Applications in the field of digital communications are becoming increasingly complex and diversified. Hence, the need to correct the transmitted message mistakes becomes an issue to be dealt with. To address this problem, error correcting codes are used. In particular, Polar Codes that are the subject of this thesis. They have recently been discovered (2008) by Arikan. They are considered an important discovery in the field of error correcting codes. Their practicality goes hand in hand with the ability to propose a hardware implementation of a decoder. The subject of this thesis focuses on the architectural exploration of Polar Code decoders implementing particular decoding algorithms. Thus, the subject revolves around two decoding algorithms: a first decoding algorithm, returning hard decisions, and another decoding algorithm, returning soft decisions.The first decoding algorithm, treated in this thesis, is based on the hard decision algorithm called "successive cancellation" (SC) as originally proposed. Analysis of implementations of SC decoders shows that the partial sum computation unit is complex. Moreover, the memory amount from this analysis limits the implementation of large decoders. Research conducted in order to solve these problems presents an original architecture, based on shift registers, to compute the partial sums. This architecture allows to reduce the complexity and increase the maximum working frequency of this unit. We also proposed a new methodology to redesign an existing decoder architecture, relatively simply, to reduce memory requirements. ASIC and FPGA syntheses were performed to characterize these contributions.The second decoding algorithm treated in this thesis is the soft decision algorithm called SCAN. The study of the state of the art shows that the only other implemented soft decision algorithm is the BP algorithm. However, it requires about fifty iterations to obtain the decoding performances of the SC algorithm. In addition, its memory requirements make it not implementable for huge code sizes. The interest of the SCAN algorithm lies in its performances which are better than those of the BP algorithm with only two iterations. In addition, its lower memory footprint makes it more convenient and allows the implementation of larger decoders. We propose in this thesis a first implementation of this algorithm on FPGA targets. FPGA syntheses were carried out in order to compare the SCAN decoder with BP decoders in the state of the art.The contributions proposed in this thesis allowed to bring a complexity reduction of the partial sum computation unit. Moreover, the amount of memory required by an SC decoder has been decreased. At last, a SCAN decoder has been proposed and can be used in the communication field with other blocks requiring soft inputs. This then broadens the application field of Polar Codes.
展开▼
