kumoh national institute of technology
Networked Systems Lab.

A Robust Automatic Wireless Signal Classification using Combined CNN and LSTM Algorithm in Multimedia IoT - ETFA 2020 Regular (R)
By :
Date : 2020-06-08
Views : 211

C. Clarity of presentation:
English grammar and spelling are proper ------------------------------------------ [1 - I disagree]
Mathematical symbols and equations are easy to understand ------------------------ [1 - I disagree]
Figures and tables are well constructed and informative -------------------------- [1 - I disagree]
The paper is well organized ------------------------------------------------------ [2 - I am neutral]
Considering the issues above, the paper is readable ------------------------------ [1 - I disagree]

T. Technical innovation and relevance
The authors cite other relevant publications ------------------------------------- [2 - I am neutral]
Authors describe relevance of work to the research field ------------------------- [3 - I agree]
The authors apply sound technical approaches ------------------------------------- [2 - I am neutral]
New ideas are convincingly and logically described ------------------------------- [2 - I am neutral]
Results are convincing ----------------------------------------------------------- [2 - I am neutral]
Considering the issues above, this work should be presented --------------------- [2 - I am neutral]

Comments:
The paper overall appears fairly incremental, adding yet another variation of the
applying-deep-learning-to-signal-classification literature. I am also not entirely
sure it is a good fit with the scope of ETFA.

The paper is not well written. While readable, the language is bumpy, with
several odd sentences. Mathematical symbols are not explained, e.g. around
Equations (3) to (7). Also the contents of Tables I and II are not explained.
Several acronyms are not introduced, e.g. ML-AMC.

The statistical significance of the results is unclear. Sometimes the differences
between the different algorithms are so small that it is uncertain whether the
differences really are significant.


C. Clarity of presentation:
English grammar and spelling are proper ------------------------------------------ [1 - I disagree]
Mathematical symbols and equations are easy to understand ------------------------ [2 - I am neutral]
Figures and tables are well constructed and informative -------------------------- [1 - I disagree]
The paper is well organized ------------------------------------------------------ [2 - I am neutral]
Considering the issues above, the paper is readable ------------------------------ [2 - I am neutral]

T. Technical innovation and relevance
The authors cite other relevant publications ------------------------------------- [2 - I am neutral]
Authors describe relevance of work to the research field ------------------------- [2 - I am neutral]
The authors apply sound technical approaches ------------------------------------- [1 - I disagree]
New ideas are convincingly and logically described ------------------------------- [1 - I disagree]
Results are convincing ----------------------------------------------------------- [1 - I disagree]
Considering the issues above, this work should be presented --------------------- [1 - I disagree]

Comments:
This paper presents a new configuration / combination of ML-
architecture for radio signal classification. I have some issues with
the paper.
- The choice of parameter configuration for the CNNs is quite ad
hoc. It would have been advisable to give an account of how you
have arrived at that configuration. Now it is just a random
configuration without arguments to its architecture.
- It is questionable why a comparison with VGG-like Net is relevant,
as this is mainly used for image recognition.
- Figure 6 is unclear, some better explanations of the confusion
matrix would have been helpful.
- For the computation time I am missing information about the
statistical reliability of your executions. Over how many executions is
the computation time averaged? As the colab-environment is quite
uncontrollable regarding allocation of GPUs, some explanatory
information about this would have been good.
- There are a number of problems with your formulations, probably
simply proficiency problems. But this leads to incorrect statements,
e.g. when describing the CNNs, and also the understanding is
hindered in places. I would recommend the input of a native speaker
or the assistance of a language expert to fine-tune the text.


C. Clarity of presentation:
English grammar and spelling are proper ------------------------------------------ [0 - I strongly disagree]
Mathematical symbols and equations are easy to understand ------------------------ [2 - I am neutral]
Figures and tables are well constructed and informative -------------------------- [3 - I agree]
The paper is well organized ------------------------------------------------------ [3 - I agree]
Considering the issues above, the paper is readable ------------------------------ [1 - I disagree]

T. Technical innovation and relevance
The authors cite other relevant publications ------------------------------------- [2 - I am neutral]
Authors describe relevance of work to the research field ------------------------- [3 - I agree]
The authors apply sound technical approaches ------------------------------------- [2 - I am neutral]
New ideas are convincingly and logically described ------------------------------- [2 - I am neutral]
Results are convincing ----------------------------------------------------------- [1 - I disagree]
Considering the issues above, this work should be presented --------------------- [2 - I am neutral]

Comments:
5G technologies and enabling methodologies are quite
compelling and relevant topics for factory communication
systems (FCS).

Unfortunately, there are far too many typos and grammar
errors throughout the manuscript, which actually makes it
not easy to read and follow. I also found the contribution
(an efficient/faster deep learning algorithm for
modulation classification) only marginally relevant for
the scope of the event; it is not clear the actual impact
of the marginally lower computation time (the results
shown in Figure 7 show only very marginal gains against
alternative solutions) at a hypothetical FCS level.