Journal of Optical Communications and Networking
Vol. 13,
Issue 4,
pp. B60-B71
(2021)
•https://doi.org/10.1364/JOCN.417434

Machine learning techniques for quality of transmission estimation in optical networks

Yvan Pointurier

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Yvan Pointurier, "Machine learning techniques for quality of transmission estimation in optical networks," J. Opt. Commun. Netw. 13, B60-B71 (2021)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Abstract

The estimation of the quality of transmission (QoT) in optical systems with machine learning (ML) has recently been the focus of a large body of research. We discuss the sources of inaccuracy in QoT estimation in general; we propose a taxonomy for ML-aided QoT estimation; we briefly review ML-aided optical performance monitoring, a tightly related topic; and we review and compare all recently published ML-aided QoT articles.

Full Article | PDF Article

More Like This

Performance comparisons between machine learning and analytical models for quality of transmission estimation in wavelength-division-multiplexed systems [Invited]

Jianing Lu, Gai Zhou, Qirui Fan, Dengke Zeng, Changjian Guo, Linyue Lu, Jianqiang Li, Chongjin Xie, Chao Lu, Faisal Nadeem Khan, and Alan Pak Tao Lau
J. Opt. Commun. Netw. 13(4) B35-B44 (2021)

Associating machine-learning and analytical models for quality of transmission estimation: combining the best of both worlds

Emmanuel Seve, Jelena Pesic, and Yvan Pointurier
J. Opt. Commun. Netw. 13(6) C21-C30 (2021)

Accurate Quality of Transmission Estimation With Machine Learning

Ippokratis Sartzetakis, Konstantinos (Kostas) Christodoulopoulos, and Emmanouel (Manos) Varvarigos
J. Opt. Commun. Netw. 11(3) 140-150 (2019)

Previous Article Next Article

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (2)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (5)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

ADC	Analog to Digital Converter
ANN	Artificial Neural Network
ASE	Amplified Spontaneous Emission
AUR	Area Under the ROC
BER	Bit Error Rate
DNN	Deep Neural Network
DSP	Digital Signal Processing
EDFA	Erbium-Doped Fiber Amplifier
FEC	Forward Error Correction
GPR	Gaussian Process Regression
KNN	$K$ Nearest Neighbors
ML	Machine Learning
NN	Neural Network
OPM	Optical Performance Monitoring
OSA	Optical Spectrum Analyzer
OSNR	Optical Signal-to-Noise Ratio
PDL	Polarization Dependent Loss
PDM	Polarization Division Multiplexing
QAM	Quadrature Amplitude Modulation
QoT	Quality of Transmission
QPSK	Quadrature Phase-Shift Keying
RMSE	Root Mean Square Error
ROADM	Reconfigurable Optical Add-Drop Multiplexer
ROC	Receiver Operating Characteristic
SNR	Signal-to-Noise Ratio
SSF	Split Step Fourier
SVM	Support Vector Machine
WDM	Wavelength Division Multiplexing
WSS	Wavelength Selective Switch

Paper	Use Case	Data Source	Learning Type	Algorithm	Type of Study	Inputs (Features)	Output	Key Results
[46]	Single EDFA	Green, brown	Data scientist	DNN	Exp.	Per channel input power	Optical power spectrum	$R M S E < 0.2 d B$
[47]	Single EDFA	Green, brown	Data scientist	DNN	Exp.	Per channel input power	Gain	0.18 dB max. gain estimation error
[48]	Single EDFA	Green, brown	Data scientist	ANN	Sim.	Per channel input status (on/off)	Optical power spectrum	$E r r o r < 0.2 d B$ in 90% of the cases
[49]	Single EDFA	Green, brown	Hybrid	DNN	Exp.	Per channel input power; per channel output power from analytic model	Optical power spectrum	$R M S E < 0.2 d B$ ; faster convergence with inputs from analytic model
[50]	Single Raman/EDFA link	Green, brown	Data scientist	DNN	Exp.	Amplifiers’ pump powers (which adjust gain and tilt)	Optical power spectrum	Typical error in output power $< 0.2 d B$ (max., 0.3 dB)
[51]	Single Raman link	Green, brown	Data scientist	ANN	Sim.	Per subband input power, pump powers	Gain and ASE noise prediction	Max. error $< 0.5 d B$ on gain and noise power

Paper	Use Case	Data Source	Learning Type	Algorithm	Type of Study	Inputs (Features)	Output	Key Results
[55]	Line	Green, brown	Data scientist	GPR	Sim./Exp.	TX power, number of spans, baudrate, interchannel spacing	$B E R / Q$	$Q e r r o r < 0.3 d B$
[56]	Line	Brown	Data scientist	DNN	Exp.	Per-channel received signal and noise	OSNR	0.5 dB max. OSNR error
[57]	Line	Green, brown	Data scientist	ANN	Exp.	End of line spectrum	OSNR	Error on OSNR $< 0.2 d B$
[58]	Line		Data scientist	DNN, transfer learning	Exp.	Amplitude histogram of received samples	OSNR	Retraining test size divided by 5, training time divided by 10
[59]	Line		Data scientist	DNN, transfer learning	Exp.	$Q$ from other lines and new line	$Q$	Retraining test size divided by 50, training time divided by 4

Paper	Use Case	Data Source	Learning Type	Algorithm	Type of Study	Inputs (Features)	Output	Key Results
[60]	Network	Brown	Data scientist	SVM	Sim.	Light path length, link lengths, wavelength, statistics on co-propagating light paths	BER (2 classes)	$> 99.9 %$ correct decisions
[61]	Network	Brown	Data scientist	ANN	Sim.	Path length, number of EDFAs, max. link length, destination node degree, wavelength	BER (2 classes)	Accuracy $> 90 %$
[32]	Network	Brown	Data scientist	KNN, random forest	Sim.	Total length, number of links, maximum link length, demand capacity, modulation format, guardband, modulation format, and traffic volume of nearest left and right neighbor	BER (2 classes)	Typical accuracy $> 95 %$
[62]	Network	Brown	Data scientist	GPR, active learning	Sim.	Total length, number of links, maximum link length, demand capacity, modulation format	BER (2 classes)	Accuracy $> 99 %$ by complementing a small data set of existing services with only a few well-chosen light paths used as probes
[63]	Network	Brown	Data scientist	SVM, ANN	Sim.	Total link length, span length, channel launch power, modulation format, data rate	SNR (2 classes)	$> 99 %$ correct decisions with ANN and SVM, ANN 10× faster (sub-ms)
[64]	Network	Brown	Data scientist	KNN, logistic regression, SVM (classification), ANN (classification, regression)	Sim.	Number of hops and spans, light path length, average and maximum link length, average span attenuation, average dispersion, modulation format	SNR (continuous or 2 classes)	Typical classification accuracy $> 95 %$ , average SNR error $< 0.4 d B$
[66]	Network	Brown	Data scientist	SVM	Sim.	Number of ROADMs, of links, of fiber spans; length of fiber span, launch channel power, EDFA pre- and EDFA post-amplifier gain, wavelength	OSNR (4 classes)	$> 95 %$ correct decisions
[65]	Network		Data scientist	SVM, transfer learning	Sim.	Total length, number of links, maximum link length, demand capacity, modulation format	BER (2 classes)	Retraining test size divided by 20

Paper	Use Case	Data Source	Learning Type	Algorithm	Type of Study	Inputs (Features)	Output	Key Results
[71]	Network	Brown	Data scientist	ANN	Exp.	Not explicit	OSNR	Microsecond estimation, RMSE $< 0.2 d B$ in 90% of cases
[19]	Network	Brown	Physicist	Filtering	Exp.	Pre-FEC BER	$Q$	0.6 dB max. estimation error on $Q$ , millisecond estimation
[72]	Network	Brown	Physicist	Gradient descent	Sim.	Pre-FEC BER, fiber input power, NF	SNR	Typical error $< 0.1 d B$
[73]	Network	Brown	Data scientist	DNN	Exp.	Channel, noise power on each link	OSNR	$< 5 d B$ OSNR error over range 20–40 dB
[74]	Network	Brown	Physicist; data scientist	Least squares	Sim.	Per span: fiber attenuation, dispersion, nonlinear coefficient (physicist); number of EDFAs and hops, total path length, baudrate, “load metric” (data scientist)	SNR	SNR RMSE down to $< 0.1 d B$
[75]	Network	Brown	Hybrid	ANN	Sim.	Gaussian noise model output, span number, maximum and average lengths, launch power, link length, residual chromatic dispersion, average fiber nonlinear coefficient and attenuation, number of channels	Nonlinear SNR	Error $< 2 d B$ in 95% of the cases
[76]	Network	Brown	Hybrid	ANN	Sim.	Fiber attenuation, dispersion coefficient, effective area, and non-linear refractive index; span length, number of active channels, launch power, channel bandwidth and frequency, output of analytic modeling tool	Nonlinear SNR	$> 99.9 %$ estimates within 0.5 dB, $< 10 m s$ computation
[77]	Network	Brown	Data scientist	ANN	Sim.	List of span lengths	SNR	RMSE $< 0.2 d B$
[78]	Network	Brown	Data scientist	DNN	Exp.	Launch power, laser bias, EDFAs input/output powers	$Q$	RMSE $< 0.02 d B$ with 600 training points
[79]	Network	Brown	Physicist	SVM regression	Sim.	Power per channel at each node, fine (sub-GHz) optical spectrum every few spans, pre-FEC BER	SNR	Typical error $< 0.2 d B$
[80]	Network	Brown	Physicist	Gradient descent	Sim.	Pre-FEC BER, spectrum at gain equalizer	SNR	Typical error $< 0.1 d B$
[81]	Network	Both	Hybrid	ANN	Sim.	Per channel power and frequency; number of spans; from physical model: ASE and nonlinear noises	SNR	Max. error $< 0.5 d B$
[82]	Network		Data scientist	DNN, evolutionary transfer learning	Exp.	OSA power profile at each WSS	$Q$	Retraining test size divided by 10
[83]	Network		Data scientist	ANN	Sim.	List of span lengths	SNR	RMSE in dB divided by 2

Machine learning techniques for quality of transmission estimation in optical networks

Author Affiliations

ORCID

Abstract

Cited By

Figures (2)

Tables (5)

Journal of Optical Communications and Networking