SerDes Equalizer Skewed Offset LMS Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current link training methods in data communications systems, particularly in high-speed serial links, face challenges in efficiently adjusting equalizer settings to optimize signal quality and minimize inter-symbol interference (ISI) due to limitations in existing adaptive control schemes like LMS, which can lead to suboptimal signal-to-noise ratio (SNR) and bit error rate (BER) performance.
Innovation Solution
Implementing a skewed offset scheme in the LMS adaptation process to lock equalizer settings to a non-zero offset from the minimum mean square error, allowing for continued tracking of system variations and noise while maintaining optimal performance, thereby stabilizing equalizer coefficients differently for increments and decrements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard LMS adaptation is used to minimize mean square error, then equalizer settings converge to minimum error, but the system cannot track variations and noise effectively
Solution Approach 1:
The patent applies asymmetry by treating increments and decrements of equalizer coefficients differently. When the coefficient changes direction (indicated by sign reversal), the adaptation algorithm applies different step sizes or rules for increasing versus decreasing the coefficient, allowing the system to maintain a biased operation point that enables tracking while preserving convergence properties
Solution Approach 2:
The patent changes the parameter of the LMS adaptation by introducing an offset to the mean square error minimum. Instead of converging to zero error, the system targets a non-zero error level that provides a operating margin for tracking variations. This parameter change transforms the adaptation behavior from pure minimization to biased tracking
2Adaptability or versatility
If equalizer settings are adjusted aggressively to track variations, then adaptability improves, but stability and convergence are compromised
Solution Approach 1:
The patent introduces dynamics by making the adaptation behavior conditional on the direction of coefficient changes. The algorithm dynamically adjusts its response based on whether the coefficient is increasing or decreasing, allowing aggressive tracking in one direction while maintaining stability in the other, thus adapting the system's inertia to the current operating conditions
3Reliability
If LMS adaptation minimizes error to zero, then signal quality improves, but the system becomes sensitive to noise and channel variations
Solution Approach 1:
The patent applies preliminary anti-action by intentionally preventing the error from reaching zero. The offset scheme proactively maintains a non-zero error floor that acts as a buffer against noise and variations, anticipating that perfect convergence would make the system overly sensitive to disturbances
Data Source
AI summary
Examples described herein include setting an equalizer tap setting and gain setting in a serializer/deserializer (SerDes). In some examples, determining an equalizer setting and gain setting occurs by causing a mean-square error cost scheme tracking to lock to an offset from a minimum of a cost of the mean-square error cost scheme without pausing error cost tracking. In some examples, the mean-square error cost scheme comprises a least mean square (LMS) scheme. In some examples, determining an equalizer setting comprises: applying increases or decreases to an equalizer setting, and an increase to an equalizer setting can be a different amount than an amount of decrease to an equalizer setting.