Session Twenty Years After:
Degradation Survey of a Large Collection of Optical Discs
Presenter Alain Carou
Bibliothèque nationale de France, département de l’Audiovisuel – Section conservation


ABSTRACT: ENGLISH

Compared to challenges such as analogic formats obsolescence or archival data management, safeguarding our contemporary audio and video heritage on stamped optical discs can look a minor issue. As far as one has experienced until now, this type of carrier benefits from chemical stabiliy, mostly superior to that of WORM carriers, for example.

However, several series of CDs and more recently DVDs are known to have degraded notably in just a few years, due to defective manufacturing processes.  Knowledge concerning this problem has been shared by publishers, archives and individuals when the degradation process has grown fast and obvious. Now, for heritage purpose, it is necessary to trace the decay of carriers as soon as possible, before they can be perceived in playback, in order to organize migration to new carriers.

Bibliothèque nationale de France’s Audiovisual Department began in 2002 a systematic survey of its collections (200,000 CD, 10,000 DVD). It started with a series known for important defects. But beyond this simple control, BNF engaged a comprehensive investigation year-by-year of samples.

This paper will adress the technical and statistical issues of the survey : equipment, method of sampling, test parameters, interpretation of the results.  It will also provide the results found so far.

ABSTRACT: FRENCH

Comparée aux défis de l’obsolescence des techniques analogiques et de la gestion des supports d’archivage, la sauvegarde du patrimoine édité sur support optique numérique peut paraître une source de préoccupation secondaire. Il est en effet avéré que ce type de support jouit en règle générale d’une bonne stabilité dans le temps, nettement supérieure à celle des supports optiques enregistrables.

Pourtant, certaines séries de CD et, plus récemment, de DVD sont connues pour s’être dégradées de manière notable en seulement quelques années du fait de mauvaises conditions de fabrication. Ces anomalies ont été portées à la connaissance du public par des éditeurs, des archives ou de simples particuliers lorsqu’elles ont été rapides et source de défauts perceptibles graves. Cependant, pour en assurer la conservation au mieux, il faut tenter de les dépister aussi tôt que possible, avant même qu’elles ne franchissent le seuil de perceptibilité, pour engager une recopie sur un support sain.

Le département de l’Audiovisuel de la Bibliothèque nationale de France a entrepris à partir de 2002 une enquête systématique dans ses collections (200 000 CD, 10 000 DVD). Celle-ci a d’abord porté sur des séries connues pour avoir fait l’objet d’alertes publiques. Au-delà de ce programme de contrôle d’informations d’origine extérieure, la BnF s’est engagée dans un programme de vérification par échantillonnage annuel.

Ce papier examinera la problématique technique et statistique d’une telle enquête : appareillage, mode de sélection des documents, paramétrage des tests, interprétation des résultats. Il fournira également les résultats qu’elle a donnés jusqu’à présent.

_______________________________

PRESENTATION

Since 1940, the Bibliothèque nationale de France (the French National Library) has been in charge of the legal deposit of sound recordings. Any sound recording published or imported to the French territory on a minimal quantity has to be sent in two copies. Today, BnF keeps 190.000 documents published on CD audio, that is 230.000 different physical units, and 400.000 if you take into account the second copy of the legal deposit. The increase rate is presently 16.000 documents / year. Preservation of this heritage in totality and without any limitation in time is one of the basic missions assigned to legal deposit. It is intended to allow any motivated person (researchers, amateurs, professionals) to have access to any sound document that has been published once.

BnF’s policy for offline born-digital documents such as floppy disks, cartridges, CD WORM and CD-RW, etc. is immediate or short term migration on mass storage media. For mass storage media are more homogeneous and more secure, thus easier to manage for long term preservation than the original carriers. Now, for CD audio, we have at least three reasons to favour preservation on the native media as long as this technology is not obsolete :

  • we have limited resources for transfers, which leads us to give priority to more fragile media ;
  • the CD collection represents a huge mass of data (approx. 100 Tb), thus migrating it would entail higher management costs of the digital archive ;
  • technical difficulty raised by accurate bit-to-bit copying of audio CDs, unlike files.

More basically, why change the media if the native one is adequate for preservation ?

It remains that objective evaluations of real-time CD audio aging still lack. It has been now 20 years since the first CD-A were received : they have reached the potential life expectancy one can find sometimes in the most careful literature on the subject. Can we trust preservation on original carriers ? For how long ? Can we identify specific pressing companies, years, chemical characteristics... for which CDs should be assessed very carefully, or even migrated immediately ?

Tests have been made since 2002 in three directions, in order to collect elements of evaluation :

  • the first series of tests dealt with a sample of CDs of years 1983 to 1994 already tested 10 years ago ;
  • the second session dealt with a more extensive sample of the first years ;
  • the third dealt with specific parts of the collection, known as or suspected to be risky.

Our paper will present the results found so far.

Basics : error corrections, main quality parameters

Error rates are the parameters more frequently used to assess the quality of a CD. CD players have two levels of error correction, called C1 and C2.

BLER is the measure of the number of blocks per second having at least one wrong 8-bit symbol detected by C1 (a block is made of 32 8-bits symbols). The limit of 220 represents 3% of wrong blocks in one second. This parameter is indicative of the general state of the CD, but fatal errors can happen far below a BLER of 220.

E22 and E32 are the number of block errors with 2 and 3 or more wrong 8-bits symbols as detected by C2. E32 different of zero means a loss of information : the correction mechanisms cannot recover a part of the sound data. E22 and the number of consecutive errors detected by C1 (BERL) can provide good indications of a risk of E32.

A second category of parameters is the characterization of electrical modulation resulting from the reflection of the spot. We are here at a more physical level of description.

I3 and I11 correspond to the amplitude of the signal for 3 successive zeros between two "1" and 11 successive zeros between two "1".

SYM is the ratio between average length of the pits and average length of the lands.

REF specifies the level of intensity of the light reflected by the metal layer.

XT specifies the ratio between the signal amplitude on the tracks and between the tracks.

"Jitter"; is a measure of irregularities in the detection of "1" (transition between pits and lands).

Some other parameters exist, but will not be examined here.

The tests were made with a CD-CATS SA3 : this CD analyser, manufactured by AudioDev Company, is one of the most reliable machines of this kind. It is equipped with a reference turntable and provides data including error rates and HF parameters. It provides figures, but also a visual representation of the distribution of values on the surface of the CD. This function helps to localize defects on the CD, and often to detect its cause (dust, scratch, metallization defect...).

1983-1994-2004-06-20

For the first series of tests, 121 CDs were tested in 2003-2004 and the results compared with those obtained in similar tests made between 1990 and 1994 by Jean-Marc Fontaine with a CD CATS SA2. The pressings were the same, the physical copy tested was the same or another. All the CDs tested had entered our collections new. They have been kept in darkness and, at least since the installation in the new BNF facilities in 1998, in controlled conditions of temperature and humidity. Air is filtered : pollutants are rare, though we may not yet be sure they are completely absent. The CDs received before 1989 were labelled : this practice was stopped then, to avoid preservation problems (however, we could not detect any noticeable interaction between label and CD). Finally, some of these discs may have been handled throughout the years for listening, hence a risk of wear, shocks, etc...

On this table, you can see the number of CDs represented in the sample for each year and each pressing company (designated by a letter).

 

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

Total

MPO

A

2

2

2

5

6

1

5

1

24

Sonopress

B

2

1

3

1

1

1

3

3

1

16

P+O

C

3

1

4

PDO Fr.

D

1

4

4

4

13

EMI

E

3

7

10

DADC Au

F

2

1

1

2

1

7

CDV

G

1

2

1

1

5

EMI Sw

H

3

2

5

SNA

I

1

1

1

1

4

AREACEM

J

2

1

1

4

SNA

K

1

1

1

1

4

RSA Germ

L

4

4

PDO Germ

M

1

1

2

Disctronics

N

1

1

2

ICM

O

1

1

2

Lordisc

P

1

1

2

Nippon Columbia

Q

1

1

Tecval

R

1

1

Dureco

S

1

1

Non déterminé

1

4

1

 

1

2

1

10

Total

1

3

4

4

8

19

14

29

9

13

8

9

121

Comparing results obtained 10 years ago with those of today requires taking into account distortions introduced by :

1/ the difference of machine : analyzers are both very reliable, but it is another machine, from another generation, and another turntable, with better performances ;

2/ the calibration disc : in changing it, the setting of HF values and jitter may differ.

We should also mention an intrinsic part of uncertainty : two consecutives tests of the same CD generally show small differences.

So we think that a variation of a few percents in a parameter or another should not be interpreted necessarily as a significant result, either at the level of each individual disc or at the level of the sample in its whole (by compilation of individual variations).

We have set a level of significant variation for each of the parameters retained, using a 1994 study comparing the results obtained by three CATS testers.

If we assume this level of uncertainty, we have the following results :

106 CDs are "stable" (that is the variation of all parameters, except crosstalk that I will comment later, stays in the limits we just defined : if a degradation has occurred, it is too low to enable us to detect it).

Average BLER is higher for 4 CDs

Average BLER is lower for 7 CDs (a surprising result we will have to explain)

There is no significant increase of E22 and BERL, nor E32 (except sometimes very locally BERL = 6 or 7 and a punctual E 32, but it could be always attributed to dust, particle or scratch, so a cause independent from CD aging).

4 CDs show non conclusive variations, as initial results and present ones are totally incoherent. However, all these CDs have rather good results.

Let us focus on the 4 CDs with a significant increase of BLER

[tableau]

The 4 CDs have been pressed by different companies. 3 of them are the pressing companies most represented in the sample, and the other discs of the same company and same year are "stable". Something noticeable is that the first CD (of 1985) is in a double case, and the other has not changed significantly in 10 years. At this stage, it seems difficult to draw affirmative conclusions on risky companies or risky years. However, these discs are not in a very risky state. Only one (the third) needs to be copied. For all 4, tests will be made every year in order to measure the rapidity of the degradation process.

Let us focus now on the 7 CDs whose BLER has decreased.

This sense of variation was a surprise. It must be eventually noticed that 5 of these discs have a minimum SYMmetry of –18, quite lower than the limit fixed by the Red Book (-15). There is no other disc with a SYMmetry lower than –17 whose error rates have not much decreased. One other disc has abnormal jitter and high SYMmetry, a bit higher than the Red Book limit (which is rare).

Apparently our present tester is more tolerant to abnormal SYM than the tester used for initial tests. This example shows that a comparison between tests uniquely based on error rates reveals rather the evolution of CD playability (that is an evolution of the disc-player performance) than the evolution of the CD itself.

A few words must be said about the problem of crosstalk.

This parameter shows an important increase on all CDs, without exception, between 40 and 60 %. This evolution seems too general to be related only to an evolution of the discs themselves. In the contrary, it can be related at least for a part to the change of tester and calibration disc. With our tests, 2 CDs overcross the limit of 50 % fixed by the Red Book, which could entail problems in accurate tracking. The evolution of this parameter should be watched with care.

The conclusions we can draw from this series of comparative tests are rather reassuring. Nevertheless over a period of 10 years, CD degradation seems a reality. It affects 3 % of the sample.

This is only an indication, for this sample was not absolutely representative. To get a better sight on the state of conservation of the eldest CDs, a second series of tests was made with more important samples of years 1983 and 1989. I can display the results for 1983.

Extensive tests : 1983

550 pressings have been received that year on legal deposit.

Each combination of one label and one pressing plant has been sampled at least once, and one supplementary item every 20 in the same group. There were 39 documents in the sample.

The noting of BLER has been refined, for high BLER max may correspond with a punctual peak, easily corrected by C2 decoder, and low average BLER on a long playing CD can conceal important variations within the CD. So we note what we could call "continuous BLER max", which means the maximum level over which BLER stays during a continuous time (one or several seconds for example).

Out of 39 CDs

24 have BLER max cont. lower than 10

9 have BLER max cont. between 10 and 29

3 have BLER max cont. between 30 and 99

2 have BLER max cont. between 100 and 220

1 has BLER max cont. higher than 220

The latter is also the only one to have E32. It has been analysed as a defect in the manufacturing process for the two copies show the same characteristic curves of error. The errors may be there from the beginning.

Apart from this specific case :

2 CDs have a RN locally higher than the limit

3 have E22 in several points

1 has a BERL max higher than 5

Out of 39 CDs, 6 can be considered as cases to watch carefully in order to know if they are in a degradation tendency. At least 4 of these 6 discs seem to come from the same pressing unit, although its name does not appear on them : same type of printing on the whole surface of the CD, same type of numbers on the central hub. 3 of them have been made for the same important recordings editor. This gives us a valuable indication for further inspection.

No comment on HF parameters, because no abnormal value has been noticed.

Specific surveys

Comprehensive tests have been made every time we have found an alert on the degradation of specific labels, year or pressing company. Stamp and Internet have relayed such alerts, raised either by the producer either by individuals. The problems related here could have been hardly detected by a general sampling of the collection, since they affect labels represented by relatively few CDs and, moreover, not every CD of these labels during the critical period.

The alert regarding CDs manufactured by PDO UK between 1987 and 1993 is well-known. It has been raised in 1997 by Hyperion. Some CDs displayed, as time passed, red marks, hence the appellation of "bronzed discs".

XXX percent of our PDO discs found so far displayed high error rates. XXXXXXXX The level of visible degradation was not always in correspondence with the level of readability degradation.

The consequences were limited in practice, since the pressing company ensured remarkably the replacement of infected CDs by new pressings.

Chemical destructive tests were carried out on 3 CDs, including 2 "bronzed" and 1 "healthy". First they passed an analysis of organic components, then an analysis of elements. It appeared that all three discs had a silver metal layer (hence their remarkably high reflectivity), but different lacquers : cellulose nitrate for the two degraded ones, acrylic for the other. Elementary analysis revealed the presence of carbon, oxygen and silver ; no sulphur. We may conclude the problem is silver oxidisation.

Another alert was raised, this time by audiophiles, about CDs issued by an Italian pressing company around 1990. Comprehensive tests made on our discs confirmed the threat. Signs of decay can be observed : radial red lines in the periphery, metal layer clearer under non-printed areas than under printed areas (so that you can read the disc label from the lower side).

Out of 24 discs so far found in our collections :

17 have a BLER higher or much higher than 220 during several minutes

15 have a BERL higher than 5 during at least several seconds

17 have E32 : out of these 17, one has a BLER < 220 ; on the contrary, 1 CD with average BLER 272 remains without E32.

Altogether, 18 CDs show dramatic error rates.

These degradations are far worse than those encountered with PDO UK CDs. The levels of E32 are spectacular on the last tracks of 5 CDs. The most degraded CD gets up to 1989 non correctible blocks per second (that is 25 % of sound information not readable) and gets over 1000 during the 6 last minutes. In XX other cases, the level of E32 remains still low and the loss is not audible when one plays the disc, thanks to interpolation mechanisms inside the players.

The most accurate way of copying the content we found was to import it on an audio workstation from a CD-ROM player working at 4x and allowed to do up to 4 attempts to read in case of failure.

[écoute]

5 CDs out of 24 show low error rates (BLER < 20, BERL < 5). Among them are 4 CDs whose REFlectivity is about 75 %, meaning they have an aluminium layer. The REFlectivity of 18 CDs showing high error rates go from 86 up to 92 %, that is a silver metal layer. Remains 1 CD with high REF but low error rates. These results agree with those obtained with PDO UK : the use of a silver layer can entail quick decay, but is not the only condition.

The cartography of error rates on the CD reveals the process of decay and tends to confirm the role of lacquers.

The attack is stronger in the periphery of the CD, so it comes mainly from the outside perimeter. But you see also that ink printing plays a protective role since degradations are lower below it : the attack comes from the upper side too, through the lacquer when it is directly exposed to the air. This process may remind you of what happens with silver plate photography. The degradation of cellulosic lacquer could entail the apparition of hydrogen peroxide, that has a corrosive function towards silver. Oxidized silver can evolve according two ways : partially reduced, it becomes reddish (colloidal silver), which reminds the colour of "disc bronzing" ; completely oxidized, it becomes colourless. This state can be observed on one disc at least.

Aluminium layers can be affected by oxidisation if they are not correctly protected by their lacquer. Here we can see an aluminium layered CD attacked by the outside periphery. Of course the process is much slower than with silver, because of the properties of the metal.

An other harm affects CDs of the first half of the 1990s : a degradation of the substrate that makes it look fogged up. These CDs are in plastic digipacks. Both copies we stack are affected. Further investigations are on the way to determine the cause.

To come back to the role of external information in the discovery of infected CDs, CD aging has become a trendy question, and one could read few weeks ago assertions in the stamp and on the Internet regarding CD perennity, coming from collectors who provided figures such as 15 or 20 % of the first CDs affected by the so-called "CD-rot". Tests made on our own copies of the pressings they quoted did not confirm the warning, at least for our copies.

The priorities of our sampling and checking program are now :

  • to seek in our collections silver layer CDs priority. This metal has been used for prestige labels in order to guarantee high reflectivity. This restricts the area of our investigations, but we are looking for information on pressing companies that have used this method and the labels concerned. These discs should be copied accurately. If it is not too late, a method to fight oxidisation could be to have the CD case made in corrosion-absorbing material – a solution already sold by a British company ;
  • to seek the presence of nitrocellulose lacquers
  • to continue extensive samplings and tests over the years 1989, then 1986 and 1992.

__________________________________

SPEAKER BIO

Alain Carou
Alain is an AV librarian by profession. He completed his studies in cinema history and librarianship, after which he worked his way in to his current position, in charge of preservation issues in the Audiovisual Department of Bibliothèque nationale de France (Paris) since 2000.

He is also a current member of the IASA Technical Committee.