EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Ensembling a Learned Volterra Polynomial with a Neural Network for Joint Nonlinear Distortions and Mismatch Errors Calibration of Time-Interleaved Pipelined ADCs

Yan Liu, Mingyu Hao, Hui Xu, Xiang Gao, Haiyong Zheng iD

Sensors · 29.06.2025 · 25 (13) · 4059

DOI: 10.3390/s25134059 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 29.06.2025. Verfasst von Yan Liu und 4 weiteren Autor:innen. Beteiligte Einrichtung: College of Electronic Engineering, Ocean University of China, Qingdao 266404, China. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 32 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Shandong Provincial Natural Science Foundation of China, TaiShan Scholar Youth Expert Program of Shandong Province.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The inherent non-ideal characteristics of circuit components and inter-channel mismatch errors induce nonlinear amplitude and phase distortions in time-interleaved pipelined analog-to-digital converters (TI-pipelined ADCs), significantly degrading system performance. Limited by prior modeling, conventional digital calibration methods only correct partial errors, while machine learning (ML) approaches achieve comprehensive calibration at a high computational cost. This work proposes an ensemble calibration framework that combines polynomial modeling and ML techniques. The ensemble calibration framework employs a two-stage correction: a learned Volterra front-end performs forward mapping to compensate static baseline nonlinear distortions, while a lightweight neural network back-end implements inverse mapping to correct dynamic nonlinear distortions and inter-channel mismatch errors adaptively. Experiments conducted on TI-pipelined ADCs show improvements in both the spurious-free dynamic range (SFDR) and signal-to-noise and distortion ratio (SNDR). It is noteworthy that in two ADCs fabricated using 40 nm CMOS technology, the 12-bit, 3000 MS/s silicon-validated four-channel TI-pipelined ADC exhibits SFDR and SNDR improvements from 35.47 dB and 35.35 dB to 79.70 dB and 55.63 dB, respectively, while the 16-bit, 1000 MS/s silicon-validated four-channel TI-pipelined ADC demonstrates an enhancement from 38.62 dB and 40.21 dB to 80.90 dB and 62.43 dB, respectively. Furthermore, a comparison with related studies reveals that our method achieves comprehensive calibration performance for wide-band inputs while substantially reducing computational complexity, requiring only 4.4 K parameters and 8.57 M floating-point operations per second (FLOPs).

Schlagwörter

Stichwörter der Autor:innen

TI-pipelined ADCVolterra polynomialcalibrationmismatch errorsneural networknonlinear distortions

Autor:innen und Institutionen

  • College of Electronic Engineering, Ocean University of China, Qingdao 266404, China.

ORCID-Profile: Haiyong Zheng

Förderung

  • Shandong Provincial Natural Science Foundation of China
  • TaiShan Scholar Youth Expert Program of Shandong Province

Zitieren

APA

Liu, Y., Hao, M., Xu, H., Gao, X., & Zheng, H. (2025). Ensembling a Learned Volterra Polynomial with a Neural Network for Joint Nonlinear Distortions and Mismatch Errors Calibration of Time-Interleaved Pipelined ADCs. Sensors, 25(13), 4059. https://doi.org/10.3390/s25134059

Vancouver

Liu Y, Hao M, Xu H, Gao X, Zheng H. Ensembling a Learned Volterra Polynomial with a Neural Network for Joint Nonlinear Distortions and Mismatch Errors Calibration of Time-Interleaved Pipelined ADCs. Sensors. 2025;25(13):4059. doi:10.3390/s25134059

BibTeX
@article{Liu2025,
  author = {Liu, Yan and Hao, Mingyu and Xu, Hui and Gao, Xiang and Zheng, Haiyong},
  title = {Ensembling a Learned Volterra Polynomial with a Neural Network for Joint Nonlinear Distortions and Mismatch Errors Calibration of Time-Interleaved Pipelined ADCs},
  journal = {Sensors},
  year = {2025},
  volume = {25},
  number = {13},
  pages = {4059},
  doi = {10.3390/s25134059}
}
Kostenlos starten →