Self-Adaptive LDPC Decoder Using Idle Accelerators for LLR Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional LDPC decoder algorithms face performance degradation due to inaccurate LLR scaling, which is challenging to optimize for varying channel conditions, leading to suboptimal wireless performance.
Innovation Solution
A self-adaptive LDPC decoder architecture that utilizes machine learning to predict optimal scaling factors by gathering posteriori decoding metrics from unused LDPC decoder accelerators, enabling continuous learning and adaptation to channel conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional LDPC decoder algorithms use fixed LLR scaling, then device complexity is reduced, but wireless performance degrades due to inaccurate scaling under varying channel conditions
Solution Approach 1:
The patent implements dynamic LLR scaling by training a neural network model that adapts scaling factors based on channel conditions. The model is trained using a priori channel statistics and posteriori decoding metrics, enabling the decoder to dynamically adjust scaling parameters rather than using fixed values, thereby improving wireless performance under varying channel conditions
Solution Approach 2:
The system employs unused LDPC decoder accelerators to perform self-training by gathering posteriori decoding metrics and updating the neural network model. This self-service mechanism allows the decoder to continuously improve its scaling accuracy using its own operational data without requiring external intervention or additional hardware resources
2Reliability
If LLR scaling is optimized for each channel condition, then wireless performance improves, but the difficulty of detecting and measuring optimal scaling increases
Solution Approach 1:
The patent introduces a neural network model as an intermediary that learns the complex relationship between channel conditions and optimal LLR scaling factors. The model translates channel statistics and decoding metrics into appropriate scaling parameters, avoiding the need for direct measurement and detection of optimal scaling under each channel condition
Solution Approach 2:
The system performs preliminary training of the neural network model using a priori channel statistics before actual decoding operations. This pre-learning phase prepares the model to quickly adapt to different channel conditions without requiring real-time optimization, thereby reducing the difficulty of detecting optimal scaling during operation
3Adaptability or versatility
If machine learning training is performed continuously, then adaptation to channel conditions improves, but loss of time for real-time decoding increases
Solution Approach 1:
The patent implements periodic training using idle periods when LDPC decoder accelerators are not actively decoding data. The system alternates between decoding operations and training phases, allowing the model to adapt to channel conditions over time without interfering with real-time decoding performance
Solution Approach 2:
The system maintains continuous adaptation by utilizing otherwise unused decoder accelerators for training during idle periods. This ensures that the learning process continues without interrupting the primary decoding function, keeping both adaptation and real-time performance operational simultaneously
Data Source
AI summary
A low-density parity check (LDPC) decoder architecture is provided for a self-adapting an LDPC control input that is used to successfully decode corresponding received code blocks. The architecture enables a machine-learning based process that facilitates learning of optimal LDPC control inputs, such as log-likelihood (LLR) terms scaling, that is required for a given channel condition and deployment scenario. This is achieved by gathering, as a background process, a posteriori decoding metrics via unused LDPC decoder accelerators, which function to process the labelled LLR data sets for different LDPC control input values. Then, such optimal LDPC control input estimates may be applied to the real-time LDPC decoding based on previous learning across UE/channel conditions.