Effects of Symmetry Restriction on the Antenna Gain Optimized Using Genetic Algorithms
Round 1
Reviewer 1 Report
The manuscript presents the GA-based optimization of gain for pixelated antennas. Also, the effect of symmetry restriction on the optimized topologies is investigated. The authors should address the following items.
1) Major concern: I am sorry, but I must say that I find it really hard to identify what the actual novel contribution of the paper is. The strategy is quite the standard procedure. The authors should clearly identify the new ideas and/or aspects of the presented strategy, and illustrate their convenience and significance.
2) The authors should clearly answer the question; what is the problem being addressed by the manuscript and why is it important to the antennas field?
3) Could the authors add some expression about the efficiency of the proposed method?
4) It is needed a better description of the results. Please avoid giving very obvious explanations.
5) Please highlight the advantages and drawbacks of the proposed method. Are there any drawbacks for the proposed filter?
6) Could the authors report the antenna polarization?
7) Could the authors report the 3D pattern of the antenna?
8) Could the authors report the surface current of the antenna at important frequency/frequencies?
9) The authors should report the gain and efficiency of the antenna in the frequency domain.
10) The authors should report the radiation pattern of the suggested antenna in H-plane and E-plane at important frequency/frequencies.
11) A summarized table in which the reported different works are compared with this proposed work is needed.
Author Response
Comment 1: Major concern: I am sorry, but I must say that I find it really hard to identify what the actual novel contribution of the paper is. The strategy is quite the standard procedure. The authors should clearly identify the new ideas and/or aspects of the presented strategy, and illustrate their convenience and significance
Comment 2: The authors should clearly answer the question; what is the problem being addressed by the manuscript and why is it important to the antennas field?
Response: We feel, that Comment 1 &2 are best answered with a single answer. We agree that our ideas were not clearly represented. We slightly re-arranged the introduction and added more text. We hope that we are now able to communicate the impact of our work, i.e. the first study of symmetry restrictions on the performance of pixelated antennas.
Comment 3: Could the authors add some expression about the efficiency of the proposed method?
Response: We are not sure if we understand the question correctly, but if the efficiency of the automated optimization method is meant, than no. We simply use the optimization technique to specifically design antennas of different symmetry.
Comment 4: It is needed a better description of the results. Please avoid giving very obvious explanations.
Response: Thank you for the suggestion! We added more discussion of the results in Section 4 “Discussion”.
Comment 5: Please highlight the advantages and drawbacks of the proposed method. Are there any drawbacks for the proposed filter?
Response: Again, we are not sure if we understand the question correctly. There is no mention of a proposed filter in the paper. If by proposed method, the reviewer means the employed optimization technique, the advantage is that the antenna structures can be optimized and simulated completely automated, thus cutting down on the required time. This was not an objective in this study, which focused on the study of symmetry restrictions on the gain and not the optimization itself. We added some text to make this more clear.
Comment 6: Could the authors report the antenna polarization?
Response: No – this was not investigated in detail.
Comment 7: Could the authors report the 3D pattern of the antenna?
Response: The simulated and measured 3D patterns are shown in Section 3.2
Comment 8: Could the authors report the surface current of the antenna at important frequency/frequencies?
Response: No, this was not investigated. In the past, we performed current density analysis of the generated antenna structures, to determine the influence “unconnected”, capacitively coupled, pixels. However, in the context of this study, no additional insight was expected.
Comment 9: The authors should report the gain and efficiency of the antenna in the frequency domain.
Response: We respectfully disagree. For antennas optimized specifically for a single resonance frequency (without optimizing the bandwidth), little to no additional information can be gained by plotting the gain/efficiency in the frequency domain.
Comment 10: The authors should report the radiation pattern of the suggested antenna in H-plane and E-plane at important frequency/frequencies.
Response: We appreciate the input however, the polarization of the antennas was not investigated, which is why we cannot make a statement about E-/H-planes. Therefore, we show the gain patters in the elevation (φ = 0° and 90°) and azimuth plane. As with Comment 9, we show the plots only for 868 MHz, because the antennas were optimized only for this resonance frequency. Additional gain patters, for different resonance frequencies will not yield any significant insight
Comment 11: A summarized table in which the reported different works are compared with this proposed work is needed.
Response: This is not possible, since this is – to the best of our knowledge – the first and only study to investigate the effects of symmetry restrictions on the gain of antennas, so there is nothing to compare them to.
Reviewer 2 Report
Authors in this research article have presented and investigated the Effects of Symmetry Restriction on the Antenna Gain Optimized Using Genetic Algorithms. The topic and concept of the paper are interesting and it includes promising results. Prior to the final acceptance recommendation, the authors are encouraged to address the following comments.
1. Its language needs some minor modifications.
2. Authors can use combination methods for optimization. For this, the following article can be used:
-“A New Design Approach of Low-Noise Stable Broadband Microwave Amplifier Using Hybrid Optimization Method”
3. Explain more about figure 4.
4. According to the title of the article (Antenna), the author can be discussing more references about the antennas. The following references may be helpful:
-“Design of a 1*4 Microstrip Antenna Array on the Human Thigh with Gain Enhancement”
- “HIGH GAIN AND WIDEBAND MULTI-STACK MULTILAYER ANISOTROPIC DIELECTRIC ANTENNA”
- “Substrate integrated waveguide leaky wave antenna with circular polarization and improvement of the scan angle”
Author Response
Comment 1: Its language needs some minor modifications.
Response: Thank you for the suggestion! We carefully proof-read the paper again and mended some errors.
Comment 2: Authors can use combination methods for optimization. For this, the following article can be used:
-“A New Design Approach of Low-Noise Stable Broadband Microwave Amplifier Using Hybrid Optimization Method”
Response: We appreciate the input – a hybrid optimization might be interesting to use. We updated the references and text accordingly.
Comment 3: Explain more about figure 4.
Response: We added more text to explain figure 4
Comment 4: According to the title of the article (Antenna), the author can be discussing more references about the antennas. The following references may be helpful:
-“Design of a 1*4 Microstrip Antenna Array on the Human Thigh with Gain Enhancement”
- “HIGH GAIN AND WIDEBAND MULTI-STACK MULTILAYER ANISOTROPIC DIELECTRIC ANTENNA”
- “Substrate integrated waveguide leaky wave antenna with circular polarization and improvement of the scan angle”
Response: While the suggested papers are clearly of high scientific value, they cover completely different topics than the submitted paper. Especially, the influence of symmetry on the gain is not examined at all in any of the suggested sources. Therefore, we feel that adding these references, does not improve our paper.
Reviewer 3 Report
The article is devoted to the issue of optimizing means of communication. The topic of the article is relevant. The structure of the article does not correspond to that adopted in the MDPI (Introduction, Models and Methods, Experiments, Discussion, Conclusions). The level of English is acceptable. The article is easy to read. The figures in the article are of acceptable quality. The article contains 28 sources, not all of which are relevant.
The following recommendations and comments can be made to the material of the article:
1. The authors say: "This study focuses on the optimization of the anteanna gain, while also investigating the effects of symmetry restrictions." At the same time, the structure of the antenna is determined by the object of its application and the physical parameters of the information signal. Methods for optimizing the structure of an antenna for the meter band will obviously differ from measures for optimizing the structure of antennas for the millimeter wave. Thus, the authors need to clearly position their research in the info-communication context.
2. Strictly speaking, genetic algorithms are not optimization methods. The result that the genetic algorithm produces is close to optimal. That is why genetic algorithms are used only when traditional optimization methods cannot be applied (a function with discontinuities, multi-extremal dependence, etc.) Just to take and apply a genetic algorithm means to spend a lot of computational resources to obtain a controversial result. The authors need to take this into account.
3. The authors sing odes to Matlab. As in the case of genetic algorithms, Matlab is good for its versatility, losing to specialized software packages in functionality and quality of the finished product. Authors should definitely pay attention to such products in the introduction.
4. The HFSS toolbox is a great thing. For example, having created an antenna model, we right-click on Optimetrics on the left in the project tree, then Add -> Optimization. It is necessary to choose an optimization algorithm (you should not choose a “quasi-Newtonian” algorithm, since this algorithm uses the gradient of the change in the S parameter, and it can fall into a local minimum), or you can choose, for example, a “genetic” algorithm. So the authors did. There is only one problem - where is the science here? Applying a ready-made software package according to the instructions is the task of a bachelor, not a scientist who wants to publish his article in Q2. I ask the authors to show that there is sufficient scientific novelty in their work.
5. In order to get closer to full antenna matching, you should do a parametric analysis (you can start by parameterizing the distance between the channels): right-click on Optimetrics -> Add -> Parametric, in the Sweep Definitions tab on the right, click Add, select the parameter YG -> Linear step and enter a range, for example, from 0.2 mm to 12 mm (the maximum value is chosen such that there is a distance to the edge of the board, say, 0.5 mm), in the Table tab there are all calculated values \u200b\u200b(it turned out to be 60), in the Options tab, check the box next to Save Fields and Mesh, this is necessary in order to later display many curves on one chart and select the appropriate one. We press OK. RMB for analysis -> Analyze.
6. When the best geometric design is chosen after parameterization, maximum results can be achieved. To do this, we will carry out one more optimization in the vicinity of the already obtained values of geometric parameters, i.e. it is necessary to reduce the range of parameters change in HFSSSDesign -> Design Properties for all changeable variables. We press the right mouse button on Optimetrics on the left in the project tree, then Add -> Optimization. You must select the Pattern Search optimization algorithm. Add the S(1,1) variable again as in the original optimization, now add a second variable by pressing Setup Calculation. And selecting Far Fields on the left in the Report type field, click on gain -> GainTotal in dB. Next, Add Calculation and enter in the Condition ">=" field, in the Goal "10" field, in the Weight "0" field, so that the first variable is more important in terms of weight, if agreement is more important to us than the gain.
Author Response
Comment 1: The authors say: "This study focuses on the optimization of the antenna gain, while also investigating the effects of symmetry restrictions." At the same time, the structure of the antenna is determined by the object of its application and the physical parameters of the information signal. Methods for optimizing the structure of an antenna for the meter band will obviously differ from measures for optimizing the structure of antennas for the millimeter wave. Thus, the authors need to clearly position their research in the info-communication context.
Response: Thank you for this input! Methods for optimizing mm-wave devices will clearly deviate from optimizing radio communication devices. We updated the text accordingly in the introduction and the discussion.
Comment 2: Strictly speaking, genetic algorithms are not optimization methods. The result that the genetic algorithm produces is close to optimal. That is why genetic algorithms are used only when traditional optimization methods cannot be applied (a function with discontinuities, multi-extremal dependence, etc.) Just to take and apply a genetic algorithm means to spend a lot of computational resources to obtain a controversial result. The authors need to take this into account.
Response: Thank you for this comment, however, we have to – respectfully – disagree. Genetic or evolutionary algorithms are optimization methods for global non-linear problems. They are meta-heuristic methods inspired by the process of natural selection and genetics that can be used to find approximate solutions to optimization problems, as you point out yourself. While the solution of an evolutionary optimization can be a local minimum and therefore not the “optimal solution”, this is also true for all other numeric multi-dimensional methods. Especially, in the context of antennas, the “best” solution is an unnecessary goal, as a sufficient solution is enough for producing well performing antennas. Just use the reflection coefficient S11 as an example: if -100dB is the global minimum, an achieved solution of -20dB, is definitely worse. However, -20dB are completely sufficient for most applications and going for -100dB would only increase the computation time unnecessarily. Therefore, we are taking this implicitly into account, when outlying the optimization goals.
Comment 3: The authors sing odes to Matlab. As in the case of genetic algorithms, Matlab is good for its versatility, losing to specialized software packages in functionality and quality of the finished product. Authors should definitely pay attention to such products in the introduction.
Response: Respectfully, we disagree. We just use Matlab as tool to implement the employed optimization scheme and stated this as a matter of fact in the paper, without any comments on our personal feelings about its usefulness. However, we agree with the general sentiment that Matlab is clearly not “the best” tool and it might prove interesting to compare different Algorithms and Tools and their influence on symmetry effects in future studies.
Comment 4: The HFSS toolbox is a great thing. For example, having created an antenna model, we right-click on Optimetrics on the left in the project tree, then Add -> Optimization. It is necessary to choose an optimization algorithm (you should not choose a “quasi-Newtonian” algorithm, since this algorithm uses the gradient of the change in the S parameter, and it can fall into a local minimum), or you can choose, for example, a “genetic” algorithm. So the authors did. There is only one problem - where is the science here? Applying a ready-made software package according to the instructions is the task of a bachelor, not a scientist who wants to publish his article in Q2. I ask the authors to show that there is sufficient scientific novelty in their work.
Comment 5: In order to get closer to full antenna matching, you should do a parametric analysis (you can start by parameterizing the distance between the channels): right-click on Optimetrics -> Add -> Parametric, in the Sweep Definitions tab on the right, click Add, select the parameter YG -> Linear step and enter a range, for example, from 0.2 mm to 12 mm (the maximum value is chosen such that there is a distance to the edge of the board, say, 0.5 mm), in the Table tab there are all calculated values \u200b\u200b(it turned out to be 60), in the Options tab, check the box next to Save Fields and Mesh, this is necessary in order to later display many curves on one chart and select the appropriate one. We press OK. RMB for analysis -> Analyze.
Comment 6: When the best geometric design is chosen after parameterization, maximum results can be achieved. To do this, we will carry out one more optimization in the vicinity of the already obtained values of geometric parameters, i.e. it is necessary to reduce the range of parameters change in HFSSSDesign -> Design Properties for all changeable variables. We press the right mouse button on Optimetrics on the left in the project tree, then Add -> Optimization. You must select the Pattern Search optimization algorithm. Add the S(1,1) variable again as in the original optimization, now add a second variable by pressing Setup Calculation. And selecting Far Fields on the left in the Report type field, click on gain -> GainTotal in dB. Next, Add Calculation and enter in the Condition ">=" field, in the Goal "10" field, in the Weight "0" field, so that the first variable is more important in terms of weight, if agreement is more important to us than the gain.
Response: We feel Comment 4 – 6 can be best answered with a single answer. We are grateful for these extensive comments, as it shows that the software aspect was severely misrepresented in our paper. We updated the text accordingly and hope that our approach is made more clear.
The reviewer is right – antenna optimization can be done quite nicely with the HFSS toolbox. However, in detailing the necessary steps, he highlights the advantage of our method. Using HFSS to optimize a single antenna, one must perform a lot of different tasks. Using our optimization scheme, this is done automatically, after entering some initial settings without the need for additional human input. In addition to that, we cannot perform a parameter sweep for a pixelated antenna, but need to employ a binary optimization procedure.
We would argue, that even using a ready-made software package to study a novel, or in this case neglected, aspect in the design and optimization of pixelated antennas, would justify a publication. There are many examples in scientific literature, where authors “only” used a ready-made software package to test out an idea.
However, the presented study goes far beyond that. It employs a custom-made optimization scheme that has been developed over the last years that uses a generic algorithm (implemented in Matlab) and a commercial software package (Sonnet) to simulate the optimized antenna structures. The optimization scheme is a complex and extensive piece of software that uses these tools and combines them for the completely automated optimization of pixelated antennas. That software has been used to optimize several antennas, whose performance was analyzed and compared. Based on these results, 4 antennas were manufactured and measured in the laboratory, to test the validity of the simulations.
Concerning the mismatch of some of the antennas: this was not a goal of this study and as outlined can be remedied with multi-objective optimization. However, it is interesting to compare the different optimization goals and their effect on S11 in addition to symmetry-restrictions on the gain.
Round 2
Reviewer 1 Report
Thanks the authors for answers and updates. The quality of the manuscript is improved. However, I think that my some question are not answered correctly.
1) Could the authors add some expression about the efficiency of the proposed method? Authors are expected to report the required time for the optimization process. Is it reasonable to use the proposed method for more complex antennas with larger dimensions? The proposed method will be faster or the optimization toolbox in full-wave simulator (e.g. HFSS, CST, etc)?
2) Could the authors report the antenna polarization? The authors can simply report the polarization of the proposed antenna by simulation. Please note that the polarization of an antenna is important feature of it.
3) The authors are expected to plot the surface current at the center frequency! The antenna polarization can be estimated using the surface current.
4) The authors are expected to plot the gain and efficiency of the antenna in the frequency domain. It is possible to comment on the performance of an antenna when its important features are available at a specified frequency interval. The return loss and antenna pattern are not sufficient for this purpose.
After making the above changes in their text, the manuscript can be accepted for publication.
Author Response
Comment 1:
We compared simulation times for antennas of different size in a previous study (10.3390/electronics9111856) - as can be seen, the optimization time varies, due to overall size and pixel size, with a decreasing simulation time for increasing antenna and pixel sizes. This is due to the fact, that the optimization goal can be found faster in a bigger solution space. The limits of this method will be tested in future studies, i.e. how small/how big the optimized topologies can be.
To fully optimize antennas of the sizes shown in the current publication 2-3 hours were needed.
However, a comparison to a full-wave simulation tool like Ansys or CST is not really possible in this case. Each optimization step in our procedure needs an external tool like Ansys or CST (or Sonnet), so per definition our method cannot be faster, but at the same time Ansys cannot perform a binary optimization for a pixelated structure.
I hope this answers your question - our method is more about finding the best geometry for pixelated antennas in an automated manner, which is not possible using only Ansys, CST or Sonnet as stand-alone simulators.
As far as we know, it might be possible to implement a similar procedure for pixelated antennas directly in CST using Visual Basic. However, we cannot comment on the efficiency of this, because there are no published results, to the best of our knowledge.
Comment 2:
We investigated the axial ratio of the antennas and conclude, that they are linearly polarized. Text has been added in the results section.
Comment 3:
We added surface current plots at 868MHz in the appendix.
Comment 4:
Additional simulations have been performed and plots have been added in the results section for the efficiency and gain as a function of the frequency.
Reviewer 2 Report
Authors have fully answered the referees, this article can be published in this journal
Author Response
Thank you very much for improving our article with your helpful comments
Reviewer 3 Report
I formulated the following recommendations for the basic version of the article:
1. The authors say: "This study focuses on the optimization of the anteanna gain, while also investigating the effects of symmetry restrictions." At the same time, the structure of the antenna is determined by the object of its application and the physical parameters of the information signal. Methods for optimizing the structure of an antenna for the meter band will obviously differ from measures for optimizing the structure of antennas for the millimeter wave. Thus, the authors need to clearly position their research in the info-communication context.
2. Strictly speaking, genetic algorithms are not optimization methods. The result that the genetic algorithm produces is close to optimal. That is why genetic algorithms are used only when traditional optimization methods cannot be applied (a function with discontinuities, multi-extremal dependence, etc.) Just to take and apply a genetic algorithm means to spend a lot of computational resources to obtain a controversial result. The authors need to take this into account.
3. The authors sing odes to Matlab. As in the case of genetic algorithms, Matlab is good for its versatility, losing to specialized software packages in functionality and quality of the finished product. Authors should definitely pay attention to such products in the introduction.
4. The HFSS toolbox is a great thing. For example, having created an antenna model, we right-click on Optimetrics on the left in the project tree, then Add -> Optimization. It is necessary to choose an optimization algorithm (you should not choose a “quasi-Newtonian” algorithm, since this algorithm uses the gradient of the change in the S parameter, and it can fall into a local minimum), or you can choose, for example, a “genetic” algorithm. So the authors did. There is only one problem - where is the science here? Applying a ready-made software package according to the instructions is the task of a bachelor, not a scientist who wants to publish his article in Q2. I ask the authors to show that there is sufficient scientific novelty in their work.
5. In order to get closer to full antenna matching, you should do a parametric analysis (you can start by parameterizing the distance between the channels): right-click on Optimetrics -> Add -> Parametric, in the Sweep Definitions tab on the right, click Add, select the parameter YG -> Linear step and enter a range, for example, from 0.2 mm to 12 mm (the maximum value is chosen such that there is a distance to the edge of the board, say, 0.5 mm), in the Table tab there are all calculated values \u200b\u200b(it turned out to be 60), in the Options tab, check the box next to Save Fields and Mesh, this is necessary in order to later display many curves on one chart and select the appropriate one. We press OK. RMB for analysis -> Analyze.
6. When the best geometric design is chosen after parameterization, maximum results can be achieved. To do this, we will carry out one more optimization in the vicinity of the already obtained values of geometric parameters, i.e. it is necessary to reduce the range of parameters change in HFSSSDesign -> Design Properties for all changeable variables. We press the right mouse button on Optimetrics on the left in the project tree, then Add -> Optimization. You must select the Pattern Search optimization algorithm. Add the S(1,1) variable again as in the original optimization, now add a second variable by pressing Setup Calculation. And selecting Far Fields on the left in the Report type field, click on gain -> GainTotal in dB. Next, Add Calculation and enter in the Condition ">=" field, in the Goal "10" field, in the Weight "0" field, so that the first variable is more important in terms of weight, if agreement is more important to us than the gain.
The authors answered all my questions. I liked the lively style of their responses. Despite the fact that the opinion of the authors differs from the opinion of the scientific community (for example, regarding the optimality of genetic algorithms according to Wikipedia), the authors express their point of view. This is a good trait for a scientist. I would like to enter into a debate with the authors and make another 5-6 iterations of the review, but I will not confuse my desire to communicate with the need to be an objective review. The article is good. I support its publication. I wish the authors creative success.
Author Response
Thank you very much for improving our article with your helpful comments!
Round 3
Reviewer 1 Report
Thanks the authors for answers and updates. The manuscript is now suitable for publication.
