A generalization of polar codes is proposed, which allows some of the frozen symbols, called dynamic frozen symbols, to be data-dependent. The obtained codes can be considered as subcodes of either algebraic codes (e.g. extended BCH), or classical polar codes. Algebraic and randomized constructions of polar subcodes are presented. The proposed codes are shown to outperform polar codes with CRC and LDPC codes. Furthermore, the proposed generalization enables application of decoding techniques developed for polar codes to other linear block codes. In particular, this allows one to perform near-ML decoding of short extended BCH codes with lower complexity compared to state-of-the-art algorithms.
Dr. Peter V. Trifonov received his PhD from Saint-Petersburg Polytechnic University in 2005. His research interests include coding theory and its applications in communication and storage systems. Currently he is an Associate Professor at Saint-Petersburg Polytechnic University.