Improving LT Decoding for Real-Time Applications

Luby Transform (LT) codes are shown to perform very well over erasure wireless channels. However, LT codes require the number of input symbols to be sufficiently high or otherwise the coding overhead becomes significant. With large number of input symbols, the packets latency and decoding complexity as well as buffer requirement may increase. Given that Gaussian elimination is computationally intensive, this paper proposes to additionally consider higher order encoded symbols during the decoding procedure. The results show up to 47% overhead reduction when degree-2 symbols are as well contributing in the decoding procedure. The proposed method is very straightforward and requires no change to the encoder and minimal modification of the decoder.