LDPC Check-Node Grouping for Faster BCJR Decoder Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional LDPC code decoding methods, such as the two-phase message passing algorithm, suffer from slow convergence speed, which affects communication performance in latency-critical applications like wireless communication systems.
Innovation Solution
A new decoding method that interprets the transmitted codeword as a punctured codeword, using the joint Maximum-A-Posteriori (MAP) probability calculation for groups of check nodes, and updating all check nodes and variable nodes connected to these groups without updating internal parity nodes, improving convergence speed by processing check nodes in groups rather than individually.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional two-phase message passing algorithm is used, then decoding complexity is manageable, but convergence speed is slow
Solution Approach 1:
The patent merges the update operations of multiple check nodes into a single joint update operation. Instead of updating check nodes individually as in conventional two-phase message passing, the invention combines several check node updates into one unified operation, thereby improving convergence speed while maintaining manageable decoding complexity through the structured joint update mechanism
2Productivity
If check nodes are updated individually, then decoding complexity is low, but number of iterations required is high
Solution Approach 1:
The patent combines multiple check node update operations into a single joint update step, reducing the total number of iterations required for convergence. This merging approach improves decoding efficiency by achieving better error correction performance in fewer iterative passes, thereby reducing the time loss associated with multiple iterations
3Speed
If all parity nodes are updated in each iteration, then decoding accuracy is maintained, but convergence speed decreases
Solution Approach 1:
The patent extracts and identifies a specific parity node that connects check nodes within the same group, and excludes this node from updates in the joint operation. By taking out this internal parity node from the update process, the invention achieves faster convergence while maintaining decoding accuracy through the specialized update rule that handles this node differently
4Speed
If joint MAP probability calculation is used for check node groups, then convergence speed improves, but computational complexity increases
Solution Approach 1:
The patent segments the set of check nodes into distinct groups, where check nodes within each group share a common connecting parity node. This segmentation allows the joint MAP probability calculation to be applied efficiently to each group independently, improving convergence speed while managing computational complexity through the structured grouping approach
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Method of decoding LDPC codes such as IRA codes used in the DVB-S2 standard, the code graph comprising first variable nodes (systematic), second variable nodes (parity) having a degree two, and check nodes connected to second variable nodes by a zigzag connectivity, the method comprising a) grouping the check nodes into at least one group, the check nodes of each group being connected by variable nodes called internal variable nodes, b) performing for each group (GRj) the following sub-steps b1) and b2): b1) jointly updating (71) all the check nodes of said group by using an algorithm of the Maximum-A-Posteriori (MAP) type on a two-state trellis, b2) updating (72, 73) all the first variable nodes and all the second variable nodes (PNi-1,i; PNi,i+1) connected to said group except said at least one internal second variable node, c) iteratively repeating step b). When using a plurality of groups, a corresponding plurality of trellis windows are LogMAP decoded without training period since connecting parity nodes are updated according to the sum-product algorithm.