Quasi-Cyclic LDPC Decoding With Dynamic Lifting Factors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current LDPC code systems face challenges in supporting flexible code lengths and code rates, particularly in achieving high encoding performance and avoiding error floors, especially in wireless communication systems where code lengths can vary significantly.
Innovation Solution
The proposed solution involves using a base matrix with specific structures and transformations, such as row/column transformations, and varying lifting factors to generate LDPC matrices that can accommodate different code block lengths, ensuring performance across multiple block lengths and code rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a fixed LDPC matrix structure is used, then the decoding complexity is low, but the system cannot support flexible code lengths and code rates
Solution Approach 1:
The LDPC matrix is segmented into a base matrix H_b and multiple circulant submatrices. Each submatrix can be independently configured with different lifting factors Z, allowing the system to support flexible code lengths while maintaining a structured approach that manages complexity. The base matrix remains fixed while only the circulant submatrices vary to accommodate different code requirements.
Solution Approach 2:
The lifting factor Z is made dynamic and configurable based on the required code length and code rate. Instead of using a fixed matrix structure, the system dynamically adjusts the lifting factor applied to circulant submatrices of the base matrix, enabling adaptation to various code lengths (e.g., 512, 1024, 2048 bits) while preserving the underlying matrix structure for efficient decoding.
2Reliability
If different base matrices are used for different lifting factors, then code performance is optimized, but the system complexity and memory requirements increase
Solution Approach 1:
A single base matrix H_b is designed to serve multiple functions across different code lengths and code rates. The base matrix contains circulant submatrices that can be expanded with different lifting factors Z to generate LDPC matrices suitable for various applications. This universal base matrix approach eliminates the need to store and manage multiple separate base matrices, reducing memory requirements while maintaining optimized performance for different code configurations.
3Adaptability or versatility
If lifting factor Z is increased to support longer code blocks, then the code block length coverage is improved, but the decoding throughput decreases
Solution Approach 1:
The decoding process is segmented into parallel processing units that handle different circulant submatrices independently. By maintaining the circulant structure in the base matrix, the system can exploit the regular pattern to implement efficient parallel decoding algorithms. This allows longer code blocks (achieved through higher lifting factors) to be decoded with throughput comparable to shorter codes, as the parallel structure prevents the decoding complexity from scaling linearly with code length.
4Reliability
If error floor is reduced through matrix optimization, then the reliability is improved, but the design complexity of the base matrix increases
Solution Approach 1:
The base matrix is designed with specific local structures in critical positions to address error floor issues. Certain circulant submatrices are configured with particular shift values and patterns that locally improve the code's error correction capability at low signal-to-noise ratios. This localized optimization approach reduces error floors without requiring complete redesign of the entire base matrix, thereby limiting the increase in design complexity to only the critical submatrices that impact error floor performance.
Data Source
Figure 1
Figure 2
Figure 3a
AI summary
The application concerns decoding of quasi-cyclic LDPC codes on the basis of a parity check matrix of the code, which is obtained from a base matrix in combination with what is known as modulo-lifting using a set of lifting factors Z being {5, 10, 20, 40, 80, 160, 320}, wherein the base matrix and its associated shift coefficients are as shown in Fig. 3b-3.