For extracting digital audio, Plextor drives manufactured before 2010 (arguably before 2005), seem to be “The Best”. (or rather, “The Least Worst”, because no single drive is completely perfect, and you can get that last 1% of accuracy by using multiple drives and combining the results.) I did some research on what it would take to extract bit-accurate archival images of optical media, and the short short answer is I would need to build my own CD reader from scratch, or piggyback my own controller onto a zombiefied CD-Drive, reading the raw FM signal off the laser pickup. (But leaving the laser focusing servos track the groove.) Building my own drive entirely from scratch would allow me to experiment with some really weird use cases, such as reading a CD backwards through the top surface of the disc. (For CDs which are badly damaged on the bottom surface, but the reflective layer and top (printed) surface are undamaged. Also, cracked, broken, and shredded CDs may be partially recoverable by reading (laser scanning) each piece, and then reassembling them in software. (And, also, with my own reader, I can automatically scan the matrix info on the central hub, and make very high resolution scans of the printed in regions on the top surface. (Only just a silhouette based on reflectance and absorption, older CDs are just black text silkscreened on with an empty (transparent) background, so this is fine.)
But anyway…
I think over the past fifteen years, there have only been two or three actual companies which have still been manufacturing CD/DVD drives, and all of the other brand names you see on the market, are just one of those cheep OEM drives with a sticker on the drive (and a string in the firmware) saying the “Well known” brand name.
Anyway, take a look at the documentation of Schily’s cdrecord, or any twenty year old CD ripping software, and marvel at the absolutely huge list of insane firmware bugs that needed to be worked around by burning and ripping software. It’s incredible that anything ever worked at all.
And yes, I realize that what I’d ultimately be building is just a scanning microscope with the ability to distinctly resolve features at least 500nm² in size. A bilevel image, with each picture element representing a spot 500nm² gives a total image size of approximately 17,210,000,000,000 pixels (2.15125TB) for the 86.05 cm² program area of a compact disc… but there are ways to cook this data down much smaller in size without loss. And also, the scanning doesn’t necessarily need to be “fast” – fast enough to playback audio in real time – so it has a lot more time to think and look around for data that regular CD players don’t. (But it can also just go fast, computers are fast now.)
(The pits and lands on an audio CD are between 850nm and 3500nm long.)
CD players use a clever trick to greatly enhance contrast between the land(s) and pits, because each of the reflective surfaces are… supposedly… half a wavelength apart in physical distance. The circularly polarized 780nm light will experience destructive interference between the two reflections off of both the pit and the land at the same time…
I think it’s the circular polarization that really cancels out the reflections (like those old anti-glare CRT screen covers) Because the thing about this that doesn’t quite seem to make sense to me though is that the pits are only 100nm high (or deep), which is not “half” of 780nm. (I mean… that will still give you partial interference, which is better than nothing.) But, when you reflect circularly polarized light, the reflection will flip from clockwise polarization to counterclockwise polarization.
ECMA-130 is less than illuminating on this topic.
Anyway, if you have the time to scan the surface hight of an object (say with Confocal techniques), you don’t really need to rely upon this polarized light trick to distinguish the hight difference between two metal surfaces. In fact, I speculate that you could possibly even capture the light scattering off the edges of each pit on a large surface area, and computationally reconstruct that edge from the diffraction pattern. (I haven’t done the math on this or anything, but intuitively, if you can record all of the scattered light projected onto a sphere, you can just computationally reverse the spherical wavefront back into the point from which it scattered from. And you know it’s just a laser going in, so the transformation between the two will give you the necessary surface shape.)
Anyway, the scanner wouldn’t necessarily need to rotate the CD either, it could just XY raster scan from left to right and top to bottom across the surface, and reassemble/decode all the pieces in the correct order using software. Spinning the CD has the advantage of reading the lands and pits in sequential order so processing/decoding of the data can be done in parallel with the reading. I would also track the rotation angle of each pit. (The radius and angle.) Because… funnily… because the raw signal is self-clocking, it has no absolute size or position… You can scale the pits and lands up and down in length, and wind the track pitch wider or narrower than 1600nm. That’s why early CDRs were only 60min (535MB), and then 74min (650MB), and then 80min (700MB), and then 99min (870MB), and GD-ROM, and Plextor’s GigaRec… etc. Also, there were copy protection techniques which would time reads and seeks between various sectors of a CD-ROM… sectors which had been carefully laid out at certain distances apart from each other on the engraved glass master. If the CD-ROM was burned to a CD-R, the sectors would typically be written into different absolute positions on the disk, which would change the time it takes for the CD Drive to seek and read from them.
Speaking of copy protection, you can write anything you want, anywhere you want when mastering a disc. Many drives make too many assumptions about things like, CD-ROM sector numbers monotonically increasing in sequential order, and every sector address being unique and used only once. If you have two different blobs of data, and you record them sequentially, but duplicating the same sector number between them. Then depending on weather you drive is seeking forward or backwards when it’s searching for those sectors, it will either return the first blob or the second blob. Fun times!
Oh, and there’s also writing deliberate correctable and uncorrectable errors, that’s an old trick. The first copy protection technique I ever learned was for the Commodore 1541, if you deliberately write a disk with a correctable sector error, and then make a copy. The copy will no longer have the error, so you can tell that it’s a copy by just checking if the read error still exists. Anyway, the same trick works for CD-ROMs. There were also “weak” sectors, but I think that with the variability in CD-ROM Drives, especially if they cache reads, it’s not consistent enough to reliably use for copy protection. (The 1541 is just dumb enough for being such a smart floppy drive.) Though actually, I think there was an audio CD copy protection scheme of mastering with deliberate C2 errors, which were inaudible (corrected) when played back in a regular CD Player, but CD drives had a terrible time extracting digital audio from.
What was I talking about… yeah so anyway, I want (me or someone else) to build a Kryoflux for optical media, and on the extreme end of the data recovery scale, possibly read data off a CD that has gone through a shredder.