Interference Estimation Using a Recurrent Neural Network Equalizer for Holographic Data Storage Systems

Round 1

Reviewer 1 Report

Ln.10 add ‘for short-term dependencies’ at the end of the sentences. The sentence is too promising as LSTM could be better suited for analysing longer.

Ln. 23 add the scientific fields that the suggested model can be used. This might enhance the readability of the paper.  

Ln. 44 indicate PR and ML as partial response and maximum likelihood in a parenthesis.  

Ln. 90 Fig 1 is the same as here:  10.3390/app12084070, ( cite this even though this is your paper ). On the other hand, Section 2 is quite close to your previous paper. Simplify the section as much as possible or merge it (after simplification) with the next section.   

Eq. 11: What is the alpha? Is it 0.85 or  explain it. 

Eq. 12: There might be an inconsistency between Figure 1 and this equation. check n value (-1) under the summation and the a (as -1, 1 )  in the Figure. 

Ln. 130 Explain why r = 30 and t = 20 are selected for each neuron. 

Ln. 139 Which value of BER indicates the best performance for the methodology? Explain in detail why you did not choose a value between  [10 20]  and [30 20] that seems to perform best.

Ln. 206 The line showing Best cannot be seen well in the plot.  Revise it. 

Ln. 209 See the comment above to fix the issue. 

Ln. 229 Highlight why any SNR value more than 15 is not available for the proposed model. 

Ln. 250 Indicate the computational power of the proposed model (time, source, etc.)

Ln. 258 I am concerned with deeper analysis of particular mechanisms, new proposals to try different methods, or simply curiosity. A future work properly guiding researchers to new research directions or ideas could be helpful in the conclusion section. 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper discusses a method for increasing the accuracy of data estimation when modeling a holographic memory system. The method is based on the use of a recurrent neural network and improves the method previously proposed by the authors themselves in [20]. The advantage of the method is that it reduces the data estimation error by 1dB compared to [20]. But the disadvantage of the method is that it is 6 times more complicated than the method [20] (Table 6). The work can be published after the authors take into account the comments.

Comments

1. The abbreviation PR is not defined, but only the abbreviation PRML is defined. This abbreviation must be defined separately.

2. The title of the second section includes the abbreviation GPR, which is defined through the indefinite abbreviation PR. It is better not to use abbreviations in section names.

3. The content of section 2 immediately begins with a description of the algorithm. Authors should add a paragraph of text with the physical statement of the problem at the beginning of this section or at the end of the Introduction. It should be explained why target is described by a 3x3 matrix, and equalizer is described by a 5x5 matrix?

4. Figure 4 shows the MSE graph, but the text does not provide the formula by which this value is calculated. An expression for MSE should be provided.

5. Line 210 states that blur is 1.85. Should I explain what the value 1.85 means?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf