Not having waded through the DVD error detection algorithm with a comb, I must ask:

Based on what has been said, Mediatek chipset can only report PIF, not PO.

A PIF error is an erroneus uncorrectable row in an ECC block.

A single row has several columns of PO EDC data.

So, a single PIF can result in multiple POs.

Right?

A PO error is likely to be fixed if only a few rows were unfixable.

So, is there a statistical correlation between number of PIFs (number of uncorrectable erroneus rows in a block) and the likelihood of getting a POF (a PO error that cannot be corrected for)?.

If so, has anybody plotted this?

regards,
halcyon

As KProbe author have said:

1. MediaTek Chip returns the PI/PO value every 1 ECC block
2. KProbe1 collects 8 PI/PO values and sums them to a PI/PO record.
3. KProbe2 collects 8 PI values and sums them to a PI record. And it saves every PO value as a PO record.

Chip returns the data every 1 ECC block(16 blocks). But chip is passive, Kprobe must issue command to get the data. But it is impossible for kprobe to issue command every 1 ECC block(16 blocks).. the best situation is exactly 1 ECC block(16 blocks), but most of situations is about 19,20 even 30 blocks at the end of disc. So Kprobe will lose its precision in low performance system ,especially when the max speed is selected.

(in reply to Halc)

Just to ask your question, according to specs:

* According to the ECMA 339, a row of an ECC Block that has at least 1 byte in error constitutes a PI error. If a row of an ECC Block contains more than 5 erroneous bytes, the row is said to be “PI-uncorrectable”.
* The total PI errors within 8 consecutive ECC sectors (PI sum 8) should not be greater than 280. This is, as it were, the error rate of the raw data. The lower it is, the more headroom is available to cope with wear and tear from scratches, marks and ageing due to heat or sunlight.
* The DVD+R specification (ECMA 349) furthermore prescribes that in any ECC Block the number of PI-uncorrectable rows should not exceed 4, after the first, inner parity, error correction (Parity Inner Fails, PIF)
* Errors that can’t be corrected by the second, outer parity, correction are finally the ones that a user will notice as actual data errors. Values greater than zero are not acceptable for the POF (Parity Outer Fails).

(in reply to emperor)

Some corrections/clarifications.

1. Mediatek chipset counts Parity Inner Errors (PIE) and Parity Inner Failuers (PIF) passively and those measures can be sampled at the maximum rate of c. 1 ECC block/sample interval

2. ECC Block = 16 scrambled main data frames (not really blocks).

3. We have an error in our thinking by enforcing PO limit of max 4 for DVD+R discs. PO is not the same as PIF, like I have explained above. We should enforce a maximum of 4 PIF (i.e. PI Unc.) / 1 ECC block. The PO rate is usually higher than PIF rate, because one PIF can be a failure of several bytes in one row. Each byte of failure in that row results in one additional PO.

So yes, I've read the specs that far, but not the level of the ECD/ECC algorithms themselves nor have I statistically modelled the distribution.

I'm more interested in stastitical correlation between PIF and PO, because it seems that most drives do not report PIF, but report PO (or vice versa).

Guess I should plot it myself, if only I had the time/skill...

regards,
halcyon

(in reply to emperor)

Halc, sorry i have no more data about PIF/PO statistical correlation, i guess by using our future test results, you could make some assumptions.

(in reply to Halc)

Although I may seem an unlikely source for this information

It was confirmed by a programer within LiteON that although the errors are referred to as PO, they are indeed PIF (on the mediatek chipset). He then clarified that the Mediatek chipset (as it stands I suppose) is incapable of testing for PO errors.

(in reply to emperor)

We already knew that Dolphy and said it most times that it cannot report PO errors..I just re-posted what KProbe author said.

(in reply to Dolphinius_Rex)

Thanks Dolph! I had mislabelled it in my second post and it's good to get a confirmation too.

I'm still worried about this 4 PO limit, which is IMHO mistaken/wrong. We should be limiting the number of PIF (i.e. rows that fail parity inner error correction).

Of course, as each PIF is always at least a single PO, this means that ( PO > PIF) always.

So limiting to 4 PO is actually too strict, but I guess it's better than having a too lax limit.

cheers,
halcyon

(in reply to emperor)