Ultra-low-latency Soft-decision Decoder Architecture using OSD approach
- Title
- Ultra-low-latency Soft-decision Decoder Architecture using OSD approach
- Authors
- 김창현
- Date Issued
- 2022
- Publisher
- 포항공과대학교
- Abstract
- The ordered statistic decoding (OSD) algorithm for short-length linear block codes provides an attractive ML approaching performance, expected to be used for the ultra-reliable low latency communication (URLLC) at the next-generation wireless solutions. To find the corrected codeword among numerous candidates, however, the decoding process requires a considerable amount of computational costs, which need to be simplified to achieve low-latency processing. In this paper, we present several optimization schemes that relax the overall complexity of the state-of-the-art segmentation discarding algorithm. Without degrading the overall error-correcting capability, our approaches basically approximate the internal steps for calculating the segment boundary and the discarding threshold. The low-latency decoder architecture is also introduced to support the proposed simplified OSD algorithm. First, we introduce the stopping rule by applying the hard-decision BCH decoder to reduce the overall decoding latency. Furthermore, we use the optimized Gaussian elimination architecture with the overlapped sorter instead of using compute-intensive serialized Gaussian elimination. In addition, the reprocessing architecture is newly proposed to support the reprocessing operation. By using our simplified algorithm, the computational cost is reduced by 2x10^5 compared to the conventional OSD algorithm. The implementation results using 28-nm CMOS technology show that the proposed OSD architecture improves the decoding latency by 28.1 times compared to the baseline realization, achieving a throughput of 631 Mbps.
- URI
- http://postech.dcollection.net/common/orgView/200000597564
https://oasis.postech.ac.kr/handle/2014.oak/112172
- Article Type
- Thesis
- Files in This Item:
- There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.