Neural network optimization of complex microwave structures with a reduced number of full‐wave analyses

An innovative technique of neural network optimization of substantially less computational cost is presented as a procedure suitable for viable computer‐aided design of complex microwave systems. The number of necessary full‐wave simulations in this optimization procedure is dramatically reduced because of a special objective function (OF) and constrained optimization response surface (CORS) sampling in the dynamic training of the decomposed radial‐basis‐function network. A high rate of convergence to an optimization objective is shown to be dependent on the combined effect of the OF and CORS technique. Performance of the algorithm is illustrated by its application to a waveguide band‐pass filter, a loaded microwave oven, and a dielectric resonator antenna; their optimal designs are found from five‐parameter optimizations requiring as little as 167, 177, and 99 analyses, respectively. © 2011 Wiley Periodicals, Inc. Int J RF and Microwave CAE, 2011.     

High-efficiency 3.4–4.4 μm light-emitting diodes based on a p-AlGaAsSb/n-InGaAsSb/n-AlGaAsSb heterostructure operating at room temperature

Light-emitting diode structures operating at room temperature were obtained based on a p-AlGaAsSb/n-InGaAsSb/n-AlGaAsSb heterostructure with high Al content in the boundary layers formed on a p-GaSb(100) substrate. This structure ensures a threefold increase in the output radiant power and the external quantum yield (～1%) as compared to the known InAsSb/InAsSbP heterostructure grown on an InAs substrate. A considerable increase in the pulsed output radiant power is explained by a more effective confinement of nonequilibrium charge carriers in the active region and by a decrease in the nonradiative recombination level, which is achieved by creating an isoperiodic structure.