Topics covered

1. sound quality of a CD-DA 
2. CD version of a recording is not as good as the LP vinyl version
3. Does my compact disk player reproduce the digital signal exactly as recorded
4. Original signal actually recorded at much higher bit rate than the CD
5. Does audio CD have the same digital read accuracy as CD-ROM?
6. Why is it so hard for a CD-ROM drive to extract the digital code for audio?
7. Can a copy of a CD-DA sound different from the original?
8. How long will CDs last?
9. Writing speed for audio CD-Rs
10. Personal experience making audio CD-Rs
11. Cleaning up dirty audio
12. Brass break up after de-clicking
13. MP3 Quality
14. Why 74 minutes of audio on a CD

Is the sound quality of a CD-DA consistently better than the very best vinyl disk, cassette tape, open reel tape, DAT or HiFi VHS?

I used to give the answer "Yes" here. Unfortunately, record companies do not take full advantage of the CD-DA medium when releasing their older analogue recording. See the next question for the reasons why many commercially produced CD-DA's do not match the quality of the same material released on vinyl.

Here's why CD-DAs are capable of better recording than the other formats. 16 bits gives a maximum signal to noise ratio of 96 dB, although unlike analogue systems, this limit applies to narrow band SNR (signal to noise ratio) as well as broad band. However, properly set-up, digital SNR is rarely audible and is overall superior to LP’s and tape. Some people claim that a professional 30 inch per second half inch width analogue tape running a little hot (over driven peaks) can be better at signal to noise than 16 bit digital audio. This is probably true. In my experience, tape has significant problems other than SNR:

1. Temporally varying (gain) response – due to the non-uniformity of the magnetic particles on the tape, the response rapidly changes as well as varying from one end of the tape to the other. If you record a mid-range or high pitched tone on audio tape, you can usually hear this effect as a sort of warble.
2. Large Total Harmonic Distortion – 2-3% is quite common. However, this is largely third harmonic distortion and is not particularly unpleasant to hear. Some people like it. This is sort of like getting used to drug store photography and liking it. In fact, some people like it better than the almost distortion free digital signal.  There is a cult following that prefers the distortion of tube amplifiers over distortion free. Oh, well.
3. Flutter. The tape slides across the record and playback heads. It can never do this with perfectly constant friction, so the tape stretches and relaxes and vibrates all the time while it is being pulled across those heads. This causes a loss of temporal coherence in the signal. This simply means that the original signal is sped up and slowed down a little bit very rapidly. This destroys some of the clarity of the original sound. Again people get used to this and think something is missing when it’s gone.
4. Print through. The signal from one layer of tape prints through to the preceding and following layers. The magnets in one layer partially magnetize the next layer. If you stay well below the Curie point, this is a minimal problem, but it is quite audible in headphones.
5. Tape deterioration. This one got me because I bought a lot of tape in the late 1970’s. At that time the manufacturers made a mistake in changing the formulation. Unfortunately the new one was slightly hydrophilic so that over time it gummed up and became unplayable. Extreme measures such as baking in an electric oven at 140 F were necessary to copy these tapes to new VHS HiFi. Too high a temperature, and you erase the tape. Too long and print through is greatly enhanced. You might not believe all of the hours spent applying special lubricants, Teflon™ tape, making special rollers and so on to save these tapes for one last playing. Yucck! I am glad that it was in the past!
6. Phase distortion. The magnetic field is put on the tape proportional to the voltage applied. Hence   (B is the magnetic field and I is the signal current in amperes.)

1. Surface noise – ticks, pops, and crackle
2. Time varying distortion
3. Wow due to off-center holes
4. Groove print through – the mechanical cutting of the master causes the signal in the current groove to appear in adjacent grooves, most noticeably the preceding groove for a pre-echo effect. I personally hate this one.
5. Rumble – this is low frequency noise caused by vibrations in the recorder or player, disk warpage, inhomogenieties in the vinyl causing non-uniform shrinkage in the cooling process, etc. My Acoustic Research XA turntable was not bad in this department, but I foolishly gave it up because 20 years in Florida caused it to rot.
6. Phase distortion. Most good phonographs employ either a moving magnet or moving coil. The principle is analogous to audio tape, and is therefore subject to the same phase distortion. Piezo pickups (crystal cartridges) are immune, but usually require large tracking forces. Laser pickups are immune but out of the price range of most of us. I am told that you can get a laser reader for LPs for about $10,000.

HiFi VHS – (my analogue favorite)

1. Signal to noise is limited to about 86 dB, but signals are not completely lost in the noise
2. There is some switching noise as the player moves from head to head. A slight tape stretching can cause an imperfect match in position, causing the phase of the recorded FM signal to be off a little bit. This is translated into a slight bump in the output signal. This is usually more than 75 dB below the signal, and can be detected only through critical listening with the volume turned extremely high. Nevertheless, when you hear it, it is a little more objectionable than a constant hiss.

DAT-

Digital audio tape suffers from read errors on the first minute or so of nearly every tape. The signal is also susceptible to wear and easy damage through mishandling by the tape drive. Tape degrades over time due to breakdown and evaporation of lubricants, binders, and plastic backing. I have been told recently, that contrary to magazine propaganda in the seventies and eighties, lubricant breakdown is rare. New tape formulations should cause it to last. However, my experience has been that sooner or later, that tape drive is going out of adjustment, and it will wreck a tape or two before you know it.

In my opinion, the compact disk beats all of these. I am not saying that it is perfect or that an improvement would not be welcome. But read the next question.

Why is it that my HiFi friends swear that a CD version of a recording is not as good as the LP vinyl version?

No. The compact disk format was chosen to be efficient at storage, and so not enough redundancy was designed in to guarantee that the exact digital code will be detected by the laser pickup and sent to the DAC (digital to analog converter.) However, in a very complicated way (those guys must have been really pushed to do this) the data is spread out over each rotation of the disk so that a loss of data in one spot, instead of affecting say, a couple of milliseconds worth of samples in a row, might actually affect 10 milliseconds worth of data, but only every 4^th sample. There are some redundancy bits, but not enough to recover all the data, so the processor interpolates the missing samples. You are unlikely to hear the effect, but, it’s possible that you could. This is actually happening here and there for short little time segments every time you play a CD because the laser pickup has read errors. Because the player has only a small buffer, and because of the lack of timing marks and address codes, it is very difficult to find the same spot again and read it over the way computer systems would do it to get higher read accuracy. Therefore, almost every CD player is incapable of reading the exact digital code that is there. However, these defects are less objectionable than the defects in, say, an LP. However, buffers are becoming popular in portable players to avoid problems with read errors due to motion of the player. A megabyte or two of buffer gives you several seconds to find and re-read the data.

Doesn’t the same CD read data perfectly for computer file storage? Shouldn’t the audio CD have the same digital read accuracy?

No. Not exactly the same CD. Both audio and data files are stored in blocks 1/75 of a second long, but audio is stored there at 176,400/75 = 2352 bytes per block, computer data is stored as 2048 bytes per block. The other 304 bytes are used for address marks and for redundancy to reconstruct data from read errors. 74 minutes of CD space is 783.216 million bytes (746.9 Megabytes), where as this is equivalent to 681.984 million data bytes (650.4 Megabytes.) So on a full CD-ROM, almost 100 more megabytes are used for redundancy than CD-DA’s. Audio CD’s do not use this much redundancy. The player can reconstruct some, but not all of the errors. The rest of the errors must be interpolated. I am unaware of any players that use sophisticated high order polynomials or Kalman filters to interpolate. Most use a very low order polynomial. If you are aware of a high end player that addresses this issue, please email me.

Why is it so hard for a CD-ROM drive to extract the digital code for audio?

Mostly because of the discussions of the previous paragraphs. But it gets worse. Because digital audio does not have block address marks, it is difficult to find the same spot where you left off reading. Most computer disks are read in chunks. When you are ready for more data, you go back and find the next block, and read it. With the CD-DA, you can’t do that because there is no identifier in the data to tell you which is the next block. The only real choice is to read the CD-DA (compact disk digital audio) in a continuous stream. This would require a buffer in the drive large enough to fill with data every time your computer went off to write data to the hard drive. Since this is not necessary for data disks (address marks are put on every track and sector) the CD-ROM manufacturers do not provide this buffer (it would add cost and since you wouldn’t know why, you’d not buy the more expensive competitor – or if you buy a new system, your computer manufacturer would install the less expensive competition unless you specified a more expensive one with a buffer.) So, if you have software to extract CD audio, it must look for discontinuities in the data stream, command overlapping reads, synch up the blocks, and delete the repeated code, all on the fly. Some software does this, but it depends on tight coordination between the computer, the CD-ROM drive, its firmware, and the program extracting the audio. Very often, missing blocks show up with ticks and pops. However, if you "play" the CD-DA in your CD-ROM drive, this is not a problem – it never has to stop to write to hard disk.

So if I make a copy of a CD-DA, it may sound different from the original?

Yes.  In addition, some CD players switch to 1X sampling mode when reading CD-Rs. It is also possible that for some older CD players that jitter in laying down the bits in the track will cause some flutter in the playback. However, every CD is different even from the same pressing. Different tiny bumps in the plastic cause different read errors for each disk. And it is not unusual for each album to be re-mastered for each pressing, so that 3 to 10 or more versions of the same album (each with different original digital codes) will exist. These will all sound different if listened to under ideal conditions by a well trained listener. But if it makes that much difference to you, go to hear the performance live.

What do you mean by that?

Well, the recording engineer intentionally alters the original sound to have a character that he likes. Sometimes it is for artistic reasons, sometimes commercial reasons, and sometimes to best live with the limitations of the final compact disk. For example, Don Kirshner made a lot of money producing groups like the Carpenters so that they sounded "good" on inexpensive AM radios. This sold records. But it was not good sound in any high fidelity sense.

Will a new compact disk produced by the record company sound similar to the LP it replaced?

Sometimes. It depends on the recording engineer. He may want to take more advantage of the compact disk dynamic range and produce a higher fidelity master than the master used for the LP. Some people are so accustomed to the restricted dynamic range of LP’s that they find the higher fidelity recordings to be a little less "professional." In fact, a lot of really critical listeners are so accustomed to the small amount of distortion and noise in LP’s that they find it disturbing that it is mostly gone in CD-DA’s. The sound seems to have lost some immeasurable quality.

I have been told that audio CD's cannot be expected to last as long as our treasured vinyl, but I can't find anyone qualified who is willing to express an opinion on it; I think the worlds consumers are probably being mislead into thinking that their discs are archival. Do you have any information and/or opinions about this?

Well, as for the longevity of vinyl and CD. I do not know how long vinyl will last, but probably longer than you will be able to find a phonograph player.

As for commercial, aluminum coated CDs. Well, the oft quoted value of 50-100 years is short compared to vinyl. The figure is arrived at by the standard tests of optical coatings (something I know a little bit about.)

It goes like this - if I take a hundred CDs and subject them to say 3 weeks of steam, but well below the melting point of the CD, I get an accelerated life test. Most chemists (oops, I was a chemistry major at one time) will claim that the chemical reaction rate constants double or triple for ever 10 degrees Kelvin. Since the temperature of boiling water is about 80 degrees Kelvin above room temperature, one can estimate that the rate is (2.5^8) or about 1500 times normal. So if half of the CD's fail after 3 weeks, the expected life under "normal" conditions would be 3 weeks X 1500= 4500 weeks or 86 years.

This is, of course, a pretty rough estimate. In Los Angles, you might guess that acid air would make them last less and in Tucson more. And you have to believe in the accelerated tests. Writing on disc labels with Sharpies and other things tends to produce bad effects. Leaving them in the car is bad, too.

You could make a nearly exact copy every 50 years as a hedge.

But, most likely, before 50 years there will be a much better medium. CDs are about 15 years old now. I would expect within the next 25 years to see a format that allows recording 100 albums onto a single disc, and it would have a bit error rate of <10^-18 and a life expectancy of 500 years. So, copy your collection onto the new format every 25 years or so. Or, baby your old vinyl. Whatever gives YOU pleasure is what counts. Not what I think or someone else thinks. But don't try to save tape for 25 years. It just doesn't work.

To avoid clicks, the tracks all need to fade to zero at the end and fade in from zero at the beginning. Fail to do this, and glitches are guaranteed.

I believe that the stuff that you describe would indicate that you have a drive that is not precisely synching up when it breaks and then starts to record again. This problem would be characterized differently between 1X and 2X because the drive accelerations and wobbles will be different. It is possible that the drivers do not fully take this into account, or that your drive design is not quite as good as you need for glitchless recording.

I sort of doubt that you are really hearing more detail consistently in discs recorded at 1X. If you really are, it may mean that your drive has a lot of drive errors at 2X and samples are lost on read, and are replaced by interpolated samples. If this is the case, forget using your drive at 2X.

It should be noted that there are a lot of reports that recording audio CDs at lower rates gives improved results. The most common explanation for this is that the drive is able to maintain better speed control at lower speeds and thus there is less jitter in the recorded bit rates.  However, most CD players read the data into a buffer and then spit them out according to a crystal controlled clock, so I tend to discount this explanation. I believe that there are probably more write errors at higher speeds.

Have you ever made a redbook audio CD-R? How was your experience?

Yes, I have. Actually, I have recorded over 250 audio CD-Rs from vinyl and tape sources. I pretty much transferred my entire vinyl, tape and CD library and my wife's (over 1000 albums) to MP3s on CDR.

Next I ran Easy CD Creator. A Wizard came up. The wizard asked for whether I wanted a data or audio CD, asked for the tracks, and then started testing. I noticed that the write light went off between tracks. I did not want this! I aborted. Then I went into the advanced tab and asked for "write at once." I then asked for testing. This time it went straight through without stopping between tracks.

It automatically started recording after the successful test. I requested 24 tracks with a total of 71 minutes of music. It recorded at 2X speed without a hitch. I pulled the disk out. I placed it in a 1985 vintage CD player. I figured this would be a stressing test. Music started to play! I let out a breath. I pressed time remaining. It read "60:38." This was not so good. But this old player never expected CD’s with more than 63 minutes, so what’s going on? I pressed skip a few times and noticed that it could not go past track 15. Track 16 puts the total time past 63 minutes! However, I could play or even fast forward all the way to the end of the disk. No gaps between tracks! In fact, I intentionally had overlapping music between two tracks, and I could not hear where one track ended and the next began. A newer CD player had no trouble with the 24 tracks and 71 minutes. In short, my first CD was a complete success. Since this experience, I have made more than three hundred more CD-R audio disks. I will eventually transfer all my vinyl recordings to CD audio. See the pitfall below.

I should mention that I recommend over two gigabytes of free hard disk space to do what I did. It could be done with less, but it would have been a real hassle to record in small chunks, edit, and then delete temporary storage.

My first CD-R system was a P5-166, 32 Mb EDO RAM, PCI bus with an IDE 4 Gbyte hard drive. I am also using a Soundblaster 16 card. The reason for using an older card was that I believe it to be a bit higher fidelity than some of the newer cards. I actually pulled this card out of my older PC and traded the new card when I got the new PC. [I have upgraded to a PIII-700MHz system with 256 MB ram and a 30 GB hard drive, a Soundblaster Live card, and a Matshita CD-RW CW-7585 drive with 8X record capability. It can record 6X reliably for disc at once, otherwise buffer under-runs are likely.  It can also extract audio at up to 12 X reliably.]

Next, I tried to extract audio to wave files from a commercial audio CD. I placed it in my Smart and Friendly, told Easy CD Creator I wanted track one on a new CD. I then requested "extract to wav file." It worked but truncated the last two seconds from the audio file. I usually use CDCopy now instead of Easy CD to extract audio, even though Easy CD has fixed this problem and added CDDB support.

In short, my experience was good. I resisted the temptation to try to make a CD-R until I was sure I was ready – reading all the documentation and carefully preparing my wav files.

 A pitfall: wave files should be exactly a multiple of 2352 bytes (588 samples). If they are not, the end of the file will be padded, hopefully with zeros, to finish it out. If your file is not "quantized" in this manner, and the end of one file is not zeros or the beginning of the next is not zeros (and I don't mean close) then there will be an audible click between files.

Can you give me a step by step instruction on making a CD-R from vinyl or an MP3 CD-R from vinyl?

Yes. There are a few things that might be cleaned up:

1. Broad band noise
2. Rumble
3. Clicks and pops
4. Short drop outs
5. Speed variations
6. Predictable distortion

Of course, the models aren't perfect, so some methods work better in some situations than others. We accept the following notions:

• Music is generally characterized by a combination of a small number of discrete frequencies.
• The frequencies in music may have a quick attack ~ 1-2 ms (but not abrupt < 500 microseconds)
• Music frequencies decay slowly (>10 ms)
• A discrete frequency treatment should work on a logarithmic frequency scale, since a ten Hertz difference is more important at 30 Hz than at 10 kHz.
• Broadband noise is usually 30 or more dB below the peak signal, hence signals stronger than that can be regarded as signal instead of noise
• Broadband noise statistics are approximately stationary - the statistics do not change over time.
• Noise impulses (clicks/pops) are short lived and can either replace the underlying music or add on top it

Therefore, a filter with a large set of frequency bands could be adjusted so that it passes through things that meet the music model, but filter out low level uncorrelated signals.

Noise impulse can be detected and removed, and the missing data must be interpolated. Extreme care must be used when setting thresholds, which probably have to be changed every few seconds of music.

Really good software is some large combination of models that require great skill to get the best results from. This is, so far, out of the reach of amateurs. The details of really good filters will remain proprietary. However, there are some available that are of moderate sophistication that are reasonably good:

Diamond Cut Millenium - This is an update to DC-Art which is now marketed by Tracer Technologies. It offers batch processing and build your own filters. The algorithms are essentially the same. It has a few bugs in version 4.10.

Note that I am a perfectionist, so I am not going to give a glowing report of anything that isn't perfect.

I personally believe that I have the experience and skill to produce algorithms which get around most of these things. Experience tells me that it’s always more complicated than you think. I believe that more automatic algorithms that do not require much user intervention can be developed. But I’d like to see how much interest there would be. Let me know. Also let me know if this page was useful to you.

New 1/9/99