Non-Binary LDPC Decoding with Reduced Memory and FFT Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing decoder structures for flash memory devices, particularly those using multi-level cell (MLC) technology, face challenges with high complexity, computational expense, and large memory requirements, which affect the endurance and retention of flash memory devices.
Innovation Solution
A decoding algorithm and structure that reduces complexity and memory needs by using confidence vectors with multiple components, passing selected vector components between variable and check nodes, and updating these vectors based on return values, eliminating the need for fast Fourier transform (FFT) computations and reducing memory requirements by storing only essential message components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If FFT-based decoding methods are used for non-binary LDPC codes, then decoding performance is improved, but computational complexity and memory requirements increase significantly
Solution Approach 1:
The patent extracts only the essential components needed for decoding by passing selected vector components (indices) between variable and check nodes rather than transmitting complete confidence vectors. This extraction principle reduces the data volume and computational burden while maintaining the core decoding functionality, directly addressing the computational complexity issue of FFT-based methods
Solution Approach 2:
The patent replaces expensive FFT computations with simpler, disposable operations such as selecting maximum/minimum values from confidence vectors and performing basic arithmetic operations. These simplified operations are computationally inexpensive and can be performed iteratively without requiring the heavy computational resources of FFT-based approaches
2Reliability
If FFT-based decoding methods are used for non-binary LDPC codes, then decoding performance is improved, but memory requirements increase significantly
Solution Approach 1:
The patent extracts and transmits only the index of the selected confidence vector component rather than the entire confidence vector between nodes. This extraction dramatically reduces the memory bandwidth and storage requirements, as only compact index values need to be stored and transmitted instead of full floating-point confidence vectors
Solution Approach 2:
The patent segments the confidence vector information into two separate parts: the confidence values (stored locally at each node) and the indices (transmitted between nodes). This segmentation allows the system to maintain decoding performance while reducing memory requirements, as only the essential index information needs to be communicated across the decoder architecture
3Measurement precision
If complete confidence vectors are passed between variable and check nodes, then decoding accuracy is improved, but communication overhead and memory usage increase
Solution Approach 1:
The patent extracts the most critical information from the complete confidence vector by identifying and transmitting only the index of the selected component (e.g., the maximum or minimum value). This extracted index contains the essential directional information needed for decoding accuracy while reducing data transmission volume from O(Q) to O(1) per message, where Q is the alphabet size
Solution Approach 2:
The patent uses the index as an intermediary representation that mediates between the full confidence vector and the decoded value. Instead of transmitting complete vectors, the index serves as a compact mediator that preserves the essential information needed for accurate decoding while minimizing communication overhead
Data Source
AI summary
A custom decoding algorithm is introduced, in connection with a corresponding decoding structure, to resolve many complexity and large memory requirements of known decoders. A system. A variable node forms a confidence vector of four components, one component for each state of a memory cell, and passes the current primary component (e.g., the largest) to one or more check nodes. The check nodes calculate a temporary component and corresponding index based on all components received from the variable nodes, which they pass back to the respective variable nodes. The variable node updates the confidence vector based on the temporary nodes received from respective check nodes, and determines the correct state for corresponding memory cell based on which component in the confidence vector is currently the primary component.