The LOS/NLOS Classification Method Based on Deep Learning for the UWB Localization System in Coal Mines

Round 1

Reviewer 1 Report

 

1.       Introduction does not state the goal of the manuscript, and does not give any idea of what the contribution of the paper is. Authors present many state-of-the-art papers, but not discuss what are the advantages of the proposed paper. I would encourage the authors to improve the introduction. Perhaps, sections 1 and 2 can be merged.

2.       It was hard to me to understand what diagnostic data physically and how can they be used for the LOS/NLOS classification. The properties of these signals/data should be discussed better.

3.       I personally liked the nice explanation of GAN, CNN and trilateral centroid, but I would make the authors noticing that these are state-of-the-art algorithms and maybe does not need so much details. Please, consider reducing the theoretical background part to make up space for other possible clarifications.

4.       Please expand  “\epsilon_T is the error involved in this process caused by physical measurement of d”: what error is modelling this variable?

5.       It is not clear how the Kalman filter is exploited to remove the DME. I would suggest the authors to expand this part.

6.       Table 4 and 5: why are there two values in each cell? Could please expand the table legend?

 

8.       English sentence structures are often not properly correct and need to be revised. In general, the English of the manuscript needs to be thoroughly revised.

Author Response

1. We have made many improvements to the introduction in our latest manuscript. It not only presents the existing problems, but also introduces the contribution of our paper.

2. According to DW1000 user manual, these diagnostic data are stored in registers of DW1000 chip, which are the parameters used to calculate the First Path Power Level and the receive power level. When NLOS occurs, the First Path Power Level and the receive power level will be affected, and these diagnostic data will change and therefore be used to classify LOS/NLOS. We have added a discussion of these figures to our latest manuscript.

3. In the latest manuscript, we have reduced some explanations of GAN, CNN and Trilateral Centroid Positioning Algorithm. Then we have reinterpreted some of the theory and reduced some of the details.

4. We have added an explanation for this error in our latest manuscript. It is the error caused by reading or operating the instrument when measured by personnel.

5. In our latest manuscript we have added a paragraph explaining what the Kalman filter does and what errors it eliminates. It is used to eliminate fluctuations in the errors generated by our device and the errors that follow normal distribution during measurement.

6. Table 4 shows the performance of MLP model and Table 5 shows the performance of CNN model. These include the use of the original data set and the new data set containing GAN-generated data respectively. When using a certain data set, function classifiation_report () can output the model's precision, recall, f1-score and support sample number for NLOS and LOS classes. So there are two values in each cell. We explain the data in our latest manuscript.

7.  
8. We have made some English modifications in the latest manuscript.

Reviewer 2 Report

The paper looks fine and has a merit. Some comments to be addressed are as follows:

1. The authors need to highlight their novelty contribution. 

2. Review all the formulas and explain all used symbols in advance

4. The introduction section should be extended.
In the introduction section, the authors should focus on the main issues here. Background, the problem statement, the motivations behind the work and its context, the main contributions and the outlines of the paper.

5. The related work should be written as section 2 after the introduction with more recent references to highlight the contribution of this paper.
6. The proposed method is inadequately described. Better start by providing the reader with a high level picture of the problem.

7. There is no analysis of the extracted results and no discussion.

8. References are very limited

Author Response

1. We have added our contribution to the latest manuscript.

2. We have reviewed and examined all the formulas and explained the symbols used.

3.  

4. In the latest manuscript, we have made a lot of changes to the introduction, adding the existing problems in the current research, the contribution of our paper and the outline of our paper.

5. In the latest manuscript, we have moved the related work described in section 1 to Section 2, and increased the number of references.

6. In the introduction, we introduce the UWB system, describe the problems existing in the current research, and then put forward the contribution of our paper.

7. We have added our analysis and discussion of the experimental results in section 6 of the latest manuscript.

8. We have added the number of references in the latest manuscript.

Round 2

Reviewer 1 Report

I am satisfied with the manuscript modifications. I suggest accepting the manuscript in this current form.

Reviewer 2 Report

The authors addressed all comments.

The paper is ready to be accepted.