BobHere

Hi,

I will read your draft with much interest. Although, I think that a trully accurate article on CIRC would be good. Most everything on WWW is either inadequate or wrong.

In above and also in the test disc article you state that the discs exceed the CD standard which is good and proper but the article hints that a drive can outperform these standards. I am not sure how that could occur. If we look at the theoretical best CIRC performance for long burst errors such as the black spot test we can see some interesting data. For complete correction of bursts the performance is totally dependant on the number of symbols corrected in C2 as shown below (for 1.25m/s)

2 symbol correction 8 frames 1.36mm
3 symbol correction 12 frames 2.04mm
4 symbol correction 14 frames 2.38mm

2 symbols can be unflagged, 3 symbols consist of 1 unflagged and 2 flagged (erasures), 4 symbols are all flagged (erasures)

From this we can see that (tracking issues aside) no drive can can correct the 3mm scratch. Interestingly, Aopen being happy with 1.2mm suggests that perhaps this drive has only 2 symbol C2 correction and can't do any better.

But then again, if you are relying on your ears alone to determine the performance then the interpolation performance is the criterion (unless you can differentiate the distortion products produced by single sample linear interpolation). Interpolation performance is generally unaffected by the CIRC configuration (possibly there are some CIRC strategies - stupid ones - that would severly reduce the interpolation performance) and is totally defined by the standard interleave and symbol scramble. Interpolation performance is a guaranteed 48 frames and 2 symbols long (8.16mm) and may be longer depending on exactly where the burst begins.

However to really determine error performance you have to go much further, something that these discs don't seem to do. The CD Check disc looks interesting but the website is somewaht lacking in detail about exactly what this disk does.

From my own testing I have determined that the Liteon LTD163 has superior burst performance to the LTR40125s and I suspect that it has 4 symbol correction against 2 symbol correction in C2. The down side is that under high error rates the LTD163 produces far more undetected errors (and thus clicks) than the LTR40125s suggesting that the flag strategy between C1 and C2 is not very good.

Also the LTR40125s can track concentric scratches better that the LTD163 which suggests better servo lock characteristics. However the LTD163 is better at reading damage to the reflective layer so things are not at all straight forward.

I suppose what I am saying is that so far I have not been able to pick a good ripper based on the currect batch of tests. More to the point, if I want a better DAE ripper than the LTR40125s, can I find comparative tests between the LTR40125s and the current batch of 52x burners?

Anyhow, I am off to read your article.

Regards,
Bob

Pio2001

I find two more frames than you, for 4 bytes correction :

2 bytes : 8 frames OK, because with 9 frames, frames 1, 5 and 9 can each send a bad byte in the same C2 frame.

3 bytes : 12 frames OK, because with 13 frames, frames 1, 5, 9 and 13 can each send a bad byte in the same C2 frame.

But wouldn't 17 frames be needed to send 5 wrong symbols in a C2 frame, from 1, 5, 9, 13 and 17 ?

This morning, analysing a burst error on the Teac 540e drive, I found two contiguous errors in the same channel. I calculated that 51 frames (rounded to one full frame) would be needed to be damaged in order for a pure burst error to produce this (damaging samples 278, and 279 of the right channel, as well as 324 and 325, of the user data, counted from the first bad sample).
That stands for 50 frames instead of 48.

I've finished the big part of my analysis (very long, due to bugs in the analyse program -thanks to Andre for fixing them-, and to confusing CIRC information online).
Now I've just to test the error correction of my CD Player, and summarize the results.

__________________
Pio2001

BobHere

Well, the joke's on me. The Lite-on tests were most interesting. It appears from your testing that the Mediatec 5 chip set is not only inferior to the 6 but also the 3 when it comes to reading accuracy.

I am most impressed with the LTR52246 performance. Its superiority over the other drives is quite dramatic. The amazing part is that all this improvement is shown in the C1 error rates with the burst error tests the most improved. At this stage, I am not sure what that means and will have to think about it but initially it does point to the EFM decoder and servo/pickup as the most likely reason for improvement. A superior C1 decoder strategy should not affect the performance on a burst error black bar test.

The document at the moment is a good compilation of the information available out there plus your own CDR-info articles. But as it is currently a compilation of data from lots of different sources (including me !!! - Danger, Will Robinson, Danger) it contains a lot of inconsistencies, repetitions and a few errors. I will have a look at it and see if I can provide some improvements to the CIRC section, especially the part from my own post.

With enough input this should become the best source of information on the web.

One thing I will discuss in here is that the section on C2 is wrong. I am not sure where this consistently wrong information came from but I suspect it was from early CDROMs that were not meant to be used for audio. The following is a simplified explanation of the 2 symbol C2 strategy formulated early on by Philips/Sony and written down in Pohlman on page 75. If C2 detects one errored symbol, it will correct it. If there is more than one errored symbol detected and more than 4 C1 error flagged symbols in the frame it will copy all the C1 error flags to the C2 error flags and pass the frame on unchanged. This is necessary or the interpolation will be severely reduced - I will explain later. If C2 detects 2 errors and these errors agree with the error positions marked by C1 it will correct 2 symbols and pass on all others unchanged. There are a few other circumstances where it will copy the C1 flags to C2 flags and pass on the frame unchanged and just a few circumstances where it will mark the entire frame errored. All this is based on the relative reliability of the error detection for various error patterns.

Why is this important to interpolation? Well, long linear interpolations have frames where every second sample is interpolated, that is, at a MINIMUM there are 6 errored symbols and up to 12 errored symbols in a single frame. So this interpolation could not occur if C2 did not allow any errored samples to leave before marking all of them wrong.

Arrrgh, I just gotta say it. The paragraph at the bottom of page 13 is rather strange. Gaussian and white yes but not Gaussian or white. Gaussian usually refers to the probability density function of a signal, this is the normal curve (bell shaped curve) or known as the "Error Function" (ErF). White is usually used to describe the Power Spectral Density of a signal what is sometimes (erroneously) called the spectrum. The two describe two totally different characteristics of a signal. Anyhow, white Gaussian noise does not play a great part in the detection errors of CD.

Also the paragraph at the bottom of page 14 makes no sense whatsoever especially the bit about pink noise.

Regards,
Bob

BobHere

There is the interaction of the first interleaver to consider, it spread the burst errors across 2 frames before C1. I will (if I find the time and the ability) sit down and work all this out precisely. A C2 of 1 symbol decoding can correct 3 frames.



Sorry but I don't follow your explanation.

The interpolation can be upset by a number of factors. The actual data is necessary to work it out. Once the interpolator reachs its limit all subsequent data is trashed not just a few isolated samples.

The theoretical interpolation length can vary (as stated before) depending on where the burst begins in relationship to the state of the interleaver. 48frames plus 2 symbols is the absolute worst case.

Here is a rough view, the interpolation can be either 14 or 16 multiplied by 4 frames (due to the de-scramble after C2 multiplied by the cross interleave per symbol) minus the effects of the first interleaver which can be either 3 or 5 frames minus anything up to 30 symbols depending on where the burst starts in relationship to a frame boundary minus 2 frames due to the post C2 interleaver. So minimum 12304 data bits (256 bits to a frame), maximum 8x8-3-2 = 59 frames - Maybe, my thinking abilities aren't what they used to be.

Regards,
Bob

cdrinfo

Hi Bob,

Thanks for your comments. I think the document says from the beginning that is uses information found around the net, compiled in a way that will provide user an easier way to read it. Of course we have added our knowledge and information and hopefully with your help (and everyone's else) it will become even better...

As you have said many times, the information about CIRC is not very accurate neither very informative, so there are problems. If we had much more money, buying the CD standards could give us the most accurate information

The comments about the LTR-52246S series are correct, we also noticed the big difference with the previous series (and of course with other Sanyo drives).

The big question now is...What each manufacturer returns as C2 error? E32 (as we have posted), E22, or E22+E32 (or something else?)

For example, RICOH said that according to them, C2 = E12+E22 while other manufacturers say C2 = E32! Another example is that the TEAC CD-552E (52X CD-ROM) returned around 10X times higher C2 errors than a LiteOn drive (0.8-1.2X times would make sense) indicating that it returns something else than the LiteOn recorders. The investigation for truth continues...

PS. To give some good news, JITTER tests are on its way

Pio2001

BobHere, I'll put online what I have done so far, even if it's incomplete. It might help you before you sit down and sort all out.

__________________
Pio2001

BobHere

Thanks Pio, I am not really getting it together at the moment and need all the help I can get

Regards,
Bob

BobHere

Actually, probably not. When I was still a practicing Engineer I used to have access to all the (online) ITU-T standards (don't ask about the cost!). Most standards are prescriptive rather than descriptive and just say how it is rather than how it works or even why it is there at all. Parts of the V90 standards were all but incompresensible to all bar a few people in the field but you could use the exact precription to design a V92 modem (but not the in exchange transmitter). You can see this tendency in the publicly available standards on CDs and I doubt that Philips/Sony are any better.

Yes that is a problem, the use of CD doctor helps from the software perspective in that it is not biased but you are still at the mercy of the drives reporting.

I rather like the way the the Philips and Sony standalone chipsets report it. It is quite specific stating exactly what each correction stage achieved. Unfortunately I have never seen a published circuit for reading out this information (which is pretty much the same as the way subchannel data is made available by the chipsets) but this would allow some cheaply built testing equipment, albiet at 1x speed.

I supose the interesting point was the C1 performance, there is much less scope for different reporting here. Also, the interesting point is the older 32x 3 series Liteons performed better than the 40x 48x 5 series Liteon, I doubt that MediaTec would have changed this but then again it is a possibility..... Anyhow if it is truly representative it is a bit irritating for someone who owns 2 LTR-40125Ss.


Good Lord! Do you really think it is out there??

Regards,
Bob

PS.

I think that the intense scrutiny that you are now showing to these drives can only push the art further, faster. Manufacturers may actually take notice and act on designs to improve their drives if the deficits in their performance are exposed. There has been little motivation for them to improve in some areas up till now. Keep pushing!

Considering what Philips achieved ealy on it is disappointing that the quality of drives is not better than it is. After Philips got out of high end audio CD players, the technical excellence of their machines seem to drop off rather badly, they went from 4/4 correction to 2/4 to 2/2 and have only relatively recently returned to 2/4 correction.

cdrinfo

From our point of view, we do everything we push manufacturers build better drives. Everyone wants to get a good review and test results, so they have to improve themselfs...

>Good Lord! Do you really think it is out there??

I am sure that if its out there, we will dig it up!

Back to the test lab now...New tests showing how VariRec, Audiomaster and low recording speeds affecting Jitter are performed, the results will be interesting so the article will be updated soon!

Pio2001

What are those numbers ?
I suppose that the second number is the number of wrong bytes that can be corrected in the C2 stage, isn't it ?

__________________
Pio2001

BobHere

2/4 is 2 corrections in C1, 4 corrections in C2. This is the way it is usually specified in the Chipset datasheets. The 4/4 occurred on early Philips chipsets where there was flag information derived from the EFM decoder. Also, I know of at least one case where 4/4 is specified in a relatively recent Taiwan Chipset (Hynix if memory serves)

Occasionally you will also see C1/C2/C1/C2 or C1>C2>C1>C2 this is a double pass system where the data is sent through each stage twice. This double pass strategy has been known from the beginning but is most commonly used in DVD. It is rare to see it in CD systems despite its apparent advantages.

Regards,
Bob