Examination Committee

Prof Volkan KURSUN, ECE/HKUST (Chairperson)
Prof Chi Ying TSUI, ECE/HKUST (Thesis Supervisor)
Prof Wei ZHANG, ECE/HKUST

Abstract

Polar codes, recently discovered by Arikan, are a breakthrough in the coding community as they provably achieve the symmetric capacity and have low encoding and decoding complexities. List Successive Cancellation Decoding (LSCD) is the most popular decoding method for polar codes, because it achieves a good error-correcting performance with reasonable complexity. Hence, LSCD attracts a lot of research interest recently and many efforts have been put on its design and implementation.

An LSCD with list size L retains L source word candidates and each is decoded with a successive cancellation decoder (SCD). SCD decodes the source word in a sequential manner as the decoding of a source word bit depends on the value of the partial sums, which are calculated from the bits that have already been decoded. For LSCD, a partial-sum network (PSN), which consists of L partial-sum generators (PSGs), is needed to generate the partial sums and coordinate the updating of the L source word candidates.

To achieve a very good performance, long code length and a large list size L are needed for LSCD. However the existing PSN architectures scale linearly (or super-linearly) with the code length and list size, and hence induce significant area and delay overhead for polar codes with long code length or large list size. In this thesis, a list high performance partial-sum network (LHPPSN) is proposed based on a folded partial-sum network (PSN) architecture of which the complexity does not scale with the code length. A design based on replicating units of high performance PSN and lazy copying is presented, first. Then, to improve the area and performance, some new methods based on indices copying and management are proposed. In addition, by exploiting the properties of the PSN, the complexity is further reduced by sharing the path copying logic of the PSN. Experimental results show that the proposed LHPPSN improves the area and delay significantly especially for long code and large list size, when compared to the state-of-the-art LSCD architectures.