Machine learning in optical

1. Indian Institute of Technology Patna
Presentation on
An Introduction to Machine Learning in
Optical Communication
Submitted to Submitted by
Dr.Preetam Kumar Vishal Waghmare

2. What is Machine Learning?
 Machine Learning
 Study of algorithms that
 improve their performance
 at some task
 with experience
 Teaching a computer to automatically learn concepts through
data observation.

3. Traditional Programming
Machine Learning
Computer
Data
Program
Output
Computer
Data
Output
Program

4.  Optimize a performance criterion using example data or past
experience.
 Role of Statistics: Inference from a sample.
 Role of Computer science: Efficient algorithms to
 Solve the optimization problem.
 Representing and evaluating the model for inference.

5. Magic?
No, more like gardening
Seeds = Algorithms
Nutrients = Data
Gardener = You
Plants = Programs

6. Optical networks
 Optical networks constitute the basic physical infrastructure of all
large-provider networks worldwide.
 Thanks to their high capacity, low cost and many other attractive
properties.
 They are now penetrating new important telecom markets as
datacom and there is no sign that a substitute technology might
appear in the foreseeable future.
 Different approaches to improve the performance of optical
networks have been investigated, such as routing, wavelength
assignment, traffic grooming and survivability.

7. ML algorithms
 Supervised-learning algorithms
 Unsupervised-learning algorithms
 Semi-Supervised learning
 Reinforcement learning

8. Supervised-learning
 We are given “labeled” data .
 Training data includes desired outputs.
 Main objective: given a set of “historical” input(s) predict an output
• Regression: output value is continuous .
• Classification: output value is discrete or “categorical”.

9.  An example: Traffic forecasts
 Given traffic during last week/month/year
 Predict traffic for the next period (regression).
 Predict if available resources will be sufficient (classification).
 Other examples
 Speech/image recognition
 Spam classifier
 House prices prediction/estimation

10.  Supervised Learning: the algorithm is trained on dataset that
consists of paths, wavelengths, modulation, and the corresponding
BER. Then it extrapolates the BER in correspondence to new inputs.

11. Courtesy of Marco Ruffini and Irene Macaluso

12. Unsupervised-learning
 Available data is not “labeled”.
 Training data does not include desired outputs.
 Main objective: derive structures (patterns) from available data
 Clustering finding “groups” of similar data.
 Anomaly detection.

13.  An example: cell-traffic classification.
 understand if some cells provide similar patterns
• Residential, business, stadium.
 This information can be used to make network resources planning.

14.  Other example
 Group the people according to their interests to improve advertisement.
 Unsupervised Learning: the algorithm identifies unusual patterns
in the data, consisting of wavelengths, paths, BER, and modulation.

15. Courtesy of Marco Ruffini and Irene Macaluso

16. Semi-Supervised learning
 Hybrid of previous two categories.
 Training data includes a few desired outputs.
 This Techniques also make use of unlabeled data for training –
typically a small amount of labeled data with a large amount of
unlabeled data.

17. Reinforcement learning
 Available data is not “labeled”.
 Rewards from sequence of actions.
 Main objective: learn a policy, i.e., a mapping between in
inputs/states and actions. Behavior is refined through rewards.
 Methodologically similar to
 Optimal control theory
 Dynamic programming
 Q-learning

18. Courtesy of Marco Ruffini and Irene Macaluso

19.  Reinforcement Learning: the algorithm learns by receiving
feedback on the effect of modifying some parameters, e.g. the
power and the modulation.

20. Overview of other applications
 Physical layer
1. Quality of Transmission (QoT) estimation
2. Optical amplifier control
3. Modulation format recognition
4. Nonlinearities mitigation

21.  Network layer
1. Traffic prediction and virtual topology design
2. Failure detection and localization
3. Flow classification

22. The general framework of a ML-assisted optical network

23. Quality of Transmission (QoT) estimation
 The concept of Quality of Transmission generally refers to a number
of physical layer parameters, such as received Optical Signal-to-
Noise Ratio (OSNR), BER, Q-factor, etc.
 Which have an impact on the “readability” of the optical signal at
the receiver.
 Such parameters give a quantitative measure to check if a
predetermined level of QoT would be guaranteed.
 Conversely, ML constitutes a promising means to automatically
predict whether unestablished lightpaths will meet the required
system QoT threshold.

24. Optical amplifier control
 When adding/dropping channels into/from a WDM system, EDFA
gain should be adjusted to re-balance output powers.
 An automatic control of preamplification signal power levels is
required, especially in case a cascade of multiple EDFAs to avoid
that excessive post-amplification power discrepancy.
 ML regression algorithms can be trained to accurately predict post
amplifier power excursion in response to the add/drop of specific
wavelengths to/from the system.

25.  ML allows to self-learn typical response patterns
Optical amplifier control

26. Modulation format recognition (MFR)
 Automatic digital modulation recognition in intelligent communication
systems is one of the most important issues in software defined radio
and cognitive radio.

27. Modulation format recognition (MFR)

28. Nonlinearities mitigation
 Optical signals are affected by fiber nonlinearities Kerr effect, self-
phase modulation (SPM), cross-phase modulation (XPM), Q-
factor, Chromatic Dispersion (CD), Polarization Mode Dispersion
(PMD).
 Traditional methods require complex mathematical models.
 ML models can be designed to directly capture the effects of such
nonlinearities, typically by creating input-output relations between
the monitored parameters and the desired outputs.

29. Nonlinearities mitigation

30. Traffic prediction and virtual topology design
 Accurate traffic prediction in the time-space domain allows
operators to effectively plan and operate their networks.
 Traffic prediction allows to reduce over-provisioning as much as
possible.
 Supervised learning algorithm can be trained to predict future traffic
requirements and consequent resource needs.

31. Traffic prediction and virtual topology design

32. Flow classification
 Traffic flows can be heterogeneous in terms of:
 protocols (http, ftp, smtp…)
 services (fixed vs mobile, VoD, data transfer, text messages…)
 requirements (latency, bandwidth, jitter…)
 network “customers” (human end-users, companies, sensors,
servers…)

33.  Distinguish between different flows is crucial for resources (i.e.,
capacity) allocation, scheduling, SLAs, QoS…
 supervised learning algorithms can be trained to extract hidden
traffic characteristics and perform fast packets classification and
flows differentiation.

34. References
 S. Shahkarami, F. Musumeci, F. Cugini, M. Tornatore, Machine-Learning-Based
Soft-Failure Detection and Identi cation in Optical Networks,"in Proceedings,
OFC 2018, San Diego (CA), Usa, Mar. 11-15, 2017
 A. Vela et al., “Soft Failure Localization during Commissioning Testing and
Lightpath Operation”, Journal of Optical Communication and Networking, vol.
10 n. 1, Jan. 2018
 A. Vela et al., “BER degradation Detection and Failure Identification in Elastic
Optical Networks”, in Journal of Lightwave Technology, vol. 35, no. 21, pp. 4595-
4604, Nov.1, 1 2017
 R.O. Duda, P.E. Hart, and D.G. Stork, Pattern Classification, New York: John
Wiley, 2001

35. Thank You!