A Real-Time BLE/PDR Integrated System by Using an Improved Robust Filter for Indoor Position

Round 1

Reviewer 1 Report

The authors proposed improving the fusion algorithm of BLE and PDR. Specifically, in the PDR method, the authors used an improved Mahony complementary filter based on the pedestrian motion states to estimate the heading angle, reducing the drift error. Then, they utilized an improved robust filter to detect and restrain the gross error of the BLE fingerprint method. The experimental results show their proposed fusion algorithm outperforms classic EKF. The paper is well organized and clearly presented, and I can easily follow the paper. However, the paper has two main drawbacks.

The first is that the proposed fusion algorithm is applied to the basic PDR and WKNN algorithms, which were proposed many years ago and have been extensively improved till now. For instance, deep learning based PDR algorithms [1-3] show promising performance in long-time tracking, without the assumptions for smartphone postures. It is the same for Fingerprinting algorithms. The novelty of this study is doubtful, since the drift error of PDR and the gross error of fingerprinting should or might have been solved.

The second is the experimental section. The author used a very simple indoor environment as the test bed and the test data is also not enough, which is just a trajectory collected from one tester. As we know, the experimental environments and testers have a large impact on the results. We cannot conclude anything from the experimental results, since the experimental data is too simple.

In a nutshell, the authors should prove that their study did contribute to the indoor positioning community based on concrete and abundant experimental data.

[1] Ronin: Robust neural inertial navigation in the wild: Benchmark, evaluations, and new methods

[2] TLIO: Tight learned inertial odometry

[3] IDOL: Inertial Deep Orientation-Estimation and Localization

Author Response

Point 1: The first is that the proposed fusion algorithm is applied to the basic PDR and WKNN algorithms, which were proposed many years ago and have been extensively improved till now. For instance, deep learning based PDR algorithms [1-3] show promising performance in long-time tracking, without the assumptions for smartphone postures. It is the same for Fingerprinting algorithms. The novelty of this study is doubtful, since the drift error of PDR and the gross error of fingerprinting should or might have been solved. 


Response 1:

Thanks for your suggestion. The fusion algorithm combing the PDR method and KNN have been researched for many years, it still has advantages of real-time, low computational load, and ease of implementation. We have carefully read the references you mentioned and added them to the citation of the manuscript. The data-driven inertial navigation technology introduced by the references has been summarized in the related work in lines 137-150, which has guided mi in our later research work. Although the data-driven inertial navigation has a good performance in long-time tracking based on the sophisticated deep learning technology, it needs a large amount of data to train and extra equipment to get the ground-truth trajectory in advance and for real-time positioning, the smartphone will be under greater computational load.

The drift error of PDR and the gross error of fingerprinting can only be suppressed and there is no general model to solve this problem under multiple conditions. Our experiments were conducted in an environment with complex multipath effects and electromagnetic interference. Under such conditions, the BLE fingerprint method is more prone to occur coarse errors. To address this problem, we proposed a robust filter for the real-time position.

 Our contributions are as follow:

1) We found that the errors of the BLE fingerprint method are not only related to the signal fluctuation but also affect by scanning numbers of BLE beacons after statistically analyzing the real-time signal data in a harsh environment. When the scanning BLE beacon numbers are few, coarse errors will more likely occur.

2) We found that the accuracy of the heading is also affected by the motion states of the pedestrian. An improved Mahony complementary filter is introduced to keep the heading angle stable by adaptively changing the control parameters in the filter after considering the different people's motion states.

3) To meet the demand of real-time position and considering the computational load of the smartphone, we adopt the EKF method to solve the nonlinear fusion problem to combine PDR with the BLE fingerprint position method to provide the real-time position service. To cope with the gross error caused by the BLE fingerprint method in a harsh environment, a robust filter based on the EKF was proposed. The robust filter detected the gross error at different granularity by constructing a robust vector changing the observation covariance matrix of the extended Kalman filter (EKF) adaptively when the application is running. The experimental results demonstrate that the proposed method has better performance at position accuracy and stability.

Point 2: The second is the experimental section. The author used a very simple indoor environment as the test bed and the test data is also not enough, which is just a trajectory collected from one tester. As we know, the experimental environments and testers have a large impact on the results. We cannot conclude anything from the experimental results, since the experimental data is too simple.

Response 2: Considering your comments that the experimental environment was simple, we have added a description of the experimental environment in lines 477-481 and increased a positioning scene to rate the proposed method, as shown below:

The new positioning scene is 331 test site, which is 21.72 m long and 7.75 m wide. The blue triangle in the Figure represents the BLE beacons. In this location scene, we conduct a fusion position experiment. The experimental result is shown as below:

From the Figure, the green line that represents the proposed method is closer to the real track and smoother in some corners compared to other methods. We can conclude that the proposed method can restrain the gross error and had a better performance in the real-time position.

Author Response File: Author Response.pdf

Reviewer 2 Report

Like others in the field of indoor position/localisation these authors use the terms localisations and positioning as synonyms, see for example line 34: “To provide a reliable, stable location in indoor environments, many types of indoor positioning technologies … But, in this paper, the author propose an indoor positioning systems, resulting in coordinates in a local reference system. To use ‘localisation’, the obtained locations should be descriptive, meaningful. See this paper: https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/III-4/89/2016/ chapter 3 on position versus location. For that, this reviewer supposes to use the term ‘position’ in this paper as much as possible, and avoid the term ‘location’ when it does not link back to “a physically defined space in a building. These can be rooms, corridors, stairs, hall etc.” So, for example in line 63, “By contrast, wireless localization such as BLE fingerprint position can obtain absolute location without cumulative error but has poor accuracy position estimation.” one has not to ask “what’s meant with absolute location”?

In the test 54 BLE beacons are installed. One can ask himself whether or not this amount of BLE access points is realistic. Are 18 BLE beacons for each floor a reasonable amount of beacons? What if each of these beacons are installed inside a room, near entrances? Would they then not serve as intended: to track (locate) someone near the (nearest) BLE?

In line 418-419 it is stated: “After eliminating the few numbers of scanning RSSI real-time fingerprints, the mean position accuracy and the root-mean-square error (RMSE) were 2.312m and 2.043m, respectively.” Given the layout of the buildings, and the layout of the corridors (width of a couple of meters) these measures are somewhat to generic, or (see discussion on position versus location) should be related to the ‘physically defined space in the building’. Which part of the corridors did perform ‘better’? The corners? Near the entrance of a room? Thus: where is an ‘accurate position’ needed to have to make a decision (to enter a room, to leave a floor, etc.).

In line 514 it is stated: “In the process of real-time positioning, bad position results are obtained by the BLE fingerprint method.”. is this conclusion not too harsh? Would BLE fingerprinting not be ‘good enough’ for indoor localisation? Thus indication someone is located inside a certain room? Walking along a part of a corridor? Thus: BLE fingerprinting in itself can be fine for localisation (stating someone is in a physically defined space in a building”, but not for accurate indoor positioning (but then you should also address the use-cases of why this kind of accurate indoor positioning is needed anyhow.

Reference 27 reads: “Zengke, L.; Chunyan, L.; Jingxiang, G.; Xin, L. An Improved WiFi/PDR Integrated System Using an Adaptive and Robust Filter for Indoor Localization. International Journal of Geo-Information 2016, 5, 224.”. But this paper is refenced as: Li et al. [27]. So, the first and surname of this Chinese author are twisted ?

Typo: line 102-103: One sentence.

Author Response

Point 1: Like others in the field of indoor position/localisation these authors use the terms localisations and positioning as synonyms, see for example line 34: “To provide a reliable, stable location in indoor environments, many types of indoor positioning technologies … But, in this paper, the author propose an indoor positioning systems, resulting in coordinates in a local reference system. To use ‘localisation’, the obtained locations should be descriptive, meaningful. See this paper: https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/III-4/89/2016/ chapter 3 on position versus location. For that, this reviewer supposes to use the term ‘position’ in this paper as much as possible, and avoid the term ‘location’ when it does not link back to “a physically defined space in a building. These can be rooms, corridors, stairs, hall etc.” So, for example in line 63, “By contrast, wireless localization such as BLE fingerprint position can obtain absolute location without cumulative error but has poor accuracy position estimation.” one has not to ask “what’s meant with absolute location”?

Response 1:

Thanks for your valuable suggestion. We had read the paper to figure out the difference between ‘location’ and ‘position’. We had corrected the ambiguities in the manuscript according to your comments. We replaced the word ‘location’ by using ‘position’ in the original sentence in lines 36 and 68.

Point 2: In the test 54 BLE beacons are installed. One can ask himself whether or not this amount of BLE access points is realistic. Are 18 BLE beacons for each floor a reasonable amount of beacons? What if each of these beacons are installed inside a room, near entrances? Would they then not serve as intended: to track (locate) someone near the (nearest) BLE?

Response 2:

Thanks for your comments. The deployment of the Bluetooth beacon in the test site is based on the principle of optimizing beacon placement by maximizing localization accuracy and satisfying a predefined coverage degree, which had been recited in the paper as reference 24. The deployment solution takes into account the various conditions in the real environment and gives the principle of optimization. The purpose of the deployment of BLE beacons is to perform the fingerprint positioning method and to verify the effectiveness of the solution at the same time.

Point 3: In line 418-419 it is stated: “After eliminating the few numbers of scanning RSSI real-time fingerprints, the mean position accuracy and the root-mean-square error (RMSE) were 2.312m and 2.043m, respectively.” Given the layout of the buildings, and the layout of the corridors (width of a couple of meters) these measures are somewhat to generic, or (see discussion on position versus location) should be related to the ‘physically defined space in the building’. Which part of the corridors did perform ‘better’? The corners? Near the entrance of a room? Thus: where is an ‘accurate position’ needed to have to make a decision (to enter a room, to leave a floor, etc.).

Response 3:

Thanks for your comments. When using the KNN method for the BLE fingerprint position, the real-time signal is collected and matched with the fingerprint database. The number of acquired signals is very small, it means that the matching operation has less useful information and the matching process will produce coarse differences. From point of view of an analysis on the signal and algorithm, eliminating the few numbers of scanning RSSI real-time fingerprints can avoid having a large error.

Point 4: In line 514 it is stated: “In the process of real-time positioning, bad position results are obtained by the BLE fingerprint method.”. is this conclusion not too harsh? Would BLE fingerprinting not be ‘good enough’ for indoor localisation? Thus indication someone is located inside a certain room? Walking along a part of a corridor? Thus: BLE fingerprinting in itself can be fine for localisation (stating someone is in a physically defined space in a building”, but not for accurate indoor positioning (but then you should also address the use-cases of why this kind of accurate indoor positioning is needed anyhow.

Response 4:

Considering your comments, we found that our expressions were not rigorous. The BLE fingerprinting method has been widely used because of its stable and reliable accuracy and other characteristics. In our location scenario, the center part of the test site is a hollow space enclosed with glass. When the radio signal of BLE propagates, there is a severe multipath effect. We found a high probability that the gross error came from the BLE fingerprint positioning method. When the accurate result was obtained by the BLE fingerprint, it was able to restrain the cumulative error caused by the PDR method by the fusion filtering method. Then, we modified the sentence ‘In the process of real-time positioning, poor position results may be obtained by the BLE fingerprint method’ to make our expressions more rigorous.

Point 5: Reference 27 reads: “Zengke, L.; Chunyan, L.; Jingxiang, G.; Xin, L. An Improved WiFi/PDR Integrated System Using an Adaptive and Robust Filter for Indoor Localization. International Journal of Geo-Information 2016, 5, 224.”. But this paper is refenced as: Li et al. [27]. So, the first and surname of this Chinese author are twisted ?

Response 5:

Thanks for your comments. I have modified the corresponding information according to your suggestion

Point 6: Typo: line 102-103: One sentence.

Response 6:

Due to the restructuring of the article, the corresponding sentences were deleted, thank you for your carefulness

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper explores the implementation of an Extended Kalman Filter (EKF) with robust filter to obtain a less erroneous indoor positioning system. The proposed EKF combines the observations from BLE fingerprinting technique and PDR using smartphone. Then, with the robust filter in the EKF, the noisy observation are highly cleaned to get improved indoor positioning system.

I would like to suggest the authors do the following revisions:

• heavily improve the writing style as it does not follow the English grammar,
• add more recent references in noise reduction of indoor positioning system with BLE and smartphone,
• add years information in the references, some were missing,
• inconsistencies in using abbreviation (CCS/CGS) (section 2.3), other similar inconsistencies may also occur in other parts,
• break long paragraphs, e.g. in introduction. I suggest to put the related works in a separate section with more focused ideas in every paragraph to easier reading,
• explicitly highlight the contribution in the introduction as opposed to the state-of-the-art research in BLE/PDR integration, some may include more sophisticated machine learning techniques,
• add more information about the positions of the BLE fingerprint data collection in Figure 8,
• with different settings of BLE beacon and movements, is it possible to have similar improvement? I suggest the authors to add more experiment scenario with various settings. In addition, a single trajectory for experiment (section 3.3) may not suffice to reflect the real situation (some activities like walking and idle, or even running, may occur due to some people in the building),
• adding some machine learning-enhanced other than KNN and indoor positioning techniques other than EKF as comparison methods in the experiment. If the proposed EKF with robust can outperform in positioning error and latency, the proposed EKF should be more suitable in the real-world indoor positioning system,
• Q: for the robust filter, is there any important aspect to use 5% significance level as limit? (section 2.5)

Author Response

Point 1: heavily improve the writing style as it does not follow the English grammar

Response 1:

Thanks for your valuable suggestion. We carefully checked the paper and corrected some grammatical errors, and had experienced colleagues guide the revision

Point 2: add more recent references in noise reduction of indoor positioning system with BLE and smartphone,

Response 2:

Thanks for your suggestions. Some recent references in noise reduction of the indoor positioning system and smartphone have been cited and summarized the references in the related-work section.

Point 3: add years information in the references, some were missing

Response 3:

Thanks for your reminder. The entire reference had been checked and add the missing year information.

Point 4: inconsistencies in using abbreviation (CCS/CGS) (section 2.3), other similar inconsistencies may also occur in other parts,

Response 4:

Thanks for your suggestions. We had corrected the incorrect misuse of abbreviations in section 2.3 and check other parts to ensure consistent use.

Point 5: break long paragraphs, e.g. in introduction. I suggest to put the related works in a separate section with more focused ideas in every paragraph to easier reading,

Response 5:

Thanks for your valuable suggestion. We had broken the introduction section into two parts, the introduction, and the related work. The related work outlined the three subsections of BLE-based position, self-contained position, and fusion position algorithms respectively. New structure and smaller paragraphs make the manuscript easier to understand

Point 6: explicitly highlight the contribution in the introduction as opposed to the state-of-the-art research in BLE/PDR integration, some may include more sophisticated machine learning techniques.

Response 6:

Thanks for your valuable suggestion, I have summarized the contributions of the paper at the end of the introduction section. The contributions are as follow:

1) We found that the errors of the BLE fingerprint method are not only related to the signal fluctuation but also affect by scanning numbers of BLE beacons after statistically analyzing the real-time signal data in a harsh environment. When the scanning BLE beacon numbers are few, coarse errors will more likely occur.

2) We found that the accuracy of the heading is also affected by the motion states of the pedestrian. An improved Mahony complementary filter is introduced to keep the heading angle stable by adaptively changing the control parameters in the filter after considering the different people's motion states.

3) To meet the demand of real-time position and considering the computational load of the smartphone, we adopt the EKF method to solve the nonlinear fusion problem to combine PDR with the BLE fingerprint position method to provide the real-time position service. To cope with the gross error caused by the BLE fingerprint method in a harsh environment, a robust filter based on the EKF was proposed. The robust filter detected the gross error at different granularity by constructing a robust vector changing the observation covariance matrix of the extended Kalman filter (EKF) adaptively when the application is running. The experimental results demonstrate that the proposed method has better performance at position accuracy and stability.

Point 7: add more information about the positions of the BLE fingerprint data collection in Figure 8

Response 7:

Thanks for your suggestion. The content about the BLE fingerprint data collection in Figure 8 had been added in lines 494-497.

Point 8: with different settings of BLE beacon and movements, is it possible to have similar improvement? I suggest the authors to add more experiment scenario with various settings. In addition, a single trajectory for experiment (section 3.3) may not suffice to reflect the real situation (some activities like walking and idle, or even running, may occur due to some people in the building),

Response 8:

Thanks for your suggestion. The proposed method can improve the accuracy under the different BLE settings and movements. The analysis of pedestrian motion states is mainly to make the heading angle more stable, which can indirectly improve the accuracy. To verify the effectiveness of the method, the new positioning scene was added to conduct the comparison experiments. The new positioning scene is shown as below: as shown below:

The new positioning scene is 331 test site, which is 21.72 m long and 7.75 m wide. The blue triangle in the Figure represents the BLE beacons. In this location scene, we conduct a fusion position experiment. The experimental result is shown as below:

From the Figure, the green line that represents the proposed method is closer to the real track and smoother in some corners compared to other methods. We can conclude that the proposed method is possible to have similar improvement with different settings.

Point 9: adding some machine learning-enhanced other than KNN and indoor positioning techniques other than EKF as comparison methods in the experiment. If the proposed EKF with robust can outperform in positioning error and latency, the proposed EKF should be more suitable in the real-world indoor positioning system,

Response 9:

Thanks for your suggestion. First, we had added some machine learning-enhanced other than KNN method to process the same group data. The comparison result is shown as below:

From the Figure and table, we could find that the KNN method yields the highest accuracy. The other machine-enhanced method may have low accuracy due to overfitting. Second, we utilized the particle filter as a comparison method in the C7 test site. The experiment added a set of data and  the result is shown below.

From the comparison of the curves with the real trajectory on the figure, the proposed method is superior to the classical EKF. The particle filter also performs well in some positions compared to the proposed method. However, the particle filter requires the construction of a large number of particles requiring a heavy computational load. The proposed method is more suitable for real-time localization than the particle filter method.

Point 10: for the robust filter, is there any important aspect to use 5% significance level as limit? (section 2.5)

Response 10:

The 5% significance level here corresponds to the double sigma principle in the normal distribution. In the paper, The analysis of the errors at different granularity is based on the assumption of normal distribution. The 5% significance level is the boundary to determine if the gross error occurs.

Author Response File: Author Response.pdf

Reviewer 4 Report

In the reviewed manuscript the real-time Bluetooth low energy (BLE)/pedestrian dead-reckoning (PDR) integrated indoor positioning system was proposed. In my opinion the reviewed manuscript is very interesting and I would like to see this paper publish in Applied Sciences Journal but I have a following suggestions:

• In order to make more comprehensible the presentation, at the beginning of the paper, before the Introduction section, a nomenclature section, not numbered, should be added with a list of all the used symbols and their meaning.
• I don't feel qualified to judge about the English language but I think that the whole paper should be corrected by native spiker.
• Figure 9a is illegible.
• The novelty of the paper must be more clearly demonstrated.
• In section Conclusions, should be mentioned advantages and disadventages of the proposed solution.
• Please supplement the paper with further options for modification/extension of the proposed solution.

Author Response

Point 1: In order to make more comprehensible the presentation, at the beginning of the paper, before the Introduction section, a nomenclature section, not numbered, should be added with a list of all the used symbols and their meaning.

Response 1:

Thanks for your valuable suggestion. We had added the nomenclature section which included a list of all the used symbols and their meaning at the beginning of the paper.

Point 2: I don't feel qualified to judge about the English language but I think that the whole paper should be corrected by native spiker.

Response 2:

Thanks for your comments. We first checked the paper carefully and then had it rechecked by experienced colleagues

Point 3: Figure 9a is illegible..

Response 3:

Thanks for your comments. We have recreated the graph and made the line of the second method thinner without covering the following data to make the whole Figure clearer. The new Figure is as shown bellow:

Point 4: The novelty of the paper must be more clearly demonstrated.

Response 4:

Thanks for your suggestions. We have broken the introduction and explicitly highlighted the novelty of the paper at the end of the introduction section.  The novelty of the paper is summarized as below:

1) We found that the errors of the BLE fingerprint method are not only related to the signal fluctuation but also affect by scanning numbers of BLE beacons after statistically analyzing the real-time signal data in a harsh environment. When the scanning BLE beacon numbers are few, coarse errors will more likely occur.

2) We found that the accuracy of the heading is also affected by the motion states of the pedestrian. An improved Mahony complementary filter is introduced to keep the heading angle stable by adaptively changing the control parameters in the filter after considering the different people's motion states.

3) To meet the demand of real-time position and considering the computational load of the smartphone, we adopt the EKF method to solve the nonlinear fusion problem to combine PDR with the BLE fingerprint position method to provide the real-time position service. To cope with the gross error caused by the BLE fingerprint method in a harsh environment, a robust filter based on the EKF was proposed. The robust filter detected the gross error at different granularity by constructing a robust vector changing the observation covariance matrix of the extended Kalman filter (EKF) adaptively when the application is running. The experimental results demonstrate that the proposed method has better performance at position accuracy and stability.

Point 5: In section Conclusions, should be mentioned advantages and disadventages of the proposed solution.

Response 5:

Thanks for your comments. We had revised the conclusion section and mentioned the advantages and disadvantages of the proposed solution. The corrected part is from line 637 and line 652.

Point 6: Please supplement the paper with further options for modification/extension of the proposed solution.

Response 6:

Thanks for your suggestions. In the conclusion section, we have made further options for modification/extendion of the proposed solution. The main content is demonstrated as follow:

In addition to the detection and suppression of coarse differences in the observation noise matrix, some modifications of the state transition matrix will be made based on the people’s motion states. How the variance matrix of process noise adaptively changes according to the people’s motion states will be investigated in-depth in future work. Combining with other positioning methods such as map matching, landmarks matching for multi-mode fusion positioning

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author added one test path (Figure 15 e). However, how could PDR-based approaches are evaluated based on just two short paths below or around 100 meters? Normally, the test paths of over several km are quite common for evaluating PDR-based approach.

Author Response

Point 1: The author added one test path (Figure 15 e). However, how could PDR-based approaches are evaluated based on just two short paths below or around 100 meters? Normally, the test paths of over several km are quite common for evaluating PDR-based approach. 


Response 1:

Thanks for your suggestion. We couldn’t agree more with you and find your comments very valuable. It is feasible to evaluate the PDR-based approach using a test path of over several kilometers. However, the test site of our experiment is relatively small, it would be a bit difficult to implement a test path of several kilometers to evaluate the PDR-based approach. We take another evaluation approach to evaluate the methods mentioned in the manuscript. The procedure of the evaluation approach is described as follows. The real-time positioning results of the PDR and other fusion methods are output separately in the system, and they do not affect each other. As mentioned in the paper in lines 570-575, the reference trajectory was designed of which each section is a straight line in advance and some test points have been constructed along the trajectory. When the experiment was carried out, the user started from the start point and reached the endpoint along the reference trajectory at a constant speed. The step length of each step was considered to be equal and the number of steps had been recorded during the experiment. Then the test points along the trajectory had been constructed based on the above conditions. For each step, the system outputs the PDR and the corresponding hybrid positioning results. Then the results are applied to compare with the corresponding test points to evaluated the accuracy of the PDR-based approach. At the same time, the output real-time points results could be connected to the path. By comparing the path with the reference trajectory, we can also visually evaluate the performance of the PDR-based approach.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have revised the manuscript accordingly.
However, I would suggest some minor adjustments to the figures for better reading:

• The font size and the points of right-hand part of Figure 3 can be enhanced
• Figure 8 can be separated with 8(a) and 8(b) where the size of triangle of 8(a) can be slightly increased
• Font size of figure 15 (a-e) can be increased
• The test depicted in Figure 16 can be slightly introduced earlier in 4.3 rather than a sudden appearance in the last 

Author Response

Point 1: The font size and the points of right-hand part of Figure 3 can be enhanced.

Response 1:

Thanks for your suggestions. The font size in Figure 3 and the points of right-hand part of Figure 3 have been enlarged to make the figure more clear after modification. The modifications have been made in lines 280-283.

Point 2: Figure 8 can be separated with 8(a) and 8(b) where the size of triangle of 8(a) can be slightly increased

Response 2:

Thanks for your reminder. We have separated Figure 8 into 8(a) and 8(b). The size of the triangle of 8(a) has been increased to the same size in 8(b).

Point 3: Font size of figure 15 (a-e) can be increased

Response 3:

Thanks for your suggestions. We have changed the word settings so that the ppi of the inserted image is higher, and the font size of figure 15(a-d) has been increased according to your comments.

Point 4: The test depicted in Figure 16 can be slightly introduced earlier in 4.3 rather than a sudden appearance in the last

Response 4:

Thanks for your suggestions. The content of Figure 15(e) is duplicated in Figure 16 after checking the manuscript. We removed the duplicate content and introduced the environment and scenario of the test in the earlier section 4.3 in lines 486-489 to make the overall content more coherent.

Author Response File: Author Response.pdf