We study the design of structured Tanner codes with low error-rate floors on the AWGN channel. The design technique involves the "doping" of standard LDPC (proto-)graphs, by which we mean Hamming or recursive systematic convolutional (RSC) code constraints are used together with single-parity-check (SPC) constraints to construct a code's protograph. We show that the doping of a "good" graph with Hamming or RSC codes is a pragmatic approach that frequently results in a code with a good threshold and very low error-rate floor. We focus on low-rate Tanner codes, in part because the design of low-rate, lowfloor LDPC codes is particularly difficult. Lastly, we perform a simple complexity analysis of our Tanner codes and examine the performance of lower-complexity, suboptimal Hamming-node decoders.
Shadi Abu-Surra, Gianluigi Liva, William E. Ryan