MRCS-Net: Multi-Radar Clustering Segmentation Networks for Full-Pulse Sequences

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1.      The grammar of the paper needs to be further checked, for example, the first letter of      "the" in line 58 needs to be capitalized, line 91 does not need to be indented.
2. In the abstract and introduction, the innovation and improvement of the work in this paper relative to the existing work need to be further elaborated.
3. There have been considerable research achievements in both SincNet and LSTM networks in radar pulse identification. Please further explain how the network structure proposed in this paper is improved and promoted on the basis of the existing work.
4. The influence of network parameters and pulse parameters on the classification and recognition performance of the proposed method should be further explained in the simulation experiment.
5. Please further explain the calculation amount of the proposed method compared with the existing method.

Comments on the Quality of English Language

The English writing of the paper is basically standardized and smooth, and some small grammatical mistakes need to be improved

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This article proposes a full pulse clustering segmentation algorithm, which has good performance under the condition of slight overlap of radar signals and has certain research significance. However, there are still the following problems that need to be improved.

1.The segmentation method in the article is not clearly stated, and it is not explained what the "feature window of the customization step" is. It is suggested to supplement and improve it. (Line 55)

2.Lines 60-62 and 68-70 of the article are highly similar, it is recommended to keep one of them.

3.The article only considers the segmentation problem when there is slight overlap, but in practical scenarios, there may be a large range of temporal overlap. It is recommended to increase relevant experiments to further verify the effectiveness of the proposed method.

4.The algorithm flow section mentions that "the processed data is a one-dimensional sampling sequence of unknown length, segmented and input into SincNet, and then input into long short-term memory networks in chronological order, and then analyzed using one-dimensional convolutional neural networks." It is not clear how the time sequence segmentation is performed, and it is recommended to supplement and improve it.

5.The dataset is not clearly stated, and only "a large amount of data" is mentioned in the article. Additional information is needed, such as the number of data samples used, the ratio of training and testing sets, etc. It is recommended to supplement and improve the information.

6.Adjust the font size of the text in Figure1 to be clearly visible. Moreover, the clarity of the figures in the experimental simulation can be further adjusted.

7.The input section in Figure1 contains a variety of targets such as airplanes and ships, is the method proposed in the manuscript applicable to all targets?Is the pre-processing process of signals from different targets the same?

8.What is the signal-to-noise ratio of the method proposed in the manuscript?Are there any differences in the results of the proposed method under different signal-to-noise ratios?

Comments on the Quality of English Language

I think the quality of language can be better! 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The core innovation of this article lies in proposing a deep learning based full pulse signal clustering and segmentation framework MRCS Net, which combines the signal filtering ability of SincNet, the sequence analysis ability of LSTM, and the feature extraction ability of 1-DCNN. It can effectively cluster and segment long duration pulse sequences, especially in complex electromagnetic environments, with high accuracy and real-time performance. My comments are as follows.

1. Please provide the computational complexity analysis. The article does not provide a detailed analysis of the computational complexity and real-time performance of the proposed algorithm. Although experiments demonstrate the algorithm's superiority, there is no explicit discussion of its computational efficiency and real-time processing capabilities on different scales of datasets, which is crucial for algorithm selection and optimization in practical applications.
2. Could you provide more diversity of experimental data? The article uses six different types of radar signals for experiments but does not mention whether real-world datasets were used or whether more types of radar signals were considered. This may affect the algorithm's generalizability and robustness.
3.  Could you provide more algorithm comparison? The article compares the proposed algorithm with single LSTM and 1-DCNN but does not provide a detailed comparison with other advanced radar signal processing algorithms. This limits the comprehensive validation of the algorithm's superiority.

Author Response

Please see the attachment.

Author Response File: Author Response.docx