An Improved Back-Projection Algorithm for GNSS-R BSAR Imaging Based on CPU and GPU Platform
Round 1
Reviewer 1 Report
Authors proposed back-projection algorithm using CPU and GPU for GNSS-R BSAR. Figure quality looks good and clear to be seen. Authors showed simulated and measured results using developed algorithms and showed some improvement in azimuth resolution. These results are quite clear to improve the azimuth direction resolutions. However, authors need to provide some missing references. Therefore, manuscript could be minor revision if authors follow the comments as below.
1. Please increase label sizes of Figures 5 and 6.
2. Please increase x and y axes size of Figure 7.
3. Please increase font sizes of Figures 8a and 11.
4. Please provide FFT in Figure 11. It might be unwanted harmonics and noises.
5. Please provide the axial resolution if possible because authors showed the resolution in azimuth direction only.
6. In the reference section, please provide city, data, country information for conference papers.
7. In the reference section, authors should use abbreviated journal names.
8. Please provide data availability section.
9. In Table 1, please make space between values and units.
10. Please provide the reference (GNSS-R BSAR is one of the emerging applications of GNSS-R technology~).
11. Please provide the reference (GNSS-R BSAR system needs to process a large amount of data, but the memory of the~)
12. Please provide the reference (the pre-processing of the raw data should be compatible with an op~)
13.Please provide the reference (Based on the Stop-Go model, the length of a rang~).
14. Please provide the reference (The BPA is a time-domain imaging algorithm whose main principle is based~).
15. Please provide the reference (BPA requires accurate time and location of the transmitter, receiver~).
Author Response
Dear reviewer
We are truly grateful for the comments and suggestions from your review. Based on these comments and suggestions, we have made careful modifications to the manuscript. You will find our point-by-point responses to the comments below. The responses are in the word document
Best regards,
Shiyu Wu
Author Response File: 
Author Response.docx
Reviewer 2 Report
The paper deals with bistatic SAR imaging using a GNSS satellite as the transmitter and a stationary antenna as the receiver. Bistatic SAR focusing is achieved by time-domain processing through an implementation of the back-projection algorithm. The topic is interesting, but the paper is very weak from some different perspectives:
1) The article is not well-written. English usage and quality are very poor. There are grammar errors and the meaning of many sentences and paragraphs is hard to catch.
2) The reference list is incomplete. Most of the sections could be drastically shortened just quoting suitable literature.
3) Last, but note least, the novelty of the paper is questionable. The authors claims that they implemented and improved BPA and that their contribution is mainly related to the exploitation of GPU processing together with CPU to speed up the processing time. Indeed, the BPA is implemented in a quite standard fashion with no specific new feature. Moreover, the use of parallelized processing via GPU to make BPA faster has already been to topic of a number of papers, see for instance,
Werner et al., “GPU-BASED PARALLELIZED TIME-DOMAIN BACK-PROJECTION PROCESSING FOR AGILE SAR PLATFORMS”, Proc. of IGARSS 2014.
Fasih and Hartley, “GPU-accelerated synthetic aperture radar backprojection in CUDA”, IEEE Radar Conf. 2010.
Que et al., “Real-Time Processing of SAR Images for Linear and Non-Linear Tracks”, nternational Radar Symposium, 2016.
- Small, “Lessons Learned Using NVidia GPUs Within SAR Applications”, SAR/GPU Workshop 2012.
The authors should therefore clarify better the novel contribution of their manuscript.
In the end, the paper cannot be recommended for publication in the present form.
My suggestion for the authors is to try to shorten the paper, reducing it to a short communication, quoting suitable literature and pointing out better their own contribution.
Detailed comments
Page 1, line 14. Define the acronym CPU, and in general define all the acronyms the first time they appear in the text.
Page 1 lines 16-18 “It was used…pre-processing”. Unclear, improve English usage.
Page 1 lines 24-25 “…with a size of on…”. It seems a number is lacking, please complete.
Page 2, line 53. Please consider replacing “and” with “but”.
Page 2, line 55 “the GNSS signal bandwidth is relatively narrow to the traditional SAR signals”. Poor English. Rephrase.
Page 2, line 68. You made a list of algorithms. According to the paragraph the reader can conclude that BPA is a frequency domain algorithm. Please fix.
Page 2, line 93 “which has improved the satellite orbit and positioning accuracy”. Unclear, please improve.
Page 3, lines 97-103 and lines 104-112. These paragraphs seem to be ill-placed. The discussion they present is more suitable to a conclusion than an introduction. Within an introduction, maybe, it is better to just highlight that both simulation and experimental results are going to be presented.
Page 3, line 118-119. Unclear/incomplete. Check the English usage.
Page 4, line 143 “SAR data reformatting”. Could you clarify what is intended as “SAR data reformatting” and the way it is accomplished?
Page 4, line 143 “which can be written in a 2D form”. The subject of the sentence is obscure. What can be rewritten in a 2D form. Moreover, maybe 2D format?
Page 4 and page 5, lines 158-164. The description is very unclear. Please quote suitable literature on BPA and explain what you mean with “coherent accumulation of signals” and “coherent accumulation of sub-images”.
Section 3.1. No reference is quoted in this section, even if most of the discussed steps are part of well-assessed and standard processing.
Page 5, line 169-170, “the data used for BPA is compatible”. Unclear. Please rephrase or explain better.
Page 5, line 175. Maybe “improve” instead of “improved”.
Page 5, line 177. Insert a blank space between “parameters” and “(“.
Page 5, line 194 “the doppler delay is skipped according to the direct signal tracking results”. The reader cannot catch the meaning of this sentence. Consider revising.
Page 6, lines 198-202. Put a reference to ISG products and product features.
Page 6, line 207. Maybe “image area” or “imaged area” instead of “imaging area”.
Page 6, lines 207-209. I am not convinced that your synthetic aperture time can start when the navigation satellite antenna beam enters the imaged area. Indeed, the beam of the satellite is huge, and as long as the satellite is in view of a given point, the satellite is covering that point. In this way you will achieve and integration time of several hours which is clearly not the case of the presented experiments. Please discuss and comment.
Page 6, lines 211-213 “To obtain…area”. Obscure. Please rephrase and improve the clarity of presentation
Page 6, lines 213-214 “Finally, determine…area”. Poor English and unclear. Please fix.
Page 6, lines 214-218. Again, hard to follow. Quote suitable references, and revise the paragraph.
Page 6, section 3.2. It is not clear to the reader why the presented BPA is referred to as improved BPA. It is rather a quite standard BPA implementation.
Page 6, section 3.1.1. The section can be significantly shortened. Actually, there is no need to explain what is range compression, just quote a good textbook or manual. Moreover, note that the correct number for the section should be 3.2.1.
Page 7, section 3.1.2. Beside the fact that the section should be numbered as 3.2.2, I found nothing new about the BPA implementation. Again, quite standard with no novelty.
Page 7, line 255. Explain the meaning of “area pixel”. Maybe pixel area?
Page 7, line 267 “slope range”. What is intended as “slope range”. Please explain or add a reference.
Page 8, line 273 “where can be expressed as”. Poor English. What “can be expressed”?
Page 8, lines 275-276 “Formula 11”. What do you mean as formula 11? Maybe equation 11? However, Eq. (11) seems to be not so useful for the determination of the computational cost. Please check and fix. The same comment applies to line 278 “Formula 12 and 13”.
Page 8, line 280. So, there is an interpolation kernel involved. It seems to be strange since you did not mention any interpolation or interpolation kernel when you described the algorithm. Please clarify.
Page 8, lines 281-282. Please be more precise. Different frequency domain algorithms lead to very different computational loads. Which type of frequency domain algorithm are you referring to? Please consider quoting suitable literature.
Page 8, line 291 “the raw data must be divided into as much as the GPU…”. The sentence looks incomplete. Please check and fix.
Page 9, lines 301-324. Is it really necessary to make range compression parallelized? In general, those steps are pretty fast and not time consuming. Your results (table 2) seem to confirm this idea. Leaving range compression to the CPU also reduces the memory requests to the GPU, since FFT and IFFF require more memory than BPA, especially if the latter is implemented in a pixel-oriented fashion as the authors did. Please discuss and comment.
Page 9, lines 335-345, about the difference between “pixel-oriented” and “(range code)-oriented”. The advantages of the pixel-oriented solution are well-known in the BPA literature. Again, adding suitable references even this section could be significantly shortened. Please also consider using a better term than “(range code)-oriented”. Just remove the parenthesis or use “code-oriented”.
Page 10, lines 355-367. These are the steps one would expect from a standard BPA applied in a parallelized fashion. No specific novel authors’ contribution seems to arise. Please discuss and comment.
Page 10, line 382, Figure 6 and Table 1. The text state that the geometric configuration of the satellite is shown in Figure 6, while the latter only shows the geometry of the receiver and the simulated scene. Even Table 1 does not clarify the simulated position or trajectory of the GNSS satellite.
Page 11, lines 440-407. The authors state that simulation results, concerning resolutions and other performance parameters, are in line with theoretical values. Please can you clarify the method used to estimate the theoretical values, or jut quote suitable references.
Page 11, Figure 7a and Table 3. The range results might be too optimistic since they only consider the autocorrelation between a PNR signal of a given satellite and a time delayed copy of the same signal. Residual cross-correlation of PRN signals relevant to different GNSS satellites still exist (even if weak) which is expected to raise side lobe level. Please check and comment.
Page 11, Table 3. I think it is useless to show that the parameters achieve the same value for both CPU and GPU&CPU. This merely verify that the algorithm was correctly implemented, which is not a result worth of being mentioned in a journal paper. The comments apply throughout the paper, any time CPU and CPU&GPU results are compared just to show that they are in agreement (e.g. lines 445-447, Figure 9, Table 6, lines 476-481).
Page 12, line 431 “GPS satellite PRN3 was selected as the optimal signal source”. Please can you explain why is PRN3 “optimal”? Maybe it is just the satellite with the easier bistatic configuration (low squint, geometry, same side as that of the receiver with respect to the scene,…). Considering bistatic scattering other Tx/Rx geometry can be even better depending on the parameter to be optimized or the scene property to be measured. Please clarify.
Page 13, Table 5. What’s the pitch angle? Maybe elevation, incidence, or grazing.
Page 13, lines 445-446 “this shows that the image quality obtained by using GPU and CPU to run BPA is the same, which is consistent with the theoretical value.”. It is not completely clear to the reader what’s the theoretical value measuring the image quality.
Page 13, Figure 9 and Table 6. I think they should be deleted from the paper, the fact that CPU and CPU&GPU lead to the same results (Figure 9) and the reduction of the computation time by parallelilzed processing have been already assessed by simulation just a few pages before. Thus, there is no need for showing them again.
Page 14, lines 457-458 “Obvi-457 ously, there is a good coincidence between the radar image and the optical photo”. It is not so obvious indeed, especially is you look at radar images only. The interpretation is made easier by the matching with a very high resolution optical image.
Page 14, lines 460-461 “Since the height of the echo-antenna is equivalent to the target area”. How is it possible that the height is equivalent to an area. Please correct and improve English quality.
Page 15, line 487 “this letter”. Please fix.
Page 15, lines 496-500. I suggest the authors to anticipe this statement in the introduction just to clarify the context and the final goal of their activity. It is also evident that more precise, phase-based, analysis are required for deformation monitoring, which is likely to be the topic of future authors’ activities.
Author Response
Dear reviewer
We are truly grateful for the comments and suggestions from your review. Based on these comments and suggestions, we have made careful modifications to the manuscript. You will find our point-by-point responses to the comments below. The responses are in the word document
Best regards,
Shiyu Wu
Author Response File: Author Response.docx
Reviewer 3 Report
see attached file
Comments for author File: 
Comments.pdf
Author Response
Dear reviewer
We are truly grateful for the comments and suggestions from your review. Based on these comments and suggestions, we have made careful modifications to the manuscript. You will find our point-by-point responses to the comments below. The responses are in the word document
Best regards,
Shiyu Wu
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
I thank the authors for their clear and exhaustive answers to my comments. The modifications they made to the manuscript are more than adequate to consider the paper ready for publication. Some minor comments are listed below. The authors can fix them without the need for a further review.
Lines 284-285 “the pixel slant range was calculated with is delay information to compensate phase by constructing one compensation factor.” Please check “with is delay information”.
Line 340. Check herei if [39] is the right reference.
Lines 344-350 “we ignore the cross-correlation interference caused by other navigation satellite signals” is repeated twice in two close lines. Please rephrase or just delete the repeated sentence.
Author Response
Dear Reviewer:
We are truly grateful for the comments and suggestions from your review. Based on these comments and suggestions, we have made careful modifications to the manuscript. You will find our point-by-point responses to the comments below. The responses are in the word document.
Author Response File: Author Response.docx
Reviewer 3 Report
The paper is ready for publication
Author Response
Dear Reviewers:
We are truly grateful for the comments and suggestions from your review. Based on these comments and suggestions, we have carefully revised several grammars in the manuscript.
Kind regards
Shiyu Wu
