Fiber communication receiver models based on the multi-head attention mechanism

Yubin Zang; Zhenming Yu; Kun Xu; Minghua Chen; Sigang Yang; Hongwei Chen

doi:10.3788/COL202321.030602

Journals >Chinese Optics Letters >Volume 21 >Issue 3 >Page 030602 > Article

Chinese Optics Letters
Vol. 21, Issue 3, 030602 (2023)

Fiber communication receiver models based on the multi-head attention mechanism

Yubin Zang¹, Zhenming Yu^2、**, Kun Xu², Minghua Chen¹, Sigang Yang¹, and Hongwei Chen^1、*

Author Affiliations

¹Beijing National Research Center for Information Science and Technology (BNRist) and Department of Electronic Engineering, Tsinghua University, Beijing 100084, China

²State Key Laboratory of Information Phonetics and Optical Communications, Beijing University of Post and Telecommunications, Beijing 100876, China

show less

DOI: 10.3788/COL202321.030602 Cite this Article Set citation alerts

Yubin Zang, Zhenming Yu, Kun Xu, Minghua Chen, Sigang Yang, Hongwei Chen. Fiber communication receiver models based on the multi-head attention mechanism[J]. Chinese Optics Letters, 2023, 21(3): 030602 Copy Citation Text

EndNote(RIS)

BibTex

Plain Text

show less

Structure of the communication scenario, the traditional receiver, and the AI-based receiver model containing multi-head attention mechanism.

Fig. 1. Structure of the communication scenario, the traditional receiver, and the AI-based receiver model containing multi-head attention mechanism.

Download full size | View in the Article

Fig. 2. Data formats and collection configurations.

Download full size | View in the Article

Fig. 3. Performance of convergence of AI-based receiver model through training.

Download full size | View in the Article

Fig. 4. BER-SNR diagram of the universal receiver model.

Download full size | View in the Article

Parameter	Value	Parameter	Value
Modulation format	OOK/PAM	Symbol rate	10/20/40 GBaud
Sampling rate	$8 \times symbol rate$	Transmission distance	0–100 km
Scale of training data set	31,744 symbols per distance	SNR	0–16 dB and +inf for OOK; 0–20 dB and +inf for PAM
Scale of validation data set	1600 symbols per distance	Scale of test data set	523,264 symbols per distance
Configurations for training set	1–100 km, interval of 1 km	Configurations for test set	0.5–99.5 km, interval of 1 km
Configurations for validation set	1–100 km, interval of 1 km	Central wavelength	1550 nm
Optimizer	ADAM	Batch size	4096 symbols
Loss function	NMSE	Valuation function	Bit error rate
Power of laser source	0 dBm	Modulator	Intensity modulator
Fiber in transmission link	SSMF (G.652)	Responsivity of PD	1 A/W
Dispersion	16.75 ps·nm⁻¹·km⁻¹	Effective area	$80 {μm}^{2}$