Thread: Matching CRCs for errored rips

This is a small example of EAC returning matching CRCs for extracted rips even though there are errors in these rips.

Before disc resurfacing, EAC did not report any suspicious positions for any of the secure mode rips, however it did light up the first row of the error bar on all secure rips. After the disc was resurfaced, all rips reported the same CRC and the error bar did not light up.

The following table shows the CRC values.

The most interesting feature is that before resurfacing the DVD drive returned 2 pairs of matching CRCs, both of which are wrong, the odd man out CRC is the correct one.

The other interesting feature is that the only correct rip before resurfacing was for burst mode although there is no way to show whether this was just coincidence, unfortunately no further testing could be carried out once the disc was resurfaced.

The final point is that the errors were repeatable, there were only 2 values that the data would take, the right one and just one wrong one.



RIP DRIVE MODE CRC
a LTD163 C2 rip1 CRC 4036274F
b LTD163 C2 rip2 CRC 4036274F
c LTD163 C2 rip3 CRC 809A3D61
d LTD163 NoC2 rip1 CRC 809A3D61
e LTD163 Burst rip1 CRC 8DAF0CA2
f LTD163 C2 rip1 resurfaced CRC 8DAF0CA2
g LTD163 C2 rip1 resurfaced CRC 8DAF0CA2
h LTR40125S No C2 rip1 CRC D60E061D
i LTR40125S No C2 rip2 CRC F7721A2C
j LTR40125S With C2 CRC 482F169F
k LTR40125S NoC2 rip1 resurfaced CRC 8DAF0CA2

e=f=g = corectly ripped wave
a=b
c=d

differences between rips
f-a = diff1
f-b = diff1
f-c = diff1 and diff2
f-d = diff1 and diff2
f-e = none
f-g = none
f-h = not done
f-i = not done
f-j = not done
f-k = none

I have attached a zip file with extracted waves showing the differences between the LTD163 rips. These waves show that all the differences were linear interpolations thus there were no crackles.

As per usual, the linear interpolations caused an increase in distortion which is most noticable in the higher frequencies. However as these bursts of distortion only last for about 15 milliseconds they are quite inaudible

Regards,
Bob

Hi BobHere,

Oh my god, I thought that getting twice the same erroneous CRC in a correctly configured secure mode was practically impossible.

You have really got me depressed now

I don't know what to say, maybe your LTD163 drive DOES cache audio after all?

Did you test it with Feurio ?

Could you redo your test checking the "drive caches audio data" option, please?

By the way, are LiteOn drives the only ones which report C2 info accurately?

Regards.

I've actually also got two pairs of matching CRCs when ripping twice in secure mode with EAC. I have a HP 8200+ drive and I'm pretty sure it doesn't have any caching or C2.

Originally posted by Estragon
Hi BobHere,

Oh my god, I thought that getting twice the same erroneous CRC in a correctly configured secure mode was practically impossible.

You have really got me depressed now

I don't know what to say, maybe your LTD163 drive DOES cache audio after all?

Did you test it with Feurio ?

Could you redo your test checking the "drive caches audio data" option, please?

By the way, are LiteOn drives the only ones which report C2 info accurately?

Regards.

I have seen that written but it is quite wrong, the probability of matching erroreous CRCs happening on lightly damaged discs is relatively high. The probability drops rapidly as the disc is more heavily damaged. In this example there were just 2 bursts of errors and as the error bursts were always the same there were only 3 combinations and thus CRCs that could have occurred.

The LTD163 does not cache audio data for the version of firmware I am currently using. I have done extensive testing of this and its error detection capacity on No C2 secure rips is intact.

I didn't test it with any other ripping software.

Unfortunately, as the disc had to be resurfaced to produce a reliable rip and thus correct CRC, the original damage has been permanently removed so no further testing is possible. I will just have to wait until I get another of these borderline CDs.

As to C2 accuracy, I can't provide any definitive data on how accurately drives report C2 errors on lightly damaged discs (which is where it is most important). My own opinion is that many drives are probably reasonably accurate in C2 reporting but that current testing procedures place far too many demands on the drives to get truly accurate C2 reporting test results. Unfortunately, the type of test disc that is required to properly test C2 can't be burnt on a normal CD burner, it would have to be mastered - expensive.

I did not mean to depress you, I just wanted to demonstrate a point. Andre's double read system for detecting errors is quite reliable but is not infallible. This is demonstarted by the secure mode scheme of rereading until you get a certain number of matching sets of data. EAC relies on reading the data differently to detect errors but also relies on reading the data the same to determine which data set is correct.

Despite all of the above, EAC usually does detect errors. If the error bar illuminates at all, it has detected errors and is rereading, in this case you may get wrong results whether EAC reports suspicious errors or not. If the error bar does not illuminate then the chances of undetected errors is very slim.

In order to reduce the number of faulty CDs getting through to as close to zero as is practical, I rip each cd on each of two drives, the LTD163 using C2 and the LTR40125S using No C2 secure mode and check the CRCs for a match. If they don't then I compare the waves to see if there are errors or whether the difference occurs due to the different offsets truncating data at the end of the file.

In the end I believe that unless you have a Plextor drive EAC is still the most accurate ripping software available but also that there is no certainty in life or DAE ripping.

Regards,
Bob

EDIT : Below data is pending confirmation, as some unexplained behaviour was observed later. http://www.digital-inn.de/exact-audio-copy-english/16843-analyse-exe-still-reports-impossible-data-v1-3-a.html
It may be a bug in the analyse program.

Originally posted by BobHere
that current testing procedures place far too many demands on the drives to get truly accurate C2 reporting test results. Unfortunately, the type of test disc that is required to properly test C2 can't be burnt on a normal CD burner, it would have to be mastered - expensive.

Good news, I've got a perfect CD for this test : a CDR perfectly readable in the middle, with absolutely no errors, but which becomes more and more damaged until the last track, where the error-channel rate rise above 40,000 per second (50% of wrong samples). There is a wide part showing very light errors.

Bad news, about 70,000 C2 errors are required to get a good statistical sample for the C2 accuracy evaluation. No one is going to read a CD with one error and compare wavs 70,000 times in order to make statistics.

OK, I understand that it is long, so here are some graphs, though not very interesting since they apply only to one uncommon drive



Memorex DVD Maxx 1648 firmware GWH2 DVD ROM. In green, C2 reporting rate in sample-channel per second. Logarithmic, one line is a factor 2, top of the picture is 88200 sample-channel per second (100% errors).
In blue, undetected errors, same scaling.



Errors, undetected errors, and undetection ratio recorded each second, averaged on 14 seconds, sorted on errors, for error rate >=55 per second.

In blue, the error rate in sample-channel per second averaged on 14 seconds.
In pink, Undetected error rate in sample-channel per second x100 averaged on 14 seconds.
In yellow, Undetected errors per 14 seconds/errors per 14 seconds x3,000,000 for each second.

The average value of the undetection ratio is between 20,000 and 25,000.
3,000,000/20,000=150
3,000,000/25,000=120

For an error rate of 10,000 per second, the Memorex DVD Maxx 1648 GWH2 fails to report one error out of 135±15.
For higher error rates, the undetection ratio falls until
3,000,000/15,000=one error out of 200 at 45,000 errors per second.

The following refers to data analysis that is not explaned here, in fact, analysis of the .DAT file in Microsoft Exel.

In order to get an accurate ratio for lower error rates, we must collect about 5,000x14 = 70,000 errors (because the yellow curve becomes stable when the blue one is around 5,000). We have them using all the recordings until the error rate of 654 per second.
We have then 70,166 errors, including 487 undetected ones. The ratio is then one error out of 144, and is the same as we got at 10,000 errors per second.

EDIT : clarifying last sentence : we have C2 missing one error out of 135 at 10,000 errors per second, one out of 200 at 45,000 errors per second, and one out of 144 for an error rate ranging from 3 to 654 per second (the smallest computable).

Originally posted by BobHere
In the end I believe that unless you have a Plextor drive EAC is still the most accurate ripping software available but also that there is no certainty in life or DAE ripping.

There is no certainty in error correction when doing playback on your standalone as well ... but your test is quite remarkable.

Did you notice if the drive's interpolations over a sample range of around 660 (15 ms) used correct samples as support elements or was it just plain replacement of the original samples ?

I am asking because a 15 ms intervall of wrong samples can be clearly audible when there are no supporting elements for interpolation ...

Originally posted by JeanLuc
Did you notice if the drive's interpolations over a sample range of around 660 (15 ms) used correct samples as support elements or was it just plain replacement of the original samples ?

I am asking because a 15 ms intervall of wrong samples can be clearly audible when there are no supporting elements for interpolation ...

Sorry, I don't understand the question, nor whom is asked.
Errors 660 samples wide are quite audible, especially if there are many of them.
Do you remember the interpolation pictures in http://www.digital-inn.de/exact-audio-copy-english/15838-discussion-reliability-daequality-test.html ? The biggest error, in the CRW3200 picture, is 100 samples wide only, and it is completely outside the wav.
What do you mean with "using correct samples as support elements" ? If it is "holding the last good sample", it is not very different than muting for such a wide gap.

Sorry ... I was stating this question in a hurry ...

@bobhere

what I wanted to know ... did your faulty rip show a range of 660 completely invalid samples or were there still valid samples in that area left to act as supporting values for interpolation ? I'm just asking because I cannot imagine a serious interpolation over a range that wide with a last valid sample and the next valid sample as "supporting elements" only ...

Interpolationg over a range of 660 invalid samples cannot be "quite inaudible" IMHO ... so the rip would be really crappy

@pio

sorry for confusing you ...

To BobHere,

Sorry, I forgot to ask you one thing: did you check the "use c2 error info for error correction" in EAC?
I can't see any mention of this option in the logs that you have provided, but I'm not sure if EAC reports this option in the log file or not.

Also, do you think that Plextools 1.15 or above (in conjunction with the newest Plex burner) would have been able to rip your CD without errors before resurfacing?

In other words, do you think that there's a higher probability of getting the correct CRC after a large number of re-reads of the offending sector?
Or, on the contrary, do you think that the possibility of getting the correct CRC is the same as the possibility of getting a wrong CRC?
(I mean before resurfacing in both questions)

Thank you very much.
Regards.

Originally posted by JeanLuc
Sorry ... I was stating this question in a hurry ...

@bobhere

what I wanted to know ... did your faulty rip show a range of 660 completely invalid samples or were there still valid samples in that area left to act as supporting values for interpolation ? I'm just asking because I cannot imagine a serious interpolation over a range that wide with a last valid sample and the next valid sample as "supporting elements" only ...

Interpolationg over a range of 660 invalid samples cannot be "quite inaudible" IMHO ... so the rip would be really crappy

@pio

sorry for confusing you ...

Hi JeanLuc,

As I stated they were all linear interpolations. I should clarify that linear interpolations can only occur where there is a single errored sample between two valid samples. Philips engineered the system through the sample scambling and block interleave to allow for quite long periods of linear interpolation, with every second sample good.

If EVERY second sample was bad and thus linearly interpolated, it would be in effect a 2X UNDERsampling of the waveform (sampling at effectively 22050 samples a sec but with the filter still at 20 khz) which is why it is mainly high frequency distortion (or hash if you like). If you look at the attached zip file you can see how few errors there were in these blocks (about 60 in each from memory) and the (expected) relatively low levels of error.

Regards,
Bob

Originally posted by Estragon
To BobHere,

Sorry, I forgot to ask you one thing: did you check the "use c2 error info for error correction" in EAC?
I can't see any mention of this option in the logs that you have provided, but I'm not sure if EAC reports this option in the log file or not.

I don't use that option on any of my drives. From what I have seen in Andre's posts it is a form of interpolation based on C2 reported errored samples. All my drives already provide the standard interpolation (1 sample linear, multiple sample hold and linear). In this case they are all 1 sample linear interpolations.

Originally posted by Estragon

Also, do you think that Plextools 1.15 or above (in conjunction with the newest Plex burner) would have been able to rip your CD without errors before resurfacing?

The Plextor could quite possibly have ripped it without problem. Based on previous experience, I believe that my Ricoh MP7120 (before it died) would have ripped the CD with EAC without problem. I have never really thought that the Liteons were particularly good performers in DAE.

Originally posted by Estragon

In other words, do you think that there's a higher probability of getting the correct CRC after a large number of re-reads of the offending sector?
Or, on the contrary, do you think that the possibility of getting the correct CRC is the same as the possibility of getting a wrong CRC?
(I mean before resurfacing in both questions)

Thank you very much.
Regards.

Ummm, not sure what you are asking here. I believe that EAC uses majority voting logic to decide which is the best waveform. The number of reads should provide more certainty about the waveform. However in previous posts I have shown that it is possible for a majority voting scheme to pick the wrong waveform due to interpolation however I do not know if EAC is susceptible to this problem.

On the other hand, if the defects on the disc cause repeatable errors then any majority voting scheme will return the wrong value as that value is constantly returned. This is the advantage of using C2 reporting.

Hope I have answered your question,

Regards,
Bob