Articles
Antenna Optimal Design Incorporating With Neural Networks And FDTD
July 13, 2009
•White Paper: Antenna Optimal Design Incorporating With Neural Networks And FDTD
By Remcom, Inc.
Particle swarm optimization (PSO) technique was recently introduced to EM community and was extensively investigated [1][2]. It is favored due to its simplicity, effectiveness and robustness. It is based on the velocity of the particles rather than the selection, crossover and mutation operations used in genetic algorithms (GA) [3]. When the PSO or GA is combined with the forward methods such as the finite difference time domain (FDTD) for antenna designs, it may require hundreds or thousands of FDTD simulations. A fast forward solution is therefore critical for the success of these optimization techniques.
However, a regular FDTD simulation for a complex structure typically takes more than one hour. To speed up the forward solution, various hardware or software techniques such as micro clusters or GPU hardware accelerator could be used. The good news is that a relatively imperfectly accurate FDTD simulation can yield results that are sufficient to determine the relative performance of similar antennas and hence the design of optimal antennas [4]. This observation not only allows the FDTD simulation to be speeded up by running for a minimum number of time steps, using a minimal FDTD grid and storing only necessary information [4], but also makes the artificial neural networks (ANN) [5] a good choice for a quick forward solution.
In this paper, we implement the PSO or GAs combined with the ANNs and Remcom's XFdtd® [6]. The combined technique is used for various antenna optimization examples and the results show that this combined technique is effective and efficient, and it can also be applied to other microwave circuit designs.
Click Here To Download:•White Paper: Antenna Optimal Design Incorporating With Neural Networks And FDTD
