Neural Network Polar Encoding for Large Codeword Spaces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional encoding and decoding processes for polar codes face high complexity due to the exponential increase in code sequences, making it difficult to traverse the entire codeword space efficiently, especially when dealing with large information bits.
Innovation Solution
An encoding and decoding method using neural networks, where an initial neural network unit is trained and updated to reduce the impact of code length on traversal complexity, allowing for generalization across the entire codeword space using a small learning sample.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the code length increases to handle larger information bits, then the encoding/decoding capability is improved, but the complexity of traversing the entire codeword space increases exponentially
Solution Approach 1:
The patent divides the entire codeword space traversal into multiple smaller subsets. Instead of traversing all 2^N possible codewords sequentially, the neural network processes subsets of codewords in parallel, significantly reducing the effective traversal complexity while maintaining the ability to handle large information bits.
Solution Approach 2:
The patent replaces the mechanical sequential traversal approach with a neural network-based parallel processing system. The neural network learns optimal encoding/decoding patterns through training, substituting the brute-force mechanical traversal with intelligent pattern recognition that scales better with code length.
2Measurement precision
If the entire codeword space is traversed to ensure optimal encoding/decoding performance, then the accuracy is improved, but the learning time and computational resources increase significantly
Solution Approach 1:
The patent applies partial action by processing subsets of the codeword space rather than the entire space. The neural network is trained on representative subsets that capture the essential patterns, achieving sufficient accuracy without the prohibitive cost of exhaustive traversal of all possible codewords.
Solution Approach 2:
The patent performs preliminary training on smaller, manageable subsets of data to build the neural network model. This preliminary action on subsets prepares the model for efficient inference, avoiding the need to traverse the entire codeword space during actual encoding/decoding operations.
Data Source
AI summary
The present disclosure relates to encoding methods and apparatus, and decoding methods and apparatus. In one example encoding method, first input information is obtained. The first input information is encoded based on an encoding neural network to obtain and output first output information. The encoding neural network comprises a first neuron parameter, and the first neuron parameter is used to indicate a mapping relationship between the first input information and the first output information.