Information about inverse modeling (neural networks)
I am currently working on the optimization of an UWB antenna using neural networks. I am trying to train a neural network to produce s11 samples at certain frequency intervals. This operation is called inverse modeling. So, does anyone have resources or papers about this subject
thanks in advance
Hi adel_48
did you see this book.
"Application of Neural Networks in Electromagnetics" by Christos Christodoulou & Michael Georgiopoulos
hope this help you.
regards
Try "Neural Networks for RF and Microwave Design" by Q. J. Zhang, K. C. Gupta.
well, frankly, I didnt undrestand your aim completely. you mean you have a certain UWB antenna (Say, a fisheye p4tch), your inputs is its physical parameters, and your output will be the S11 in certain frequensies? if it is the case, so your net has to learn a certain model (a certain antenna), and so you can use RBF or MPL. Otherwise, if it is more complicated than just this, and you want to take into account the modeling you already know about the antennas, or if it is so general; you have to do a lot of more, if this is the case, please reply, and I will guide more (and couse interesting, maybe do some cooperation).
have great time there,
.m
Hi mamali,
Here are my job exactly
1. I have a UWB antenna, that is giving good performance but requires more optimization.
2. The structure is electrically large, and has a lot of parameters (mainly 3 or 4 of primary effect), so EM simulation takes very long time.
3. Available optimization techniques are Genetic Algorithms, Neural Networks, Local Optimizers (which don't give good results)
4. Genetic algorithms will require very long time, specially for such highly dimensional problem. So I chose neural networks.
5. Instead of using the neural network as a EM simulation, I decided to train it to be a synthesis tool. i.e. its inputs are the S11 and its output are the physical dimensions. This is called inverse modeling
6. I am now having some problems in choosing the number of hidden neurons and choosing the sample space.
Best Regards,
Adel_48
For you, the best book is:
"Application of Neural Networks in Electromagnetics" by Christos Christodoulou & Michael Georgiopoulos" as mahdithdn has told. The IO nets, such as ARTMAP is best for this porpose as I know, couse you can feed it in one way, it learns, then you can use it in both ways, direct or reverse.
.m
Is this book available in ebook? If it is, I appreciate if you let me know the link.
Thanks,
Yid