Max-Pooling PD: A Machine Learning-Based Timing Recovery Method

doi:10.1109/TCSI.2024.3524336

Max-Pooling PD: A Machine Learning-Based Timing Recovery Method

2025-01-01

0

PRE

AI

DOI：10.1109/TCSI.2024.3524336

Original

Original request for help

Save

Abstract

En 中文

This paper presents a max-pooling phase detector (PD) designed to significantly reduce power consumption and area compared to conventional baud-rate detectors such as the Mueller-Mueller (MMPD) and Sign-Sign MMPD (SS-MMPD). The PD was originally developed using a convolutional neural network (CNN) model and trained within a machine-learning (ML) framework optimized for the provided training dataset. The design method was completed using a minimum mean square error (MMSE) approach to address discrepancies between training conditions and real-world operation, enabling traditional adaptation methods to enhance the performance and robustness of the PD. Synthesized in the GF 22-nm FD-SOI process, the max-pooling PD consumes 40% less power and occupies 60% less area than a standard SS-MMPD.

Keywords:

Noise

Timing

Clocks

Jitter

Detectors

Training

Receivers

Standards

Decision feedback equalizers

Power demand

Phase detector

CDR

SerDes

wire-line

machine learning

CNN

Max-Pooling PD: A Machine Learning-Based Timing Recovery Method

Abstract

Journal

Organization