That one of the researchers from his point of view of his experience in the subject
Hi Daryl, I have used genetic algorithm to optimize the 4G LTE antennas for automotive applications. Just as an emphasis, you need to know which parameter optimization would drastically change your results (e.g. propagation pattern, gain,..). Since genetic algorithm sweeps parameters in a given interval, the optimization could be computationally complex and time-consuming. To be more specific about my project, I designed a PIFA antenna for IP (Instrument Panel) of a vehicle and noticed that it would be best if one knows the exact antenna parameters and their effects if they change in a specific antenna geometry and then he can apply e.g. Genetic Algorithm or Nelder-Mead,...
Keep in mind that based on the smartness of the genetic algorithm's cost function the algorithm could act smartly or otherwise, it could run for days without really doing a good job.
Side lobe level optimization in phased array antennas using genetic algorithm
Genetic algorithms in the design and optimization of antenna array patterns
Side Lobe Reduction of Phased Array Antenna using Genetic Algorithm and Particle Swarm Optimization
http://inside.mines.edu/~rhaupt/journals/ACES%20Nov%202006.pdf
few of the papers I found online.. hope this helps
Hi Daryl, yes evolutionary methods can perform very well in antenna applications, however as noted by some of the other responses, you have to be very careful in how you configure the decision vector structure and the objective functions. I have used EAs (i.e. not just Genetic Algorithms, but Evolutionary Programmes etc. too) for optimising array taper functions, and also for working out good partitions for sub-arrays. For large arrays with thousands of elements, using a decision vector structure that employs a mapping process is often very effective, rather than attempting to adjust each element individually. For example, if you wanted to generate an amplitude taper for an array with 4000 elements, you could either use a decision vector of 4000 real values (or binary equivalent if using a classic GA), or recognise that most good taper functions have a strong correlation in the weightings of neighbouring elements and therefore optimise a parametric function that is mapped to each element. For example using a basic 2D cubic spline would require 16 real values and each element would then be set to the value from the spline function at the element location. Using a spline will prevent sudden sharp transitions in the local amplitude taper, so higher order splines, or multiple splines concatenated may be needed if a large amount of variation is needed across the pattern profile. For the objective function, often multiple criteria are to be optimised, such as minimising sidelobes whilst maximising gain,so multi-objective algorithms can be very useful too. Evan
- Badges
- Science topic