Machine learning techniques for selecting paths in multi-vendor reconfigurable optical add/drop multiplexer networks

Patent metadata

Field	Value
Publication number	US-11316607-B2
Application number	US-202117347383-A
Country	US
Kind code	B2
Filing date	Jun 14, 2021
Priority date	Sep 19, 2018
Publication date	Apr 26, 2022
Grant date	Apr 26, 2022

Abstract

Devices, computer-readable media and methods are disclosed for selecting paths in reconfigurable optical add/drop multiplexer (ROADM) networks using machine learning. In one example, a method includes defining a feature set for a proposed path through a wavelength division multiplexing network, wherein the proposed path traverses at least one link in the network, and wherein the at least one link connects a pair of reconfigurable optical add/drop multiplexers, predicting an optical performance of the proposed path, wherein the predicting employs a machine learning model that takes the feature set as an input and outputs a metric that quantifies predicted optical performance, and determining whether to deploy a new wavelength on the proposed path based on the predicted optical performance of the proposed path.

First claim

What is claimed is:

1. A method comprising:
defining, by a processing system including at least one processor, a feature set for a proposed path through a wavelength division multiplexing network, wherein the proposed path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link connects a pair of reconfigurable optical add/drop multiplexers;
predicting, by the processing system, an optical performance of the proposed path, wherein the predicting employs a machine learning model that takes the feature set as an input and outputs a metric that quantifies a predicted optical performance of the proposed path; and
determining, by the processing system, whether to deploy a new wavelength on the proposed path based on the predicted optical performance of the proposed path.

2. The method of claim 1 , wherein the machine learning model is a penalized linear regression model.

3. The method of claim 1 , wherein the machine learning model is an ensemble of regression trees.

4. The method of claim 3 , wherein the machine learning model is a random forest model.

5. The method of claim 1 , wherein the metric is a bit error rate of the proposed path.

6. The method of claim 5 , wherein the bit error rate is a bit error rate before forward error correction.

7. The method of claim 6 , wherein the bit error rate is a log 10 (bit error rate).

8. The method of claim 1 , wherein the determining comprises:
determining the metric satisfies an optical performance standard; and
deploying the new wavelength on the proposed path.

9. The method of claim 8 , wherein the optical performance standard is a threshold value for the metric.

10. The method of claim 8 , further comprising:
collecting data regarding an optical performance of the proposed path, after the deploying.

11. The method of claim 10 , further comprising:
training the machine learning model using the data.

12. The method of claim 1 , wherein the determining comprises:
generating a new proposed path, when the metric fails to satisfy an optical performance standard;
defining a new feature set for the new proposed path, wherein the new proposed path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link of the new proposed path connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the new proposed path using the machine learning model, wherein the machine learning model takes the new feature set as an input and outputs a new metric that quantifies a predicted optical performance of the new proposed path; and
determining whether to deploy the new wavelength on the new proposed path based on the predicted optical performance of the new proposed path.

13. The method of claim 1 , further comprising:
defining a feature set for an existing path, wherein the existing path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link of the existing path connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the existing path, wherein the predicting employs the machine learning model, wherein the machine learning model takes the feature set for the existing path as an input and outputs a metric that quantifies a predicted optical performance of the existing path; and
determining whether to move an existing wavelength from the existing path to an alternative path based on the predicted optical performance of the existing path.

14. A non-transitory computer-readable medium storing instructions which, when executed by a processing system including at least one processor, cause the processing system to perform operations, the operations comprising:
defining a feature set for a proposed path through a wavelength division multiplexing network, wherein the proposed path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the proposed path, wherein the predicting employs a machine learning model that takes the feature set as an input and outputs a metric that quantifies a predicted optical performance of the proposed path; and
determining whether to deploy a new wavelength on the proposed path based on the predicted optical performance of the proposed path.

15. A system comprising:
a processor; and
a non-transitory computer-readable medium storing instructions which, when executed by the processor, cause the processor to perform operations, the operations comprising:
defining a feature set for a proposed path through a wavelength division multiplexing network, wherein the proposed path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the proposed path, wherein the predicting employs a machine learning model that takes the feature set as an input and outputs a metric that quantifies a predicted optical performance of the proposed path; and
determining whether to deploy a new wavelength on the proposed path based on the predicted optical performance of the proposed path.

16. The system of claim 15 , wherein the metric is a bit error rate of the proposed path.

17. The system of claim 15 , wherein the determining comprises:
determining the metric satisfies an optical performance standard; and
deploying the new wavelength on the proposed path.

18. The system of claim 17 , the operations further comprising:
collecting data regarding an optical performance of the proposed path, after the deploying and training the machine learning model using the data.

19. The system of claim 15 , wherein the determining comprises:
generating a new proposed path, when the metric fails to satisfy an optical performance standard;
defining a new feature set for the new proposed path, wherein the new proposed path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link of the new proposed path connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the new proposed path using the machine learning model, wherein the machine learning model takes the new feature set as an input and outputs a new metric that quantifies a predicted optical performance of the new proposed path; and
determining whether to deploy the new wavelength on the new proposed path based on the predicted optical performance of the new proposed path.

20. The system of claim 15 , the operations further comprising:
defining a feature set for an existing path, wherein the existing path traverses at least one link in the wavelength division multiplexing network, and wherein the at least one link of the existing path connects a pair of reconfigurable optical add/drop multiplexers;
predicting an optical performance of the existing path, wherein the predicting employs the machine learning model, wherein the machine learning model takes the feature set for the existing path as an input and outputs a metric that quantifies a predicted optical performance of the existing path; and
determining whether to move an existing wavelength from the existing path to an alternative path based on the predicted optical performance of the existing path.

Assignees

• At & T Ip I Lp

Inventors

• Birk Martin
• Lynch David Frederick
• Thakur Gaurav
• Tse Simon

Classifications

• H04L45/08 Primary

  Learning-based routing, e.g. using neural networks or artificial intelligence · CPC title

• H04J14/02122

  Colourless, directionless or contentionless [CDC] arrangements · CPC title

• H04J14/021 Primary

  Reconfigurable arrangements, e.g. reconfigurable optical add/drop multiplexers [ROADM] or tunable optical add/drop multiplexers [TOADM] · CPC title

• H04J14/0257

  Wavelength assignment algorithms · CPC title

• H04J14/0267

  Optical signaling or routing · CPC title