Neural Network Read Channel Nodes for Data Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data detection methods in read channels, particularly in data storage and transmission devices, face inefficiencies in error rate performance due to reliance on outdated techniques such as PRML and LDPC, which limit the adaptability and accuracy of signal processing.
Innovation Solution
The implementation of interconnected neural network circuits that can be separately trained to preprocess and modify read signals for soft output detectors, allowing for improved data detection and adaptation through dynamic training and retraining during runtime operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional PRML and LDPC detection schemes are used in read channels, then device complexity is reduced and ease of manufacture is improved, but data detection accuracy and error rate performance deteriorate
Solution Approach 1:
The patent divides the read channel into multiple independent neural network circuits, each performing specific detection functions. This segmentation allows parallel processing of different signal aspects (equalization, detection, decoding) simultaneously, improving detection accuracy while distributing computational complexity across modular components rather than requiring a single complex detector.
Solution Approach 2:
The patent replaces traditional signal processing algorithms (PRML, LDPC) with neural network-based processing. The neural networks learn optimal detection strategies from training data and adapt to varying channel conditions, providing superior detection accuracy compared to fixed algorithmic approaches while managing complexity through hardware implementation of the trained networks.
2Adaptability or versatility
If a single neural network is used for read signal processing, then device complexity is reduced, but adaptability to changing data conditions deteriorates
Solution Approach 1:
The patent implements multiple specialized neural network circuits (equalization network, detection network, decoding network) that can be independently trained and optimized for specific functions. This functional segmentation enables each network to adapt to its specific processing task while the overall system gains versatility through the combination of specialized components.
Solution Approach 2:
The patent enables dynamic adaptation by allowing neural network parameters to be updated during runtime based on incoming data characteristics. The system can retrain or fine-tune network parameters in response to changing channel conditions, data patterns, or error rates, providing continuous adaptation without requiring complete system redesign.
3Measurement precision
If extensive training is performed on all neural network parameters, then detection accuracy is improved, but training time increases
Solution Approach 1:
The patent divides the training process into separate stages for different neural network components. Each network (equalization, detection, decoding) is trained independently on its specific function using relevant training data, reducing the overall training time compared to training a single monolithic network on all aspects simultaneously. This modular training approach maintains accuracy while significantly reducing computational burden.
Solution Approach 2:
The patent performs preliminary training of neural network parameters during manufacturing or initialization using extensive training datasets. Once trained, these parameters are stored and used during runtime with minimal additional training required. This preliminary action transfers the heavy training burden to the manufacturing phase, allowing fast operation during actual data processing.
4Measurement precision
If multiple neural network circuits are implemented with separate training, then adaptability and detection accuracy are improved, but device complexity increases
Solution Approach 1:
The patent employs universal neural network building blocks that can be configured to perform different functions (equalization, detection, decoding) through parameter settings and connection configurations rather than requiring completely separate hardware for each function. This universality reduces manufacturing complexity by using standardized components while maintaining the benefits of multiple specialized circuits.
Data Source
AI summary
Example systems, read channels, and methods provide multiple neural network training nodes for processing read data signals prior to symbol detection and decoding. A plurality of neural network circuits receive read data signals and modify them based on different neural network configurations and sets of trained node coefficients. Each neural network circuit may pass modified read data signals directly to another neural network circuit or determine a parameter for modifying processing of the read data signals by another component. In some configurations, the last neural network circuit may pass the output read data signal to a soft output detector for determining the symbols in the read data signal.