Research Of Decoder Of LDPC Codes Based On FPGA

LDPC codes are a class of channel codes with strong error correction capability.The intrinsic characteristic is that these codes can obtain better performance and lower cost as the processing technology develops,therefore it may reduce possibility of the new technologies replacement in the future.This thesis introduces the definition of QC-LDPC codes and related theories,listing the belief propagation decoding algorithm and other algorithms.Combining the advantages and disadvantages of these several algorithms,check nodes self-update algorithm is applied in this thesis.Compared with other algorithms,the performance of new algorithm is slightly reduced,but the resource consuming is greatly reduced.For the disadvantages of large memory resource consumption and low throughput of serial decoding structure and full parallel decoding structure.This thesis adopts a layered partial parallel decoder structure.The structure maintains a good compromise between storage resources and throughput.The layered partial parallel decoder includes a receive channel information storage module,a node message storage module,a check node update module,a decode output module,and a control module.The receive channel information storage module and the decode result storage module both adopt the ping-pong structure,which better improves the efficiency of the decoder.The check node storage module uses an optimized compression bit-width storage method to store the check node information,which can simultaneously support three kinds of code rates on the premise of saving storage resources,and increase the flexibility of the decoder.The check node storage module updating model uses a three-stage pipeline structure,which reduces latency and increases operation frequency.The control module uses a limit state machine to reduce the complexity of the decoder.The simulation results show that the layered partial parallel decoder works at 308.2 MHz,and the throughput capacity can reach 516.8 Mbps when the maximum number of iterations is 20.