links, it is clear that jitter is real. It can definitely make a noticeable difference in sound quality. What's more, the effect that individual components have on quality when combined into a system can be quite unpredictable.

You're absolutely right that the author is wrong to espouse the "bits is bits" mentality. Two digital signals can contain exactly the same bits, but differ substantially in quality.

It's fascinating that jitter took such a long time to come to light. It seems like a modern understanding of jitter didn't come to light until around 1993. I'm sure that part of the reason is that jitter straddles the categories of analog and digital. When designing purely digital systems, as long as you're inside tolerances, you just don't care about jitter. Conversely, the responsibility for jitter lies almost entirely outside the analog chain - it's a characteristic of the digital signal.

The reason why system performance is so unpredictable is that various components can have quite different properties of passing jitter through.

Consider CD player A with picosecond jitter, player B with modest jitter at 20kHz, and player C with modest jitter spread over the spectrum.

Now consider DAC X with a very high quality PLL to essentially "resynthesize" the clock signal with a minium of jitter, DAC Y with a PLL with a cutoff frequency somewhere in the audio band, and DAC Z that essentially passes the jitter through.

Here's what quality looks like, on an arbitrary 1-9 scale:

- | A B C
-+----
X | 9 9 9
Y | 9 8 5
Z | 9 1 4

Thus, the owner of DAC Y will get better results from player B than player C, while the owner of DAC Z will see the opposite.

One thing I found interesting about the StereoPhile link is that the Audio Alchemy DTI, which ostensibly reduces jitter from a digital signal, in many cases actually made it worse, especially when dealing with higher quality sources.

Now that jitter is much better understood, I think it's likely that you'll see excellent performance on all non-crap equipment. Unfortunately, sorting out the crap from the non-crap is not exactly trivial. I learned this the hard way when I got a Creative AWE128 sound card in my last PC - it has noise problems all over the place, and I wouldn't be surprised if it had jitter problems too.

It is interesting that the really expensive gear bought you that excellent performance even before jitter was widely understood. Quality always makes a difference, even if you don't understand exactly why.

[ Parent ]

It might not even be your sound card's fault (4.00 / 4) (#38)
by Trepalium on Sun Jul 15, 2001 at 02:59:31 AM EST

Virtually anything in a system can be responsible for such noise. Everything from the power supply (and it's rated wattage), motherboard, drives, expansion cards, case, monitor, keyboard or mouse. Using virtually any device in or connected to a computer system can cause the power supply to dip, or introduce noise. Adding capacitors can help normalize the power, but only to an extent. Unlike most stereo equipment, A PC isn't a closed system where interference can be easily minimised. Virtually everything's a problem here -- crosstalk from cabling or cards seated too close to the audio card, electromagnetic noise that some components may be emitting. It's a wonder you can get mediocre sounding audio out of a PC at all!

It is interesting that the really expensive gear bought you that excellent performance even before jitter was widely understood. Quality always makes a difference, even if you don't understand exactly why.
I imagine it depends on if that equipment was professional equipment, or audiophile equipment. The professional equipment is often very expensive, but also rationally designed. Audiophile equipment is also expensive, but usually designed around whatever is the "big thing" is today. Quality does matter, but it's usually better when you can pin it down to something in particular. The people who buy either really cheap crap, or those who actively seek the most expensive thing they can find are often the people who get burned the worst. In the first case, they usually end up replacing it because it doesn't perform very well, and in the second case, the person usually ends up constantly trying to rationalize why he or she paid so much for so little. It's sad that so many people still make this mistake.

[ Parent ]
PS Noise (none / 0) (#53)
by sigwinch on Tue Jul 17, 2001 at 04:14:29 AM EST

Virtually anything in a system can be responsible for such noise. Everything from the power supply (and it's rated wattage), motherboard, drives, expansion cards, case, monitor, keyboard or mouse. Using virtually any device in or connected to a computer system can cause the power supply to dip, or introduce noise.
Exactly. That's why you often see DC-DC converters on good instrumentation cards (such as the ones National Instruments makes). They can't count on a stable power supply, so they make their own.

--
I don't want the world, I just want your half.
[ Parent ]

Doubtful (4.66 / 3) (#47)
by JonesBoy on Mon Jul 16, 2001 at 11:57:22 AM EST

A sound card is placed in a box with dozens of high frequency lines bouncing all over the place, unstable power, and virtually no shielding on anything. No kidding you had noise. Your sound card is also driven by a crystal oscillator when playing a file. No substiatial jitter there. Attributing the lack of quality on a sound card to jitter is like blaming a bad 0-60 run on a dirty windshield. If you are playing a CD, its all analog coming from the CDROM. Blame that, the cable to the sound card or the cheap final drive on the audio board, but not jitter.

Telling me that nobody knew about jitter until '93 is an insult to anyone working on telecommunications equipment.

Your CD player is in no way depending on the rate that data is being pulled from the CD to determine the DAC speed. Anyone telling you that is a crackhead. Data on a CD is not stored linearly, but highly interlaced and hamming encoded. When data is read from a CD, the interlacing is removed, and the data is checked for errors. If any are found, they are corrected. If a burst error is found, say, from a speck of dirt, then the missing data is interpolated so you don't hear a pop or click. If the burst error is too big, yes you will hear it. Then the data is pulled from the buffer and sent to the dac. The same clock the controlls the pulling of the data controlls the rate of the dac. Actually, this rate is fed back, and controlls the rate of the disc motor. The rate of the DAC is always controlled by an oscillator. They are very accurate and very cheap. If anything would go awry here, it would be the quality of the dac and the power supply drop to an instantaneous load of a high speed transient. More expensive systems use better chips, filtering, and isolation between stages. For example, the voltage wavering indicated by the first link, on the first comment, reports of internal logic rippling the power supply to the dac and fouling the output. I guess an expensive system would use optoisolators and dual power supplies. Easy fix, but they will charge you....


Speeding never killed anyone. Stopping did.
[ Parent ]
Transport jitter is now irrelevant (4.00 / 1) (#41)
by jwb on Sun Jul 15, 2001 at 04:21:20 PM EST

I've listened to everything Madrigal makes, and it sure sounds good. It does not sound better than equipment that a competent person can make at home for less than $1000, or much less depending on amount of wankery desired. The Crystal CS8420 is a sample rate converter with programmable 8-108 kHz sample range and up to 24 bit word size. The 8420 takes an AES or SPDIF signal, jitter and all, converts it to the desired sample rate and word size, which may be the same as the input frequency, and outputs it to a DAC via a serial connection. The 8420's output clock is controlled via an external crystal oscillator, which of course is very very accurate. The result is that a DAC utilizing the 8420 or similar chips from other manufacturers has NO input jitter at all, when used with any reasonable transport. I use mine with a $99 JVC DVD player, and the jitter in the SPDIF output is eaten by the 8420. No jitter is seen at the DAC. The result is wonderful resolution and phase accuracy in high frequencies.

Let's face it, Madrigal equipment is expensive not because of its technical superiority, but simply because the market will pay it. Any competent hardware hacker can wire up a DAC that will beat Madrigal, and the hardware hacker gets the satisfaction of absolutely no comprimises. He can use a separate transformer for the digital side and each analog output. He can put the power regulators right up next to the ICs. He can use $5 Vishay resistors if that is what he wants. And he can use stepped attenuators which even Madrigral won't give you!

Ah, back to my original point, with a modern DAC design jitter induced by the source transport is not worth worrying about.

[ Parent ]

10k for anti jitter? (none / 0) (#56)
by EriKZ on Tue Jul 17, 2001 at 08:53:25 AM EST


I don't understand, wouldn't copying the CD to a small hard drive in the system and playing the music from there stop all jitter problems?

[ Parent ]
Boogeyman for audiophiles (none / 0) (#64)
by wnight on Wed Jul 18, 2001 at 03:32:47 PM EST

The the data path from CD -> Reader -> Buffer -> DSP -> Buffer -> DAC, the only place that any delay matters is in the DAC.

I think you'll agree that if you're playing from a buffer and that buffer contains a certain bit stream, then anything that came before it, if it reproduces the same bit stream, is irrelevant. A buffer can't contain a 'late' or 'early' bit, only a 1 or a 0. If the 1 or 0 isn't so late as to not be what it is supposed to be, your signal is perfect.

The whole jitter issue is caused by a DAC holding a signal at a level too long, perhaps because it was using the incoming data as its clock.

Any issues before the DAC are either so small as to be fixed (bit read errors) or so big as to prevent a proper signal (whole frame corrupt).

Old CD players kept only one frame in memory, just enough to apply the ECC codes. They then sent the signal straight to a DAC. They also usually streamed data to the DAC, so it was always receiving a bit at a time.

Today everything is done differently. The CD player would burst data to the DSP a frame at a time (to allow the DSP to mask any errors the ECC couldn't fix). The DSP would burst data to the DAC a few bytes at a time, etc.

Because of this send, wait, send, wait, method of data transfer all that needs to be done is for the receiving device to ask a little sooner, or later than expected, but easily withing the tolerance of this wait cycle.

Because of this, if a device falls under its low-water mark for data, it sends a message back asking for more. If this signal comes a microsecond sooner, the data gets sent a microsecond sooner.

This means that if the DAC timing is subtly off, it will play the music either faster or slower, though only by a few parts per billion.

Any 'jitter' than exists now is in the DAC itself and is caused simply by a low-quality DAC not producing a perfect signal. Nothing to do with any previous parts of the equation.

Read post #63 for more details.

As I said in that post, almost 100% of quality issues (post-mastering) are in the Amp and speakers, amp->speaker cables and power filtering are the rest of the equation.

Any other results are from biased tests.


[ Parent ]
Easy way to test multiple Codecs (4.25 / 4) (#22)
by jbridges on Sat Jul 14, 2001 at 09:06:00 AM EST

Take a perfect RIP of a CD, burn original WAV as CD A

Encode with MP3, then decode, burn the WAV as CD B

Repeat with MP3Pro, burn decoded WAV as CD C

and so on for each CODEC.

Then you end up with a stack of CDs that look identical, and can be played on ANY system.

Give each tester a stack of these CDs with random labelling. They will appear identical, there will be no clues at all as to which is which.

Now let's get the testers to pick out the differences!! (grin)

(and of course, forget 64kbit, use at least 128kbit, maybe do each codec at a couple bit rates making more CDs).



audio testing (4.40 / 5) (#24)
by greycat on Sat Jul 14, 2001 at 11:55:26 AM EST

Having to switch CDs in order to get to the next sample is tedious and time-consuming. It also introduces a substantial delay between the hearing of the two samples, during which time the listener can easily "forget" the exact nature of the sound he heard in the previous sample.

One of the best approaches to testing that I'm aware of is called the double-blind A/B test. Double-blind should be familiar to many of you already: it means that both the tester and the test administrator have no knowledge of which sample is being played until after the test is over, so that the administrator can't accidentally reveal information to the tester. A/B testing means that the listener can switch back and forth between two samples rapidly, as many times as necessary, so that he can compare them with no "lag".

Randomization is also important -- across multiple tests, the samples must be presented in a random order (in other words, don't always present the Vorbis sample first, then the MP3Pro, then the WMA; you have to mix it up). This is related to the double-blind aspect, naturally.

Finally, the selection of the samples for encoding is not at all easy. Different codecs behave differently when fed different inputs; sometimes very differently. Someone who wants to construct a misleading test could select samples that are known to give one particular codec a hard time, or which are handled best by some other codec. Yet at the same time, you have to choose samples which are difficult to encode; otherwise, there will be no perceptible difference between the compressed audio and the original.

I'm still learning about this field (like most of us). In my experience the most difficult pieces to encode fall into two broad groups: noisy recordings, and "industrial"/"techno" music. Noisy recordings (such as those originally made on analog tapes -- e.g., Jimi Hendrix) are difficult because the encoder has to try to reproduce the noise in the recording -- but most codecs are designed for normal musical tones, not noise. "Industrial" and "techno" music usually features a lot of "noise" of its own, in the form of synthesized sounds that aren't typically produced by normal acoustic instruments; and in addition to this, it usually has a wide spectrum of active frequencies (high and low sounds all over the place).

MP3 was optimized for a 16kHz lowpass filter; sounds above that frequency are difficult to encode, and most encoders at low bitrates will drop them or distort them. This leads to artifacts when encoding high-frequency sounds such as those produced by cymbals. Therefore music that has a lot of cymbal hits (typically rock or metal) is also challenging for the encoder.

Then, of course, there's the human voice. The people who designed MP3 spent a long time optimizing it for the human voice. Nevertheless, the range of pitch and timbre that the human vocal "subsystem" can produce is remarkable, and can often be as challenging as "industrial" music.



[ Parent ]

If a few seconds delay makes a difference... (4.00 / 4) (#25)
by SIGFPE on Sat Jul 14, 2001 at 12:42:04 PM EST

If it's that hard to tell what sounds better then quite frankly - who cares? I find all this talk of 'scientific testing' comletely ludicrous when at the end of the day we're after a subjective experience.
SIGFPE
[ Parent ]
Overtones (4.00 / 1) (#27)
by ToneHog on Sat Jul 14, 2001 at 02:51:51 PM EST

In my experience the most difficult pieces to encode fall into two broad groups: noisy recordings, and "industrial"/"techno" music.
Encoding electronic music should be much easier to encode than most other forms of music. Synthesized music is comprised of sound algorithms, even the "noise" of which you speak; wouldn't that make the raw audio easier to compress?

From my experience, one of the most difficult types of music to compress with a more true reproduction have instruments that have a lot of overtones, such as single reed, double reed, and percussion instruments. Ever listened happily to a Miles Davis mp3 compressed at 128kbps? I haven't.
Breeze,
TH
[ Parent ]

Why techno is hard (3.50 / 2) (#28)
by mbrubeck on Sat Jul 14, 2001 at 04:28:34 PM EST

Electronica sees significant distortion with most lossy codecs because it uses a much larger frequency range than most audio (no orchestral music has a significant portion of its power concentrated in pure 10Hz bass tones), and because it uses timbres like those of square- and sawtooth-wave synths that are not found in "natural" sound. Some of the techniques used in audio coding are ill-suited to synthesized music; more importantly the specific codecs are not tuned to work well with the more unusual properties of techno music.

[ Parent ]
re: overtones (none / 0) (#55)
by treefrog on Tue Jul 17, 2001 at 08:20:07 AM EST

I am not an audio codec designer... but I do work in telecoms with some pretty hot AMR codec types.

I also used to be an audio sound engineer, working with digital and analogue equipment, 10 years ago or so.

In my experience the best tests are not rock or metal, or techno, which usually are quite busy, in that there is a lot going on. Something with lots of space is often a harsher test of a system. I agree with the comment below on Miles Davis - Kind of Blue or Sketches of Spain would be a good test. My own favourite though as a test is Rickie Lee Jones first album. If that sounds good, then anything will (IMHO)!

Best regards

Treefrog
Twin fin swallowtail fish. You don't see many of those these days - rare as gold dust Customs officer to Treefrog
[ Parent ]
Double blind if they want - it doesn't matter (2.33 / 3) (#39)
by jbridges on Sun Jul 15, 2001 at 06:02:59 AM EST

The point is, the testers have the CDs, they can use double blind, or they can put one in then another. As long as the testers don't know which CD is which it doesn't make any difference, there can be no bias, nothing subjective.

These tests that use MP3 players, or other silly mechanisms to playback digital audio are crazy. They introduce all kinds of effects based on the conversion to analog part of the playback process! They should be using the exact same playback mechanism for all the samples (CDs are just a convenient format to do this).



[ Parent ]
Delay doesn't matter (5.00 / 1) (#49)
by valency on Mon Jul 16, 2001 at 03:15:50 PM EST

Having to switch CDs in order to get to the next sample is tedious and time-consuming. It also introduces a substantial delay between the hearing of the two samples, during which time the listener can easily "forget" the exact nature of the sound he heard in the previous sample.

Any error introduced by the delay will not be biased for/against any particular codec -- so as long as you hand them the CD's in purely random order, any delay in switching CDs will introduce purely random noise into the results -- which will be nullified simply by using a sufficiently large group of listeners.

The rest of your comment is great, btw.



---
If you disagree, and somebody has already posted the exact rebuttal that you would use: moderate, don't post.
[ Parent ]
I'm not sure what you are saying... (3.22 / 9) (#26)
by mindstrm on Sat Jul 14, 2001 at 02:06:09 PM EST

I mean, I understand the issues regarding jitter; but most people have normal, run of the mill CD players, so it makes sense to compare this to what an audio codec available to normal humans can do.

As for why people choose to encode at 384 when 256 is enough. Well, technically, 384 can produce better output. Why do people encode at 256 when VBR is sufficient? Does it make them stupid? I record a number of tracks at 384 back in the day, because I didn't know exactly which implementation of the codec to use, at which bitrate.

People don't pick a codec because it's 'the best'. They pick it because it is available, and meets their needs. mp3, in whatever form (typical 128Kbps stuff, for example) is obviously NOT cd-quality, but it's GOOD ENOUGH to listen to, certainly better than FM-radio in most cases, certainly high enough quality for most people unless you have a thumpin stereo.

http://www.r3mix.net/




/. discussion for insight on how to do it right? (1.50 / 8) (#34)
by untrusted user on Sat Jul 14, 2001 at 08:31:24 PM EST

Talk about things that sound stupid...

Why the 1 (3.00 / 4) (#37)
by regeya on Sat Jul 14, 2001 at 09:21:05 PM EST

Well, it was pretty obvious you were just mojo whoring, ya whore.

Had you read the article on /., and had you read any of the comments, you would have found that, yes, many readers (and far too many did the "me too" thing of posting the right answer) knew a little something about conducting a study.

The only thing stupid here was your need to mouth off about something you apparently know nothing about.

[ yokelpunk | kuro5hin diary ]
[ Parent ]

Codecs as a black art. (4.33 / 6) (#46)
by Sawzall on Sun Jul 15, 2001 at 11:38:31 PM EST

I work in the broadcast industry. Currently, we are spending millions, and will spend billions, on new digital equipment. Many, if not most, of those buying decisons will be driven to some degree by the codec and similar stuff.

Why? Because we are now all beginning to work on pure digital streams. We are all insisting on sticking to one standard codec and encoding scheme since transcoding just wipes out the sound - information theory says that each of those steps back and forth loses valuable information. Today, the industry is almost all MPEG II. There may be better ones out there, but it still does the job.

Meanwhile, there are several new codecs competing to become the new standard. Lucent has spent billions on developing thier PAC codec. It is assumed to be used in at least one of the new Sat. Digital Radio systems. It is being pushed in the new InBand On Channel (IBOC) digital radio concept here in the states. Totally different system than the one used in Europe for thier Digital Radio. If it wins those two battles, it is likely that it will become the standard since having more than one is just disaster on the air signal.

Now here is the bad part. While it claims to be "near CD quality", it sounds like shit. It is very lossy. It rings.

There will be no opportunity to record these airstreams in the digital mode. There will be no digital output - The codec will be on the same chip that authorized you to "hear" the bits. It will be so private and so buried that RIAA will have a field day on keeping out the hacks - this is non-trival, unlike the DVD issue. If not illegal, it would still be hard. So only analog will be the output. This means it sucks to be you - don't throw away your cassette decks yet. If you attempt to re-encode these, like into MP3's, there will be so many bits missing that it will be worse than a bad tape recorder. So thats the battle front on the codec issue - They are creating as system that will make it impossible to exercise fair use in the digital world. This war is being fought in front of the FCC now. If Lucent wins out, prepare to be raped on the fees they will want.

Calm down (4.33 / 3) (#50)
by raph on Mon Jul 16, 2001 at 03:19:56 PM EST

Let me clarify a few things:

1. I am not arguing that jitter is a plausible mechanism for the idea that the green magic marker improves sound quality. In fact, I rather doubt that the pit pattern on the CD has anything to do with jitter in any case. CD mechanisms have their own clock, and sample the analog signal from the optical detector at that clock rate, prior to doing the (nontrivial) decoding and error correction. For a CD mechanism to derive the clock from the optical pickup would be insane.

2. Green magic markers don't, of course, work. The only problem is with the refutation, which espouses the now known to be incorrect "bits is bits" point of view.

3. Of course jitter has been understood for a while. What's new as of ca 1993 is the realization that even jitter in the range of tens of picoseconds can have a significant effect on audio quality. The acceptable range for jitter in telecom is orders of magnitude larger, as the concern there is flipping bits, not causing audio artifacts.

4. Yes, new high quality DAC's should generate their own, extremely stable clock, and thus be almost completely insensitive to jitter. With such a DAC, your generic $50 CD mechanism should produce identical sound as your $10,000 Mark Levinson boondoggle.

5. The noise problems in my sound card are no doubt caused by poor shielding. They are obviously not jitter, as I can hear the noise even when there is no signal. I stand by my assertion that the card is a piece of junk in terms of audio quality - the noise is much higher in the right channel than the left. In a high quality piece of equipment, that would never happen.

6. Because my card is a piece of junk, I find it entirely plausible that it suffers from jitter, ie I find it believable that the same engineers who didn't bother to reduce the noise in the analog chain would also not have spent the extra effort to make sure that the D/A clock is accurate to within tens of picoseconds.

7. Yes, pro equipment is rationally designed, while audiophile equipment is designed to extract money from suckers. Quality is important, but just because it costs a lot of money doesn't necessarily mean that quality is what you get. It is an amazing feat of consumer electronics engineering that such high quality components are available at dirt-cheap prices today. At the same time, it's also possible to buy junk, at any price range.

Quality was the focus of my original post. Brilliant engineers such as Chris Montgomery and the LAME team have put a lot of blood, sweat, and tears into making audio codecs that render sound with the highest quality. It is an insult to their talent and hard work to then choose a codec because of inaccurate, biased perceptions of quality. That is what I was trying to say in my (somewhat rambling) essay.

Thanks to everyone who responded! I might use the feedback to write a more tightly focussed draft.

P.S. I found some excellent technical papers on jitter at www.nanophon.com

Picosecond jitter (4.66 / 3) (#54)
by sigwinch on Tue Jul 17, 2001 at 04:38:33 AM EST

What's new as of ca 1993 is the realization that even jitter in the range of tens of picoseconds can have a significant effect on audio quality.
I find this to be utterly unbelievable. Assuming a cumulative jitter of 10 ps/cycle, and a clock rate of 50 kHz, the total cumulative time differential over a one second period would be a mere 0.5 microseconds.

Assuming the speed of sound is 350 m/s, an equivalent frequency change would be caused by the listener moving their ears 0.2 mm over a period of one second. The doppler shift from that motion is the same as the frequency shift from 10 ps/cycle jitter.

If golden ears can hear a 'significant effect' from 10 ps/cycle jitter, then they can hear the same 'significant effect' from the doppler shift due to an almost imperceptible motion of their head. This means that a large motion of the head would cause horrible distortion that renders the sound almost unintelligible. Since that is clearly not the case, the claim is disproven.

--
I don't want the world, I just want your half.
[ Parent ]

Maybe (3.00 / 2) (#59)
by coryking on Tue Jul 17, 2001 at 10:33:59 AM EST

But there is the phycology of hearing. Perhaps your brain is compensating for this frequency shift when you are moving, much like your eyes automatically move as you turn your head. However, when the music does this, and you are not moving, perhaps your brain does not compensate.

This may sound far fetched, but I don't think is really is that unreasonable of a counter argument. There is a lot about the brain we don't know, but we do know that the brain does a lot of processing to our senses before "we" get to "look/hear" them so I dont think you can just flat out say that we would notice the frequency difference when moving.

[ Parent ]

Psychoacoustics (5.00 / 2) (#61)
by sigwinch on Tue Jul 17, 2001 at 12:02:36 PM EST

The doppler shift analogy was to tie the claim to an understandable effect that we have all experienced. I should have stated it in terms of the audio source moving rather than the ears, to avoid the issue of psychoacoustic compensation by the brain.

Anyway, the point is that 10 ps/cycle is equivalent to a 0.2 mm/s doppler shift, and is claimed to have a 'significant effect on audio quality'. However, we all routinely experience doppler shifts four orders of magnitude larger than that *without* experiencing much degradation of audio quality. This tells me that 10 ps/cycle is utterly negligible.

Even at radio frequencies, 10 ps/cycle is fairly small. At 1 MHz, it causes a mere 10 Hz frequency deviation. That's right, if you applied this big horrible 10 ps/cycle noise directly to the carrier of a 1 MHz FM radio transmitter, the received signal would be just fine. That means that it is utterly negligible at audio frequencies.

--
I don't want the world, I just want your half.
[ Parent ]

Audio codecs: the green magic marker of the '00s | 64 comments (64 topical, 0 editorial, 0 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!