Electro-Optical Sensors for Atmospheric Turbulence Strength Characterization with Embedded Edge AI Processing of Scintillation Patterns
Round 1
Reviewer 1 Report
In this paper, the author uses a deep neural network model to evaluate the path-integrated refractive index structure parameter with commercial sensors. Also, the DNN model training, validating, and testing in real-time inference experiments was presented. Interesting results and solid works. Thus, I support the publication of this paper in Photonics.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The article utilizes DNN to evaluate the Cn2 value by real-time processing of exposure beacon light flicker patterns (images) captured by the TurbNet sensor photoreceiver. The experimental process of the article is clear, but there are still many problems that need to be further explained.
1. I have consulted the relevant papers mentioned in the author's introduction. The research content of this paper is similar to the Ref. [11] and [12]. The author should explain the difference and novelty of this paper and the above-mentioned papers.
2. What is the research background and application scenarios of this paper?
3. There are many self-citations in the paper, and the analysis of the research status of other researchers is not comprehensive. Authors should expand the Introduction section.
4. In Part 2, The author introduces the composition of the TurbNet sensor “A TurbNet sensor is comprised of optical receiver modules and a laser beacon located at opposite ends of an atmospheric propagation path.”. So, what is the difference between TurbNet sensors and commonly used point-to-point laser communication systems?
5. The DNN-based prediction method described in the article is suitable for systems with fixed conditions. Are the conclusions of the article still general when the link length, receive aperture, etc. vary?
6. Figure 7 presents the results for the LED, and the authors should also give the results for the LB to ensure the integrity of the experiment.
7. The author proposes that the DNN method can increase the prediction rate of Cn2 by 33-43 times compared to the measurement of Cn2 with a scintillator. But the authors seem to ignore the issue of training time.
8. Compared with the direct measurement of Cn2 by the scintillator, the method proposed in this paper needs to collect a large amount of data and the system construction is complicated. Is this suitable for practical engineering applications?
9. The title of Figure 10 is the predicted value of Cn2 for weak and strong turbulence within 60s under TurbNet-LED. However, A and B in Figure 10 are characteristic grayscale scintillation images of LB, which should be corrected by the author. Further, the authors should also supplement the Cn2 pictures corresponding to weak turbulence and strong turbulence within 60s under LB to ensure the integrity and reliability of the experiment.
10. Figure 10 uses Cn2 to directly illustrate weak turbulence and strong turbulence, but the usual turbulence intensity calculation also needs to consider the influence of link distance, etc. The author should give the judgment basis for weak turbulence and strong turbulence (such as atmospheric coherence constant).
11. The authors have only verified a good correspondence between the scintillator measurements and the TurbNet sensor's turbulence intensity assessment (correlation coefficients range from 0.7 to 0.9) with only 3 experiments. The authors should conduct multiple experiments to verify the accuracy of the conclusions.
12. There are many writing errors in the article, such as "TurbNet-LP" on page 3, etc. Authors should carefully check the full text for spelling and grammatical errors.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The refractive index structure parameter C2n represents the sole measure of atmospheric turbulence strength, which is essential for predictive performance assessment of various electro-optical systems including directed energy, free-space laser communication, active imaging and optical surveillance.
In this article, a deep machine learning-based electro-optics systems (TurbNet sensors) were developed to measure C2n at a high temporal resolution by processing short-exposure intensity scintillation patterns originating from a laser or LED beacon 7 km away. The results of the inference trials demonstrated correlation coefficient ranging from 0.7 to 0.9 between the scintillometer measurements and turbulence strength assessment with the TurbNet sensors.
However, two issues need to be discussed:
1. The intensity scintillation level of laser beam is commonly characterized by the Rytov variance. Is Rytov variance appropriate to characterize intensity scintillation level of LED?
2. In the text, lines 175 to 177 explain, “In the case of the TurbNet-LED dataset containing a large number of instances, the desired uniformity in data distribution was achieved by selective partial removal of data instances belonging to the medium strength turbulence range.” In order to obtain the desired data distribution, is it appropriate to discard some data?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
This manuscript introduces the use of LED to obtain the atmospheric structure constant Cn2, and the use of CNN network training calculation. The whole article seems very meaningful, but as the reviewers know, there have been many similar reports before. However, the author's CNN calculation results are relatively new, and most of the questions asked by reviewers have been answered. It is recommended to accept,
but the following questions need to be answered
1. LED and LD have different degrees of coherence. This is how to calculate Cn2 and ensure that Cn2 is accurate.
2. For this flare image, what is the formula for calculating Cn2? Suggestions are given in detail.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Based on the author's reply, I agree to publish it in its present form.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
