Specific Emitter Identification Based on Ensemble Neural Network and Signal Graph

Round 1

Reviewer 1 Report

In this paper an Specific Emitter Identification Based on Ensemble Neural Network and Signal Graph has been developed. The paper is well written and for it to be accepted, certain issues have to be addressed as follows.

1) What is the novelty of this work. The authors did not clear state why they use ensemble neural network in comparison to the other machine learning techniques. There are better techniques like probabilistic neural networks and support vector machines. An ensemble neural network is a deep learning technique that can classify complex dataset and requires large dataset for better accuracy. Why is deep learning compared with unsupervised machine learning techniques like KNN. The comparison should be with either another neural network or a deep learning technique.

2) There is no critical evaluation of the literature in the introduction. What is the rationale behind the work and the gap in research that the work is addressing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper “Specific Emitter Identification Based on Ensemble Neural Network and Signal Graph” needs careful improvement in order to be published.

1. References must be numbered in the order of the citations and placed before the full stop.
2. The formulation between lines 27-30 is not clear and it is not clear presented what the two categories
3. What is IQ imbalance.
4. Line 75 “Through a number of ablation studies’’. Through which studies?
5. Line 100: “graph convolution formula was improved in this study”. Please integrate information regarding how it was improved.
6. Please explain all terms from eq.3.
7. Sequence feature extractor, Graph feature extractor, Feature fusion module and Classifier are more generally explained. Please add important technical information.
8. Line 199 “1,300 signal”. What is “1,300 signal”?
9. Lines 200-201. Is it correct formulation?
10. Methods used for comparison are incomplete presented. As example:

• Authors never mention which was the values of k for the result of k-neighbors classification algorithm.
• Which was the percentage of data used for training the k-neighbors algorithm
• How many hidden layers are considered? How was optimized the number of neurons in the hidden layer?

11. How is the capacity of a proposed model affected by underfitting and overfitting

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

No further comments