A Design Of Multidimensional Data Reconciliation System Based On CPU/GPU Heterogeneous Platform

Data reconciliation is an important part of the Continuous Variable Quantum Key Distribution(CVQKD)system.The reconciliation efficiency and the speed of the reconciliaton greatly affect the practicability of the CVQKD system.For the multidimensional data reconciliation of CVQKD,the Low Density Parity Check(LDPC)error correction code used can approach the Shannon limit when the code length is long,thus improving the reconciliaton efficiency.But this brings about a problem of large amount of calculation in the reconciliation part,especially the multiple iterations of the decoding part,the increase in time cost.As a result,the reconciliation speed is slowed down and the practicality of the CVQKD system is reduced.What this paper needs to do is to improve reconciliation speed while ensuring reconciliation efficiency.The main work is as follows :1.Using the idea of parallel computing,the CVQKD multidimensional data reconciliation acceleration is realized on the CPU/GPU heterogeneous platform.The specific process is to hand over the complex logical operation part of the data reconciliaton to the CPU for serial execution,and hand over the logic simple and data-intensive decoding part to the GPU for parallel execution.For the GPU decoding part,variable node initialization,check node message update and variable node message update three kernel functions are designed to realize parallel acceleration decoding.2.The parameters of the kernel function in CUDA do not allow the use of three-level pointers,and the large-scale sparse check matrix is transformed into a static two-way cross-linked list storage that meets CUDA requirements.The specific method is to store the sparse check matrix and the non-zero node by constructing two structures respectively,and change the position of the adjacent node in the non-zero node structure to the int data type.This storage method only stores 1 location greatly reduces the memory footprint.