Performance of WLAN in Downlink MU-MIMO Channel with the Least Cost in Terms of Increased Delay
Round 1
Reviewer 1 Report
This paper is interesting, it deals about the frame size optimization using a machine learning approach, I can see little simulation work due the authors use three traffic models and they study the effect of few variables, the paper contains issues with English, I am indicating some of them. I recommend the publication of the paper after minor corrections, should the authors correct the following details.
In abstract, what is AP?, STA?, the meaning of all the acronyms should be given in a table.
Page 2.- “and wireless users interact with wireless communication systems are significantly increased”….please check redaction, probably “interaction” instead of interact.
Page 3, line 122.- “Finally, the conclusion is given in Section 5.”…please correct…”Finally, the conclusions are given in Section 5”
Page 4, 137.- line “the throughput increases with increasing the frame size…”….Please correct….”the throughput increases with the frame size”
Page 5, line 170.- “the same frame length in all spatial streams could maximize the system throughput performance”……please correct…”the same frame length in all spatial streams that could maximize the system throughput performance”
Page 5, line 173.- “Some works in the literature have also studied focusing on the padding problem”….please correct….”Some works in the literature have also been studied focusing on the padding problema”
Page 5, line 188.- “none of them have proposed a machine-learning”….please correct….”none of them has proposed a machine-learning”
Page 7.-Line 283.- “machine learning model which is used to obtain the estimated gradient”…please correct….”machine learning model to estimate the gradient----"
Page 8, in equation 4, what is the meaning of vj?
Page 9-Figure 2, in one box it is written “knowledge (model) buildeing”….please revise if the word “buildeing” is correct. Probably “building”
Page 9, line 329.- “Then, Neural Network perform the training and adjusting the weight by employing the collected data set….”….please correct….”Then, Neural Network performs the training and adjusts the weight by employing the collected data set”
Page 10, figure 2 and legend must be in the same page
Page 11.- line 377.- “to examine how much is the Proposed ML Approach effectively optimized the frame size to the maximum system throughput comparably.”….please re – work this text, probably…..”to examine how much is the Proposed ML Approach effectively optimized with the frame size to the maximum system throughput comparably.”
Page 12, line 419.-“As the result shows in Figure 3, the proposed approach achieved the maximum performance in all traffic models.”….please re-work this text……probably…”As the results shown in Figure 3, the proposed approach achieved the maximum performance in all traffic models.”….which is the proposed approach?.....what is the maximum performance?
Page 12, line “The maximum performance of 820Mbps is achieved in 420 Weibull traffic model achieved the as compared to the Pareto and fBM…”….please re- work this text.
Page 12, line 421.- “Whereas the lowest performance is achieved in whereas the lowest performance of…”…..please re-work this text.
Page 12.- line 424.- “traffic which is less bursty as compared”….please check if the word “bursty” is correct in this context.
Page 14.- Legend of Figure 4, “Figure 4. Illustrates System throughput versus SNR when under different traffic models such as Pareto….”….please correct…..”Figure 4. Illustrates System throughput versus SNR for different traffic models such as Pareto…”
Page 12, line 443.- “However, the proposed approach achieved the maximum performance of 892Mbps the Weibull traffic model and…”…..please correct….”However, the proposed approach achieved the maximum performance of 892Mbps using the Weibull traffic model and…”
Page 13, line “by [16]. Table 3 illustrates quantitative comparative performance results of average…”…please correct….”by [16]. Table 3 illustrates quantitative performance results of average”
Page 16, line 544.- “affects the performance of the system throughput performance under the conditions of heterogenous…”….please correct…..”affects the performance of the system throughput behavior under the conditions of heterogenous”
Page 16.- Same observation, line 550.- “The results in Figure 6 (a), (b), and (c) demonstrate the performance of system throughput performance with increasing optimal frame size….”…please correct….”The results in Figure 6 (a), (b), and (c) demonstrate the performance of system throughput behavior with increasing optimal frame size….”
Page 17.- line “According to the results in Figure 6 (a), (b), and (c) show, the proposed ML….”…please correct….”According to the results in Figure 6 (a), (b), and (c), the proposed ML…”
Page 18.- Conclusions are just a summary of the paper. What do we deduct from the material presented in the paper?
Author Response
Response to Reviewer 1 Comments
Comment 1. In abstract, what is AP?, STA?, the meaning of all the acronyms should be given in a table.
Response: - In this approach, the Access Point (AP) performs the maximum Line18
Comment 2. Page 2.- “and wireless users interact with wireless communication systems are significantly increased” …. please check redaction, probably “interaction” instead of interact.
Response: - and wireless users’ interaction …...Line 62.
Comment 3. Page 3, line 122.- “Finally, the conclusion is given in Section 5.”…please correct…”Finally, the conclusions are given in Section 5”.
Response: - Finally, the conclusions are given in Section 5. Line 122.
Comment 4. Page 4, 137.- line “the throughput increases with increasing the frame size…” …. Please correct….”the throughput increases with the frame size”.
Response: According to the simulation results, the throughput increases with the frame size…Line 135.
Comment 5. Page 5, line 170.- “the same frame length in all spatial streams could maximize the system throughput performance” ……please correct…” the same frame length in all spatial streams that could maximize the system throughput performance”.
Response: - the same frame length in all spatial streams that could maximize the system throughput performance”. Line 168-169
Comment 6. Page 5, line 173.- “Some works in the literature have also studied focusing on the padding problem” …. please correct….”Some works in the literature have also been studied focusing on the padding problem”.
Response: - Some works in the literature have also been studied focusing on the padding problem
Comment 7. Page 5, line 188.- “none of them have proposed a machine-learning” …. please correct….”none of them has proposed a machine-learning”
Response: - none of them has proposed a machine-learning”
Comment 8. Page 7.-Line 283.- “machine learning model which is used to obtain the estimated gradient”…please correct….”machine learning model to estimate the gradient----"
Response: - machine learning model to estimate the gradient.
Comment 9. Page 8, in equation 4, what is the meaning of vj?
Responses: - Where in equation 4 is the weight sum of synaptic input plus bias of neuron j in layer l.
Comment 10. Page 9-Figure 2, in one box it is written “knowledge (model) buildeing”….please revise if the word “buildeing” is correct. Probably “building”.
Response: - Corrected as ''building''.
Comment 11. Page 9, line 329.- “Then, Neural Network perform the training and adjusting the weight by employing the collected data set….”….please correct….” Then, Neural Network performs the training and adjusts the weight by employing the collected data set”
Response: - Then, Neural Network performs the training and adjusts the weight by employing the collected data set.Line 311-312
Comment 12. Page 10, figure 2 and legend must be in the same page
Response: - Corrected.
Comment 13. Page 11.- line 377.- “to examine how much is the Proposed ML Approach effectively optimized the frame size to the maximum system throughput comparably.”….please re – work this text, probably…..”to examine how much is the Proposed ML Approach effectively optimized with the frame size to the maximum system throughput comparably.”
Response: - examine how much is the Proposed ML Approach effectively optimized with the frame size to the maximum system throughput comparably. Line 351-352.
Comment 14. Page 12, line 419.-“As the result shows in Figure 3, the proposed approach achieved the maximum performance in all traffic models.”….please re-work this text……probably…”As the results shown in Figure 3, the proposed approach achieved the maximum performance in all traffic models.”….which is the proposed approach?.....what is the maximum performance?
Response: - As the result shows in Figure 3, the proposed ML approach achieved the maximum performance in all traffic models.
Comment 15. Page 12, line “The maximum performance of 820Mbps is achieved in 420 Weibull traffic model achieved the as compared to the Pareto and fBM…”please re- work this text.
Response: - As the result shows in Figure 3, the proposed ML approach achieved the maximum performance in all traffic models. For instance, in the Weibull traffic model the maximum performance of 820Mbps is achieved as compared to the Pareto and fBM.
Comment 16. Page 12, line 421.- “Whereas the lowest performance is achieved in whereas the lowest performance of…” …...please re-work this text.
Response: - Whereas the lowest performance of 708Mbps is achieved using the Pareto traffic model. Line 379-380.
Comment 17. Page 12.- line 424.- “traffic which is less bursty as compared”….please check if the word “bursty” is correct in this context.
Response: - …Weibull traffic which is less bursty as compared to.
Comment 18. Page 14.- Legend of Figure 4, “Figure 4. Illustrates System throughput versus SNR when under different traffic models such as Pareto….”….please correct…..”Figure 4. Illustrates System throughput versus SNR for different traffic models such as Pareto…”
Response: - Figure 4. Illustrates System throughput versus SNR for different traffic models such as Pareto…’’
Comment 19. Page 12, line 443.- “However, the proposed approach achieved the maximum performance of 892Mbps the Weibull traffic model and…” …..please correct….”However, the proposed approach achieved the maximum performance of 892Mbps using the Weibull traffic model and…”
Response: - However, the proposed approach achieved the maximum performance of 892Mbps using the Weibull traffic model and the lower 732Mbps is achieved in the Pareto traffic model.
Comment 20. Page 13, line “by [16]. Table 3 illustrates quantitative comparative performance results of average…” …please correct….”by [16]. Table 3 illustrates quantitative performance results of average”.
Response: - Table 3 illustrates quantitative performance results of average…Line 411.
Comment 21. Page 16, line 544.- “affects the performance of the system throughput performance under the conditions of heterogenous…”….please correct…..”affects the performance of the system throughput behavior under the conditions of heterogenous”.
Response: - Affects the performance of the system throughput behavior under the conditions of heterogenous
Comment 22. Page 16.- Same observation, line 550.- “The results in Figure 6 (a), (b), and (c) demonstrate the performance of system throughput performance with increasing optimal frame size….”…please correct….” The results in Figure 6 (a), (b), and (c) demonstrate the performance of system throughput behavior with increasing optimal frame size….”
Response: - The results in Figure 6 (a), (b), and (c) demonstrate the performance of system throughput behavior with increasing optimal frame size.
Comment 23. Page 17.- line “According to the results in Figure 6 (a), (b), and (c) show, the proposed ML….”…please correct….” According to the results in Figure 6 (a), (b), and (c), the proposed ML…”
Response: - According to the results in Figure 6 (a), (b), and (c), the proposed ML…
Dear Reviewer, Thanks so much for your fruitful comments and suggestions. Kind Regards,
Reviewer 2 Report
The authors propose an approach using a machine learning to improve system performance such as throughput in downlink MU-MIMO WLAN environments. The approach uses the gradient information gathered from a multi-layer perceptron to adjust the aggregated frame size in downlink MU-MIMO.
However, there are several weak points:
1. Machine learnings require time to learn. Therefore, the time spent for learning and the errors occurred during the learning process should to be analyzed and included for fair performance comparison with other schemes especially in constantly changing wireless communication environments.
2. In the performance comparisons, authors did not compare performances with other schemes except for the base case (FIFO) and the throughputs from the authors’ previous works [16]. The performance of the proposed scheme should to be compared with performances of schemes from others (e.g., algorithms in the related works) for proper assessments.
3. The usefulness and contribution of the proposed scheme are not clear. In addition, the scheme depends on the authors’ previous work [16] because the throughputs obtained from the system of the previous work are needed to train the proposed neural network.
4. For the time-varying channel conditions, channel conditions should change over time. Just using fixed different SNRs are not enough to be considered as time-varying channel conditions.
5. There are lots of errors and typos. Some of them are:
- In Eq. (1), “max” should be “argmax”.
- In Eq. (5), notations are not consistent with Figure 1.
- In Line 96, "cannel" should be "channel".
- In Line 257, "or (batch" should be "(or batch".
Author Response
Response to Reviewer 2 Comments
- Comment 1. Machine learning requires time to learn. Therefore, the time spent for learning and the errors occurred during the learning process should be analyzed and included for performance comparison with other schemes especially in constantly changing wireless communication environment.
Response: - As we have tried to mention, to cope with the effects of time-varying channel conditions and heterogeneous traffic patterns, we have adopted online machine learning strategy as it is the feasible approach to achieve the data collection, knowledge building, and frame-size adjustment kept online. To achieve this, instant training data set is collected once every 50 seconds in considering different network scenarios such as channel conditions and traffic patterns and number of stations to train the neural network. Forward and backward passes are iteratively performed until the stopping criteria of Mean Square Error (MES) fall below 0.00001 or when the training epoch exceeds 1000 times according to our assumption in the experiment. The error threshold and the maximum number of iterations determine the accuracy of the function and the computing cost. When we compare the performance of the proposed ML approach with the maximum performance, we are intended to demonstrate how much the proposed ML approach can a performance compatibly with maximum throughput such that the cost of error in the prediction is lesser.
Comment 2. In the performance comparison, the authors did not compare performance with other schemes except for the base case FIFO and the throughput form authors’ previous works [16]. The performance of the proposed scheme should be compared with the performance of schemes from others (e.g., algorithms in related works) for proper assessments.
Response: - Because the objective of this study is to propose ML based adaptive approach mainly by extending the performance of our previous algorithm which was evaluated in comparing with the FIFO aggregation algorithm. But we will consider the compression with other approaches in future works.
Comments 3: The usefulness and contribution of the proposed scheme are not clear. In addition, the scheme depends on the authors’ previous work [16] because the throughputs obtained from the system of the previous work are needed to train the proposed neural network.
Response: - ML is an innovative approach that can autonomously extract patterns and predict trends based on environmental measurements and performance indicators as input. Moreover, that can respond the demand of today network performance by maintaining a self-driven networks that can configure and optimize themselves and reduces the drawbacks in using mathematical formulations and complex data analysis algorithms in the traditional approach and speed up the decision-making process, thus, in this regard, the proposed ML approach is significant to enhance the limitation of our previous work [16] which lack the benefits in adopting machine learning adaptive approach mentioned above.
We have made the CORRECTIONS ON THE PAPER TOO for this comment.
Comment 4: 4. For the time-varying channel conditions, channel conditions should change over time. Just using fixed different SNRs is not enough to be considered as time-varying channel conditions.
Response: - To examine the effect of time-varying channel conditions in WLAN and to study the effects we consider the lowest different SNR values for the lowest 3dB, 10dB and the maximum 20 dB in considering Additive White Gaussian Noise (AWGN) channel model. As a future work, we have a plan to consider different channel models such as Rayleigh and Rician as well and study the effects of different time-varying channel conditions.
Dear reviewer, Thanks so much for your fruitful comments and suggestions.
Kindest regards,
Reviewer 3 Report
Paper in subject of Electronix journal of MDPI. Numberand quality of references are acceptable
Article need review for improve its readability
I prepared my advice in some comments.
Comment 1
In introduction you wrote subsection
1.2 Machine Learning Adaptation in WLAN
Where is previous
1.1. Subsection ???
Comment 2
You wrote
Figure 4. Illustrates System throughput versus SNR when under different traffic models 493
such as Pareto, Weibull and fBM when Num STAs = 4 .
I propose more detail descript of tis figure
Figure 4. Illustrates System throughput versus SNR when under different traffic models 493
such as: a) Pareto, b) Weibull and c) fBM when Num STAs = 4 .
That same
Figure 5. Performance of system optimal frame size versus number of stations when the 537
channel condition is SNR =10dB in the Weibull, Pareto, and fBM traffic models.
I propose change on
Figure 5. Performance of system optimal frame size versus number of stations when the 537
channel condition is SNR =10dB in the: a) Weibull, b) Pareto, and c) fBM traffic models.
Comment 3
In line 353 need space “ “
The training data set is collected by adopting the simulation environment proposed by [16]as a pattern of “frame size - system throughput”.
See below
The training data set is collected by adopting the simulation environment proposed by [16] as a pattern of “frame size - system throughput”.
Comment 4
Please check size of font description under graph in Figure 4a.
“Throughput vs. SNR( Pareto Traffic) FIFO (Baseline Approach) Maximum Throughput
Proposed ML Approach”
look a small. Please use the same size for all figures the same as in Figure 5
Comment 5
You wrote
5. Conclusion and Future Works
To improve radability I propose use of bulleting. Also need to tick begining of Future works description
Author Response
Response to Reviewer 3 Comments
Comment 1. In introduction, you wrote subsection
1.2 Machine Learning Adaptation in WLAN
Where is previous
1.1. Subsection???
Response: - Correction made on the paper
Comment 2. You wrote
Figure 4. Illustrates System throughput versus SNR when under different traffic models 493
such as Pareto, Weibull and fBM when Num STAs = 4 .
I propose more detail descript of tis figure
Figure 4. Illustrates System throughput versus SNR when under different traffic models 493
such as: a) Pareto, b) Weibull and c) fBM when Num STAs = 4 .
That same
Figure 5. Performance of system optimal frame size versus number of stations when the 537
channel condition is SNR =10dB in the Weibull, Pareto, and fBM traffic models.
I propose change on
Figure 5. Performance of system optimal frame size versus number of stations when the 537
channel condition is SNR =10dB in the: a) Weibull, b) Pareto, and c) fBM traffic models.
Response: -Figure 4. Illustrates System throughput versus SNR for different traffic models such as Pareto, Weibull and fBM when NumSTAs= 4.
Figure 5. Performance of system throughput versus number of stations when SNR =10dB for different traffic models such as Weibull, Pareto, and fBM.
Comment 3
In line 353 need space “ “
The training data set is collected by adopting the simulation environment proposed by [16]as a pattern of “frame size - system throughput”.
See below
The training data set is collected by adopting the simulation environment proposed by [16] as a pattern of “frame size - system throughput”.
Response: Correction made on the manuscript
Comment 4. Please check size of font description under graph in Figure 4a.
“Throughput vs. SNR( Pareto Traffic) FIFO (Baseline Approach) Maximum Throughput
Proposed ML Approach”
look a small. Please use the same size for all figures the same as in Figure 5
Response: - Correction made on the manuscript
Comment 5 You wrote Conclusion and Future Works
To improve readability, I propose use of bulleting. Also need to tick beginning of Future works description
Response: We cannot use bullets, but we made the correction for the expression about the Future work in new paragraph.
Dear Reviewer, thanks so much for the comments and suggestions you gave us. Kindest Regards,
Reviewer 4 Report
1) Figures need re-formatting especially, figure 1 and figure 2 are blurred and squeezed.
2) Novelty of the research work is not clear. Authors need to explain clearly how this method is novel or better as compared to previously presented scenarios.
Author Response
Response to Reviewer 4 Comments
Comment 1. Figures need re-formatting especially, Figure 1 and Figure 2 are blurred and squeezed
Response: - Correction made on the manuscript.
Comment 2. Novelty of the research work is not clear. Authors need to explain how this method is novel or better compared to previously presented scenario.
Response: - ML is an innovative approach that can autonomously extract patterns and predict trends based on environmental measurements and performance indicators as input. Moreover, that can respond the demand of today network performance by maintaining a self-driven networks that can configure and optimize themselves and reduces the drawbacks in using mathematical formulations and complex data analysis algorithms in the traditional approach and speed up the decision-making process, thus, in this regard, the proposed ML approach is significant to enhance the limitation of our previous work [16] and also most of the existing approaches which have lacked the benefits in adopting machine learning adaptive approach.
Dear Reviewer, Thanks so much for the comments and suggestions.
Kindest regards,
Round 2
Reviewer 2 Report
- The latency of a scheme for the packet aggregation is perhaps more important than the accuracy of it especially in dynamically changing wireless networking environments. If a scheme makes a perfect decision after a few seconds, can it be said that it is better than a scheme that is less accurate but decides in milliseconds? The tradeoff between the accuracy and the latency is the reason why researches are needed in the environments.
- The proposed scheme depends on the output from the authors' previous scheme as a reference for the learning. Therefore, the time spent for the reference also should to be considered in the latency analysis. In addition, if the author's previous work could make perfect decisions, why is it needed to make another less accurate scheme? Just the use of a machine learning approach is not enough justification for it.
- The article did not provide comparisons of performances (including time delay) with schemes from other authors mentioned in the related works. Without it, it is not possible to assess the contribution or the usefulness of the proposed scheme if there is any.
Author Response
Comment 1- The latency of a scheme for the packet aggregation is perhaps more important than the accuracy of it especially in dynamically changing wireless networking environments. If a scheme makes a perfect decision after a few seconds, can it be said that it is better than a scheme that is less accurate but decides in milliseconds? The tradeoff between the accuracy and the latency is the reason why researches are needed in the environments.
Comment 2 - The proposed scheme depends on the output from the authors' previous scheme as a reference for the learning. Therefore, the time spent for the reference also should to be considered in the latency analysis. In addition, if the author's previous work could make perfect decisions, why is it needed to make another less accurate scheme? Just the use of a machine learning approach is not enough justification for it.
Comment 3- The article did not provide comparisons of performances (including time delay) with schemes from other authors mentioned in the related works. Without it, it is not possible to assess the contribution or the usefulness of the proposed scheme if there is any.
General Response to the comments
Dear reviewer,
Based on your supportive comments, we are forced to redo the work in a new form by changing the research approach including the topic. Because, according to the comments, the issue of delay was our future work which we planned to do in the future. Therefore, we revised the manuscript accordingly and tried to respond to your comments in considering the expense of latency.
Kindest Regards,
