Dynamic Error Feedback Quantization for DSL Precoder Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

DSL systems face performance limitations due to crosstalk interference, particularly during precoder initialization, where high data rates for error feedback signals are often not met, leading to increased initialization time and reduced accuracy.

Innovation Solution

The system adjusts the quantity of bits and quantization accuracy of error feedback signals based on the error range, allowing for lower data rates without sacrificing convergence speed or performance, by dynamically representing error feedback using fewer bits and higher quantization accuracy as the precoder output converges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of energy

If error feedback signals are transmitted using a subset of tones to reduce data rate requirements, then feedback overhead is reduced, but precoder initialization time increases and convergence speed decreases

Engineering Contradiction:
Improvefeedback overheadVSAvoidprecoder initialization time
Core Design Contradiction:
Loss of energyVSLoss of time

Solution Approach 1:

The patent changes the parameter of quantization bits dynamically during precoder training. Initially, full quantization precision is used to ensure accurate feedback for fast convergence. As training progresses and error ranges decrease, the system reduces the number of quantization bits, thereby reducing feedback overhead and data rate requirements while maintaining sufficient accuracy for the converged state.

Inventive Principle:
Principle #35Parameter changes

2Loss of energy

If quantization bits for error feedback are reduced to lower data rate requirements, then feedback bandwidth is reduced, but precision of error representation decreases

Engineering Contradiction:
Improvefeedback bandwidthVSAvoiderror feedback precision
Core Design Contradiction:
Loss of energyVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts the quantization precision based on the training phase and error magnitude. During early training phases with large error ranges, full precision is maintained. As the precoder converges and error ranges shrink, the system transitions to lower precision quantization, optimizing the trade-off between bandwidth efficiency and measurement precision throughout the training process.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent explicitly changes the quantization parameter (number of bits) based on the error range observed during training. When error ranges are large, more bits are used to maintain precision. When error ranges become small after convergence, fewer bits suffice, reducing feedback bandwidth requirements while maintaining adequate precision for the current error level.

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If fixed quantization accuracy is used throughout precoder training, then implementation is simpler, but feedback data rate remains high even when not needed

Engineering Contradiction:
Improveimplementation simplicityVSAvoidfeedback data rate
Core Design Contradiction:
Ease of manufactureVSLoss of energy

Solution Approach 1:

Rather than using a fixed quantization accuracy, the system implements dynamic adjustment of quantization bits based on the training phase and error magnitude. This adds some complexity to the implementation but dramatically reduces feedback data rate requirements, especially during later training phases when the precoder has already converged and high precision is no longer necessary.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3402116B1Method and apparatus for reducing feedback overhead
Publication Date: 2019.12.25 HUAWEI TECH CO LTD
  • EP3402116B1 patent drawingFigure 1
  • EP3402116B1 patent drawingFigure 2
  • EP3402116B1 patent drawingFigure 3

AI summary

An apparatus comprising a receiver coupled to a digital subscriber line (DSL) between an exchange site and a customer premise equipment (CPE) and configured to send a feedback error message to train a precoder coupled to the exchange site, wherein the feedback error message comprises a plurality of error components and an indication of a quantity of bits per error component, a quantization accuracy per error component, or both. Included is a method comprising determining a range of error for a plurality of error components of a pilot signal; determining a quantity of bits for representing error such that a full error range is preserved, a quantization accuracy such that the full error range is represented by a fixed number of feedback bits, or both based on the range of error for the error components; and transmitting an error feedback signal that comprises the error components and indicates the quantity of bits, the quantization accuracy, or both.