Experimental Machine Learning Approach for Optical Turbulence and FSO Outage Performance Modeling
, Keith Cohn 4, Andreas Tsigopoulos 2, Kostas Peppas 1, Konstantinos Aidinis 5 and Hector Nistazakis 3
Round 1
Reviewer 1 Report
The Manuscript describe about the Machine learning approach. Below are my comments:
Mention the error percentage as obtained in Abstract section.
Equation 2, 3, 4 are all derived from [22] reference!!! If yes, then I think only mentioning reference will be good.
Is your CDF equation is your proposed model??
Only one year of measurement data is sufficient?? If possible provide the measurement data performed for more years.
Arrange the Figure 1 position.
3.776 E-14 isn’t the presented data in Table 1 is very very small. Need some explanation. What are the count values signifying?
Figure 3 is not clear. Increase the resolution of this figure.
What is the relationship between R squared and RMSE??
Your R square values are less than 0.8!!! so, is your data trust worthy!!! Need some explanation.
Figure 10 need to re plotted as the equation is not clear. Increase the resolution.
Figure 12 need to be explained more.
Figure 14 and 15 labels in not clear.
How about Free-space propagation model!! Can this be applied in your case??
Error calculation need to be performed more. Follow the paper of 1-minute rain attenuation papers and 1-minute rain rate papers. ITU-R P. 838-3 need to be followed for error calculation.
Can we use any ITU recommended rule and regulations??
Author need to have more study on the research papers.
What is your proposed model is not clear. Where are your proposed formulas?? Is the paper just applying already proposed methods!!
Most importantly, provide the comparison table against the existing models.
Where is your conclusion section!!! Is this the silly mistake as not to have conclusion statement in manuscript?
Manuscript without conclusion and comparison table. I would prefer to reject this manuscript. However, if author revise this paper as suggested then would be happy to go through the revised version.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
1. In the study, the results of machine learning algorithms were compared by modeling the refractive index parameter using the experimental data of Free-space optical communication (FSOCs). This study compares six machine learning (ML) regression algorithms comprehensively. Atmospheric effects in FSO systems are highly variable and estimation/classification of data with machine learning will contribute to the literature. 2. The article is about applying machine learning to a field. It fills a gap in this field. 3. In the field of FSO, there are recent studies on applications such as machine learning and artificial intelligence. The authors have a work. In this study, a different parameter is modeled. 4. In the study, a channel model was developed for the atmospheric turbulence model. Then, performance metrics were determined and experimental data were analyzed. The methodology is appropriate. 5. Data from the FSO system for about a year has been used. Thus, all seasonal effects were investigated. Then, seasonal regression analyzes were made and presented in the figures. Features Consistent results have been obtained for the RF algorithm. 6. References are appropriate. 7. Figures are difficult to read. Their resolution should be increased.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Manuscript:
Experimental Machine Learning Approach for Optical Turbulence and FSO Outage Performance Modeling
submitted by:
Antonios Lionis, Antonios Sklavounos, Argyris Stassinakis, Keith Cohn, Andreas Tsigopoulos, Kostas Peppas, Konstantinos Aidinis and Hector Nistazakis
A laser beam propagating in the free space suffers numerous degradation effects. In the context of free space optical communications, this results in reduced link’s availability. The Authors' study provides a comprehensive comparison between six machine learning regression algorithms for modeling the refractive index structure parameter. The structure of the manuscript is considered and clear. In the introduction, the background and comprehensive review of the problem's literature were presented. The Authors present atmospheric otical turbulence, channel modeling and performance metrics. To execute the regression modeling analysis, the Authors utilized six well-known machne learning algorithms, Artificial Neural Networks, Random Forest, Gradient Boosting Regressor, k-Nearest Neighbor, Decision Trees and Deep Neural Networks. Results of the research have been presented in graphic form. Conclusions, on the basis of the research, are comprehensive and clear.
Following suggestions should be taken into consideration:
Line 32: gamma is written twice
Line 129 - Equation (1): a dot should be used for algebraic multiplication. x is used for vector one
Line 145: as above
Line 160: exp can be replaced by e with superscript
Line 171: Equation is written as abbreviation and should be standardized - see line 166
Line 191 and 193:as line 160
Line 234: CO2/H2O should be writte using subscripts: CO2/H2O
Line 253 and 254: Producers of the radiometer and GPS receiver should be added. Probably they are Apogee Instruments and Garmin
Table 1: E-14 can be replaced by e-14
Figure 2: There are not units in axes
Figure 3: The unit in Cn2 is unclear
Line 330: as line 129
Line 420: as line 171
Line 445: Figure is written as abbreviation and should be standardized - see line 452
Line 496: a dot should be used for algebraic multiplication. * is used for convolution. Previously x was used (line 129)
Line 497: as line 171
Additionally:
- a dot is not necessary after Figure. Figure 1 should be rather than Figure.1.
- quality (resolution) of figures should be improved, especially 3 and 15
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
This manuscript reports on a machine learning modellisation of the refractive index structure parameter of a free-space optical communication link. The authors use of year-long measurements of various meteorogical parameters at the host institution and compare the modelling of the refractive index structure parameters using 6 standard machine learning algorithms. This allows for predicting the outage probability of a a free-space optical link in a turbulent atmospheric environment.
The topic investigated is relevant for free-space optical communications and of interest for a journal like Electronics. The benchmarking of machine learning algorithms for understanding channel transmission in turbulent atmosphere is timely and the presented results are original. In general, the paper is correctly structured and the results decently presented. I would thus recommend this work for publication in Electronics provided the authors address/clarify the following points.
-Figure 4: all the parameters forming the entries of the correlation matrix should be clearly defined and related to what is measured and how in the experimental setup description (sec. 3).
-line 420: the authors refer to the expression for the predicted outage probability given Eq. 20, which comes much later. This is confusing. It is also not completely clear what value for gamma_th is used there.
-Figure 10 is of too poor quality to be readable. I would also suggest that the authors briefly explain in the caption the main ideas of the flowchart .
-Why are the max values all identical in Table 2?
-page 14: it is not clear why the authors chose such an unbalanced availability for their channel. How do the results depend on the 99% required availability?
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 5 Report
This paper study provides a comprehensive comparison between six machine learning regression algorithms for modeling the refractive index structure parameter. This paper is innovative and the authors have done a great deal of work to provide technical support to practitioners. However, I also noticed some flaws that need the authors' attention.
It is suggested that the authors give the contribution points of this paper so that the reader can easily understand the content of this paper.
The C in Eq. 2 is suggested to be changed, just like the representation of permutations in mathematics, which can easily confuse the reader.
The gamma distribution in line 180, how is the gamma gamma distribution?
The figure in the experimental results is not clear, suggest using high resolution.
What is the meaning of P T etc. in the DNN input layer in Figure 11?
Missing literature on deep learning, e.g., Forest fire recognition based on feature extraction from multi-view images. traitement du Signal.
DOI: 10.1109/TITS.2022.3203800
Towards high accuracy pedestrian detection on edge gpus. sensors.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Manuscript is now good to go. Requires just slight modification as Conclusion statement should be in last section.
Author Response
Moved the Conclusion section to the end.
