Enhancing Tactile Internet Reliability: AI-Driven Resilience in NG-EPON Networks
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
The manuscript proposes a new architecture based on SEBA for AI-enhanced anomaly detection that improves the overall network operation. The introduction gives a lot of background information and gives a good overview of the tackled problem. The proposed method is well-described with informative figures. The experimental evaluation illustrates the usefulness of the proposed system. The paper is well-written in terms of language.
I have a few suggestions to improve the manuscript:
– all acronyms should be explained on first occurrence, even the “common knowledge” ones, to enable broad audience to understand the work. Some of them (e.g., VOLTHA, NEM, SDN, OLT) are explained but not on first occurrence;
– in the ML-based part of the system in the experimental part (Section 4.2), the dataset should be described in more detail, as data is the basis of any ML model. The number of acquired samples is given but it is not clear if the dataset is balanced and how the failure scenarios are distributed.
Author Response
Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for Authors
Dear authors
I would like to address some of the suggestions
1. In the abstract section, there is some lack of problem-solving and empathy with your proposed methods.
2. The introduction section from lines 35-37 should be considered, it is too short for one paragraph. The paragraphs in the introduction section should be within 2 pages. Please separate them into related work sections and make a table of affiliations.
3. Addressing your paper contribution and summarizing each section.
4. In lines 205-212, there are some mistakes with the writing flow.
5. Yourwork lacks convincing evidence, please add more flow charts and transition procedure.
6. In Section 5, please indicate your comparison with the baseline approach, and give your pros and cons.
7. Please emphasize this in your future work in the conclusion section.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors
According to the initial purposes of integrating AI, SDN, SEBA, VOLTHA, NEM into NG-EPON networks it seemed like a very interesting article, but basically the result is a simulation of NG-EPON networks using the OptiSystem simulator, which is far from the interesting intentions intended in the abstract.
Regarding the different parts of this paper:
1. Introduction: good introduction, well structured and documented, but I think that this paper has not developed what is indicated in lines 129 to 131 (detects and localizes fiber faults).
2. Network Fault Detection and Localization: Throughout this paper, the development of what was planned and indicated in lines 183 to 187 (where: localization of anomalies?, where: combining both OTDR and the eye diagram?) has not been carried out.
3. SEBA Architecture: I think there is an error on line 211: ASX instead of ASF. The description of Network Edge Mediator is too long, what is the meaning of FCAPS?.
4. Proposed Architecture: The infrastructure to detect and locate fiber faults is not sufficiently descibed (where are the network performance monitors located: BER?, SNR? and OTDR). The white box hardware and the RFoG system features must be described.
On line 76, the types of faults and degradations (only fiber cuts?) must be described.
In this paper what is used to measure QoT (BER Analyzer or EYE Diagram monitoring)?
What measurement resolution and accuracy of OTDR is necessary for localization of the different faults and degradations?
4.2. Experiment and Evaluation: Instead of Experimental evaluation it is rather a software Simulation.
This whole section needs to be rewritten (from line 338 to line 370), and some figures need to be included so that it can be understood better. Table 2 is not necessary, the information is already in the text.
4.3. Resilience Dynamic Bandwidth Allocation: What information does ML (OTDR, BER, Eye Diagram,...) work with? (line 385)
5. Performance Evaluation: It is necessary to describe the optical network, otherwise only the Ethernet layer is mentioned (where is the optical communication network?)
5.1 Mean Packet Delay: Indicate the meaning of EF (line 425) and describe the faults (one and three) in the network.
In simulation results (5.1 Mean Packet Delay, 5.2 System Throughput, 5.3 Packet Drop Rate , 5.4 Bandwidth Waste ) where is the description (components, features, etc.) of the fiber optic network that supports NG-EPON?
6. Conclusions: The information indicated in lines 501-502 has not been fully developed: the network topology has not been specified, nor how faulty branch anomalies are detected and located. Where is the experimental data?, all data comes from simulations. Regarding AI, the structure of the neural network needs to be better described and it should be indicated which data from the optical communication network it works with.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Comments and Suggestions for Authors
Dear Authors
I am pleased to say that your manuscript is well done.
Author Response
Concern #1: I am pleased to say that your manuscript is well done;
Author response: Thank you very much.
Reviewer 3 Report
Comments and Suggestions for Authors
First of all, we must congratulate the authors for their efforts in improving this second version of the paper.
But, it is an exaggeration to introduce the word Revolutionizing in the title of this paper.
Some recommendations:
It should be better explained how the eye diagram analyzer data helps in the localization (improving the accuracy and efficiency) of faults (which cannot be detected by the OTDR), since the eye data are used for anomaly detection.
Explain why the Neural Network only has 2 Neurons in the input layer? What data from OTDR and eye diagram are needed in its input?
To train the MLP (neural network) for binary classification tasks, explain how the data set (OTDR and eye diagram data) is obtained.
Author Response
Please see the attachemnt
Author Response File: 
Author Response.pdf
