Performance Analysis of Improved LT Codes with the Removal of Low Degree Input Symbols

Luby transform (LT) codes suffer from undesired error floors since a certain amount of input symbols may remain uncovered by output symbols due to the uniformly random encoding process. Several variants of LT codes presented previously aiming at removing low-degree input symbols assure that all input symbols can be covered by output symbols after encoding in high probability and have been verified to effectively reduce error floors for short and moderate block lengths. However, in this work, we show that those variants might not be asymptotically good even with the removal of low-degree input symbols in the encoding process.