Multi-Channel Nonlinearity Mitigation Using Machine Learning Algorithms

doi:10.1109/TMC.2023.3259880

2024-04-01

PRE

Haotian Zhao *

Julian Camilo Gomez Diaz

Sebastián Hoyos

DOI：10.1109/TMC.2023.3259880

原文链接

原文求助

摘要

En 中文

This paper investigates multi-channel machine learning (ML) techniques in the presence of receiver nonlinearities and noise, and compares the results with the single-channel receiver architecture. It is known that the multi-channel architecture relaxes the sampling speed requirement of analog to digital conversion and provides significant robustness to clock jitter and front-end noise due to the bandwidth-splitting property inherent in these receivers. However, when a high-voltage swing signal is used in a wireline communication link, the received signal suffers from third-order harmonic distortions and inter-modulation products caused by the nonlinearity profile of the analog front-end (AFE). To this end, this paper proposes the channel decision passing (CDP) algorithm in combination with nonlinear feedback cancellation as a low-complexity candidate for nonlinearity mitigation and compares the performance of this solution with other well-known ML algorithms. Simulation results show significant improvement in a multi-channel receiver architecture equipped with nonlinear feedback cancellation and CDP in comparison with its single-channel counterpart under practical nonlinearity profiles and noise conditions.

Keyword:

Receivers

Clustering algorithms

Classification algorithms

Machine learning algorithms

Bandwidth

Supervised learning

Unsupervised learning

Multi-channel receiver

nonlinearities

machine learning

supervised learning