EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Low-Complexity AdderCNN Equalizer for 200 Gbaud RRM-Based IM/DD Transmission

Dan Li, Yevhenii Osadchuk, Armands Ostrovskis, Tianyu Jiang, Francesco Da Ros, Darko Zibar, Richard Schatz, Lu Zhang, Xianbin Yu, Vjaceslavs Bobrovs, Xiaodan Pang, Oskars Ozolins

Journal of Lightwave Technology · 15.05.2026 · 44 (10) · 3975-3982

DOI: 10.1109/jlt.2026.3670709 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Lightwave Technology, erschienen am 15.05.2026. Verfasst von Dan Li und 11 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: China Scholarship Council, LZP FLPP, Vetenskapsrådet.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Neural networks (NNs) have emerged as an effective equalization approach for high-speed intensity modulation and direct detection (IM/DD) optical links, where chromatic dispersion, limited bandwidth, and device-induced nonlinearities degrade signal quality. However, the heavy reliance on multiplications in conventional NNs leads to high computational complexity, limiting hardware deployment. Addition-based convolutional NN (AdderCNN) was originally developed for image classification, where it reduces complexity by replacing multiplications with addition-based operations. In this work, we propose a multiplier-free AdderCNN equalizer for up to 200 Gbaud on-off keying IM/DD links using a 26 GHz O-band ring resonator modulator. After 500 m single-mode fiber transmission, AdderCNN achieves bit error rate (BER) below the 7% overhead (OH) hard-decision forward error correction (HD-FEC) threshold of 3.8 ×10−3, outperforming conventional equalizers. Compared to classical CNN, AdderCNN eliminates all real multiplications (RM), reduces bit operations (BOP) by 87.5% and the number of adders and shifters (NABS) by 93.9% per equalized symbol under 32-bit full-precision with comparable performance. With 7-bit quantization, it further reduces both BOP and NABS by 71.7% per equalized symbol, while keeping BER below the 7% OH HD-FEC threshold. Therefore, AdderCNN has the potential to become a hardware-efficient NN equalization solution for next-generation optical interconnects.

Förderung

  • China Scholarship Council
  • LZP FLPP
  • Vetenskapsrådet
  • Sweden’s Innovation Agency

Zitieren

APA

Li, D., Osadchuk, Y., Ostrovskis, A., Jiang, T., Ros, F. D., Zibar, D., Schatz, R., Zhang, L., Yu, X., Bobrovs, V., Pang, X., & Ozolins, O. (2026). Low-Complexity AdderCNN Equalizer for 200 Gbaud RRM-Based IM/DD Transmission. Journal of Lightwave Technology, 44(10), 3975-3982. https://doi.org/10.1109/jlt.2026.3670709

Vancouver

Li D, Osadchuk Y, Ostrovskis A, Jiang T, Ros FD, Zibar D, et al. Low-Complexity AdderCNN Equalizer for 200 Gbaud RRM-Based IM/DD Transmission. Journal of Lightwave Technology. 2026;44(10):3975-3982. doi:10.1109/jlt.2026.3670709

BibTeX
@article{Li2026,
  author = {Li, Dan and Osadchuk, Yevhenii and Ostrovskis, Armands and Jiang, Tianyu and Ros, Francesco Da and Zibar, Darko and Schatz, Richard and Zhang, Lu and Yu, Xianbin and Bobrovs, Vjaceslavs and Pang, Xiaodan and Ozolins, Oskars},
  title = {Low-Complexity AdderCNN Equalizer for 200 Gbaud RRM-Based IM/DD Transmission},
  journal = {Journal of Lightwave Technology},
  year = {2026},
  volume = {44},
  number = {10},
  pages = {3975--3982},
  doi = {10.1109/jlt.2026.3670709}
}
Kostenlos starten →