Issue link: https://read.uberflip.com/i/1464683
LoRaWAN ® Fragmented Data Block Transport Specification TS004-2.0.0 ©2022 LoRa Alliance ® Page 29 of 32 The authors reserve the right to change specifications without notice. 741 It can be seen that when the number of coded fragments received equals M, the matrix 742 cannot be inverted (is not of rank M) for 70% of the cases. But this probability falls very 743 rapidly once only a few additional fragments are received. With M+7 fragments, the matrix 744 can be inverted in 99% of the occurrences. On average, it takes M+2 coded fragments to 745 recover the original data block. Therefore, the proposed fragmentation works better (with a 746 lower statistical overhead) with a larger number of fragments. With a fixed overhead, this 747 coding scheme provides close-to-ideal performance when the number of fragments is large 748 (>100), because adding 2 fragments to a 100-fragment session only represents 2% relative 749 statistical overhead. The overhead increases when the number of fragments is lower. This 750 coding scheme should not be used for less than 20 fragments because the average 751 overhead is 10% in this case. 752 A.4 End-Device Memory Requirement 753 754 This coding scheme has been optimized to allow the transportation of a large binary file with 755 minimum memory overhead on the receiving end (performing the defragmentation). 756 757 The following formula gives the decoding memory requirements for an optimized end-device 758 implementation expressed in octets, on TOP of the memory required to store the final 759 reconstructed data block: 760 761 Parity matrix memory (octets) = × ( + 1)/2/8 + 2 × 762 763 For a data block fragmented using M fragments, is the number of coded fragments lost by 764 the end-device among the first M coded fragments (fragments P M 1 to P M M ). 765 766