Nonuniform microstrip lines analysis using neural networks

A technique based on an artificial neural network is presented for determining the reflection and transmission characteristics of nonuniform microstrip lines. The width of the microstrip line is expressed as a truncated Fourier series, whose coefficients are combined with the analysis frequency and input to the neural network to determine the S₁₁ and S₂₁ parameters. A multilayer perceptron with two hidden layers and resilient error backpropagation training is used in this work. It was trained with 180 randomly generated microstrip lines whose S-parameters were determined by a different technique. Then, 60 randomly generated microstrip lines were used for validation. The obtained neural network results are in excellent agreement with those obtained by full-wave simulation.