Adaptive FFE Preset Selection for High-Speed Serial Links
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data communication technologies using high-speed serial expansion bus standards face challenges in identifying optimal filter preset coefficients for feed-forward equalization (FFE) due to non-monotonicity and non-orthogonality, leading to suboptimal performance in data links.
Innovation Solution
Adaptive feed-forward equalization (FFE) is implemented using a finite impulse response (FIR) filter with a lookup table and stochastic gradient descent (SGD) method to dynamically select FIR coefficients based on residual errors, optimizing signal quality by reducing intersymbol interference (ISI).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If preset filter coefficients are applied for feed-forward equalization, then signal quality is improved by compensating insertion loss, but the non-monotonicity and non-orthogonality of the coefficient sequence makes it difficult to identify optimal coefficients
Solution Approach 1:
The patent transforms the preset configuration table into a lookup table where coefficients are organized with monotonicity and orthogonality properties. This reparameterization allows the stochastic gradient descent algorithm to efficiently search for optimal coefficients by systematically varying parameters in an ordered sequence, resolving the difficulty of identifying optimal coefficients while maintaining signal quality improvement.
2Adaptability or versatility
If a plurality of preset filter configurations are used to handle different channel conditions, then adaptability is improved, but the ordered sequence lacks monotonicity and orthogonality making optimization difficult
Solution Approach 1:
The patent segments the coefficient optimization problem into manageable parts by creating a lookup table structure where each row represents a preset configuration. The table is organized with monotonic variation in certain coefficient groups and orthogonal relationships between different coefficient sets, allowing systematic search through configurations while maintaining adaptability to different channel conditions.
Solution Approach 2:
The patent introduces dynamic adaptation through the stochastic gradient descent algorithm that can dynamically select and adjust filter coefficients based on real-time channel conditions. The lookup table provides a structured framework that enables dynamic optimization while maintaining the benefits of multiple preset configurations for different scenarios.
3Ease of manufacture
If traditional preset configuration tables are used, then implementation is simplified, but performance optimization is hindered by non-monotonic coefficient sequences
Solution Approach 1:
The patent introduces a lookup table as an intermediary structure between the preset configuration table and the optimization algorithm. This intermediary organizes coefficients with monotonicity and orthogonality properties, enabling the stochastic gradient descent algorithm to efficiently optimize performance while maintaining the implementation simplicity of using preset configurations. The lookup table acts as a bridge that preserves ease of implementation while achieving precision optimization.
Data Source
AI summary
This application discloses adaptively setting feed-forward equalization (FFE) for a data communication channel. An equalization signal is generated using a finite impulse response (FIR) filter that has a plurality of FIR coefficients configured to be defined by one of a plurality of preset configurations. A lookup table has a plurality of rows, and each row is associated with a different preset configuration of the FIR coefficients and identifies a subset of respective preset configurations corresponding to a subset of FIR coefficients. In some implementations, a temporal sequence of preset configurations of the FIR coefficients is selected from the lookup table, until a predefined equalization criterion is satisfied. In some implementations, residual errors are determined and correspond to signal samples of the equalization signal, and a sequence of preset configurations of the FIR coefficients is selected from the lookup table based on the residual errors.