Research On High Speed Decoding Of LDPC Code For CV-QKD

In a continuous variable quantum key distribution(CV-QKD)system,the two parties in the communication need to share the key under conditions of long distance and low signal-to-noise ratio.Therefore,the codes with low code rates and long block lengths are required to ensure effective communication between the two parties.In this thesis,we proposed three high-speed graphics processing unit(GPU)based decoders for decoding the multi-edge type low density parity check code with the quasi-cyclic structure(QC-MET-LDPC);that is,the flooding belief propagation(BP)decoder,the layered BP decoder in which threads map check nodes(called CN-type layered BP decoder)and the layered BP decoder in which threads maps variable nodes(called VN-type layered BP decoder).The decoders store the required base matrix information in a combined way for reducing the GPU memory consumption.Three decoders are applied to a code with length of 610,rate of 0.1,and an expansion factor of 2500.Simulation results show that the flooding BP decoder achieves a throughput of 30.32Mbits/s when decoding 64 codewords simultaneously without the early termination,and 100 iterations.The scheduling of algorithm has a great impact on the decoder's convergence rate,the complexity and the error correction performance.The two kind of decoders were implemented with a layered belief propagation algorithm(LBPA)and obtained a faster convergence than the flooding BP algorithm,which reduced the number of required decoding iterations.Both decoders were used to decode a QC-MET-LDPC code with a large expansion factor and thus reduced the number of sub-matrices.The CN-type layered BP decoder allocates GPU threads mapping to check nodes,while the VN-type layered BP decoder allocates GPU threads mapping to variable nodes.Simulation results demonstrate that the former and the latter achieved 64.11Mbits/s and 41.50Mbits/s when decoding 128 and 16 codewords at the same time,respectively,under the condition of 50 iterations without early termination.It can also be seen from simulation results that the flooding BP decoder has the largest GPU thread utilization but the lowest decoding speed among three decoders,while the CN-type layered BP decoder has the fastest decoding speed but the lowest GPU threads utilization among them.