System Development and VLSI Implementation of High Throughput and Hardware Efficient Polar Code Decoder

Polar code is the first channel code which is provable to achieve the Shannon capacity. Additionally, it has a very good performance in terms of low error floor. All these merits make it a potential candidate for the future standard of wireless communication or storage system. Polar code is received increasing research interest these years. However, the hardware implementation of hardware decoder still has not meet the expectation of practical applications, no matter from neither throughput aspect nor hardware efficient aspect. This dissertation presents several system development approaches and hardware structures for three widely known decoding algorithms. These algorithms are successive cancellation (SC), list successive cancellation (LSC) and belief propagation (BP). All the efforts are in order to maximize the throughput meanwhile minimize the hardware cost. Throughput centric successive cancellation (TCSC) decoder is proposed for SC decoding. By introducing the concept of constituent code, the decoding latency is significantly reduced with a negligible decoding performance loss. However, the specifically designed computation unites dramatically increase the hardware cost, and how to handle the conventional polar code sets and constituent codes sets makes the hardware implementation more complicated. By exploiting the natural property of conventional SC decoder, datapaths for decoding constituent codes are compatibly built via computation units sharing technique. This approach does not incur additional hardware cost expect some multiplexer logic, but can significantly increase the decoding throughput. Other techniques such as pre-computing and gate-level optimization are used as well in order to further increase the decoding throughput. A specific designed partial sum generator (PSG) is also investigated in this dissertation. This PSG is hardware efficient and timing compatible with proposed TCSC decoder. Additionally, a polar code construction scheme with constituent codes optimization is also presents. This construction scheme aims to reduce the constituent codes based SC decoding latency. Results show that, compared with the state-of-art decoder, TCSC can achieve at least 60% latency reduction for the codes with length n = 1024. By using Nangate FreePDK 45nm process, TCSC decoder can reach throughput up to 5.81 Gbps and 2.01 Gbps for (1024, 870) and (1024, 512) polar code, respectively. Besides, with the proposed construction scheme, the TCSC decoder generally is able to further achieve at least around 20% latency deduction with an negligible gain loss. Overlapped List Successive Cancellation (OLSC) is proposed for LSC decoding as a design approach. LSC decoding has a better performance than LS decoding at the cost of hardware consumption. With such approach, the l (l > 1) instances of successive cancellation (SC) decoder for LSC with list size l can be cut down to only one. This results in a dramatic reduction of the hardware complexity without any decoding performance loss. Meanwhile, approaches to reduce the latency associated with the pipeline scheme are also investigated. Simulation results show that with proposed design approach the hardware efficiency is increased significantly over the recently proposed LSC decoders. Express Journey Belief Propagation (XJBP) is proposed for BP decoding. This idea origins from extending the constituent codes concept from SC to BP decoding. Express journey refers to the datapath of specific constituent codes in the factor graph, which accelerates the belief information propagation speed. The XJBP decoder is able to achieve 40.6% computational complexity reduction with the conventional BP decoding. This enables an energy efficient hardware implementation. In summary, all the efforts to optimize the polar code decoder are presented in this dissertation, supported by the careful analysis, precise description, extensively numerical simulations, thoughtful discussion and RTL implementation on VLSI design platforms