This page gives some information on the erasure recovery performance that can be expected from LDPC-Staircase codes. It follows that N1=7 is usually a good choice.
| conditions | average overhead | overhead for a failure probability < 10-4 |
|---|---|---|
| CR=2/3, N1=5 | 0.636% | with 2.1% overhead (22 extra symbols), Pr_fail=5.9*10-5 |
| CR=2/3, N1=7 | 0.238% | with 1.5% overhead (15 extra symbols), Pr_fail=8.2*10-5 |
| CR=1/2, N1=5 | 01.228% | with 3.1% overhead (28 extra symbols), Pr_fail=7.2*10-5 |
| CR=1/2, N1=7 | 0.378% | with 1.7% overhead (17 extra symbols), Pr_fail=9.6*10-5 |
Results are excellent and approach that of an ideal code where the overhead is 0. The details are given in the curves below, and one can see that if an error floor is still visible when using N1=5, it is no longer visible (i.e., if any is below 10-5) with N1=7. Note that higher values of N1 can still be used if needed...
This curve indicates the probability decoding fails as a function of the overhead (said differently the number of symbols received). Here code rate=2/3; N1=5 or 7.
(NB: don't be fooled by the horizontal scale, an extra 40 symbols only represents a 3.9% overhead when dealing with k=1024. This curve is therefore a close-up.)
(NB2: on the left of the vertical line, no decoding is possible since fewer than k symbols have been received)
Same conditions as above for code rate=1/2.
This curve indicates the probability decoding fails as a function of the loss rate (said differently the number of symbols received/n). Here code rate=2/3; N1=5 or 7.
(NB: on the right of the vertical line, no decoding is possible since fewer than k symbols have been received)
Same conditions as above for code rate=1/2.