LDPC Decoding with Weighted Message Passing and Reduced Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing decoding methods for low density parity check (LDPC) codes face performance penalties due to quantization in hard decision decoding and resource-intensiveness in soft decision decoding, particularly requiring complex belief propagation and node searches.
Innovation Solution
A hybrid decoding method that employs repetitive message passing with variable node output messages including a 'best guess' and confidence weights, and check node output messages with predetermined magnitudes, simplifying computations and eliminating the need for node searches and conventional belief propagation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If hard decision decoding with quantization is used, then decoding complexity is reduced, but decoding performance deteriorates
Solution Approach 1:
The patent changes the parameter representation from hard quantized bits to soft information in the form of log-likelihood ratios (LLRs). Each received bit is represented by an LLR value that preserves the magnitude of confidence, allowing the decoder to operate on continuous-valued soft information rather than quantized hard decisions. This parameter transformation maintains decoding performance while enabling simpler iterative decoding algorithms.
2Reliability
If soft decision decoding with belief propagation is used, then decoding performance is improved, but computational resources and complexity increase
Solution Approach 1:
The patent extracts only the essential information needed for decoding from the full belief propagation algorithm. Instead of computing and passing complete probability distributions between check nodes and variable nodes, the method extracts and utilizes only the log-likelihood ratio values and their magnitudes. This extraction simplifies the computational burden while retaining the performance benefits of soft decision decoding.
Solution Approach 2:
The patent employs simplified check node operations that can be implemented with low computational cost. The check node messages are computed using simple functions of the incoming LLRs without requiring complex probability calculations. These simplified operations can be performed efficiently in each iteration, making the overall decoding process computationally feasible while maintaining good performance.
3Device complexity
If conventional bit-flipping methods are used, then decoding complexity is reduced, but decoding performance is insufficient
Solution Approach 1:
The patent introduces dynamic adaptation in the decoding process by using iterative message passing where the check node messages are updated based on the current state of variable nodes. The magnitude of check node messages dynamically adjusts according to the confidence levels of incoming variable node messages, allowing the decoder to adapt its behavior to the specific error patterns in each received codeword. This dynamic approach significantly improves performance over static bit-flipping methods.
Data Source
AI summary
An improved decoder and decoding method for low density parity check (LDPC) codes is provided. Decoding proceeds by repetitive message passing from a set of variable nodes to a set of check nodes, and from the check nodes back to the variable nodes. The variable node output messages include a “best guess” as to the relevant bit value, along with a weight giving the confidence in the guess (e.g., weak, medium or strong). The check node output messages have magnitudes selected from a predetermined set including neutral, weak, medium and strong magnitudes. The check node output messages tend to reinforce the status quo of the input variable nodes if the check node parity check is satisfied, and tend to flip bits in the input variable nodes if the check node parity check is not satisfied. The variable node message weights are used to determine the check node message magnitudes.