Parameterized Message Passing Decoder for Short Dense Codes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Iterative error correction coding schemes, such as LDPC codes, perform poorly for short codes due to assumptions of statistical independence between messages, which become inaccurate as codes become shorter and denser, leading to degraded performance.
Innovation Solution
A parameterized iterative message passing decoder is developed that learns parameters to adapt to statistical dependencies in the code's graph, using machine learning to train the decoder on specific graphs, data properties, and physical properties of memory devices, improving read throughput and convergence speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional iterative message passing decoding is used, then the decoder is simple to implement, but performance degrades for short and dense codes due to inaccurate statistical independence assumptions
Solution Approach 1:
The patent modifies the conventional message passing decoding by introducing learnable parameters into the message computation rules. These parameters are trained using machine learning to account for statistical dependencies in short and dense codes, thereby improving decoding performance without fundamentally changing the iterative decoding architecture.
Solution Approach 2:
The patent replaces the fixed mathematical computation rules in conventional decoders with machine learning-based parameterized rules. This substitution allows the decoder to adapt to specific code properties and statistical dependencies that conventional fixed rules cannot capture, particularly for short and dense codes.
2Reliability
If code length is increased to improve error correction performance, then decoding accuracy improves, but latency and computational complexity increase
Solution Approach 1:
By training parameters on specific code configurations and properties, the decoder can achieve better error correction performance with shorter codes. The learned parameters compensate for the reduced code length by adapting to the specific statistical characteristics of short and dense codes, thereby maintaining reliability without increasing latency.
3Speed
If code density is increased to reduce code length, then latency is reduced, but decoding performance degrades due to stronger statistical dependencies
Solution Approach 1:
The patent replaces fixed computation rules with machine learning-based parameterized rules that can adapt to the statistical dependencies introduced by higher code density. This allows the decoder to maintain good performance even when codes are short and dense, thereby preserving both speed and reliability.
4Ease of manufacture
If fixed computation rules are used in the decoder, then the decoder is easy to implement, but it cannot adapt to varying physical properties of memory devices
Solution Approach 1:
The patent introduces learnable parameters into the message passing computation rules, allowing the decoder to adapt to varying physical properties of memory devices. These parameters are trained offline on data representing different physical conditions, enabling the decoder to handle variations without changing its fundamental structure or implementation complexity.
Solution Approach 2:
The patent performs parameter training in advance during an offline phase, before actual decoding operations. This preliminary action allows the decoder to be pre-adapted to specific code properties and memory device characteristics, so that during online operation, the decoder can directly use the pre-trained parameters without requiring real-time adaptation, thus maintaining ease of implementation.
Data Source
AI summary
Technology is described herein for learning parameters for a parameterized iterative message passing decoder, and to a corresponding parameterized iterative message passing decoder. Learning the parameters may adapt the decoder to statistical dependencies introduced by the specific code's graph. Taking into account the statistical dependencies may allow the code to be shorter and/or denser. Note that the statistical dependencies in the graph may be extremely complex. Machine learning may be used to learn the parameters. The parameters may be learned when decoding data stored in the memory device. Learning the parameters may adapt the decoder to properties of data stored in the memory device, physical properties of the memory device, and/or patterns in host data.