Novel neural network modeling method and applications

Abstract

Neural networks play an important role for designing the parametric model of electromagnetic structures. The current neural network methods are unfit for a circuit model with many input variables because it is costly to extract a large number of the training data and test data to complete the highly nonlinear mapping approximation. This article proposes a new neural network modeling method-the multidimensional neural network model, which can be used to solve the issue of multivariable radiofrequency and microwave passive device modeling. The entire multidimensional neural network modeling problem is simplified into a set of neural network submodels through decomposition method. Then the submodels are combined into an equivalent model, and the final entire model is produced through the neural-network mapping model developed with the submodels and equivalent model. A microstrip hairpin filter model is developed using the proposed method. The simulation results show the correctness and the effectivity of the proposed method.

Neural network of calibrated coarse model and application to substrate integrated waveguide filter design

“…In order to illustrate the advantages of the proposed model, we design the same SIW filter using a conventional dimensional synthesis method, which incorporates the equivalent deembedding technique with the inverse neural network and an optimization method based on equivalent deembedding and multidimensional neural network. 16 The results are summarized in Table 5, which demonstrates that the proposed method takes less time than the conventional synthesis method and the optimization method of multidimensional neural network. The geometric parameters synthesized by the inverse neural network can only be used as initial values, then we use the Quasi-Newton optimizer to optimize the return loss, which cost more than 10 hours.…”

“…However, it is difficult to train the neural network for a high-order filter because it corresponding to a high-dimensional space mapping relationship and a large amount of training data will be used. To solve this problem, the segmentation finite element (SFE) method 22 can be used to decompose the whole filter into substructures, which will reduce the dimension of the input-output relationship, and several examples of this method to decompose the filter are shown in References [2,4,16]. We assume that the input of a filter to be an m-dimensional vector as follows:…”

“…Neural models are developed for each module instead of the entire filter. 15,16 This method not only speeds up the generation of data but also reduces the size of the entire neural network, thereby obtaining higher accuracy. However, this neural network is still designed based on the purpose of optimization.…”

Modeling of heterojunction bipolar transistors based on novel Wiener‐type dynamic neural network

“…2 Furthermore, computer-aided design (CAD) is widely used in the development of microwave device models, and artificial neural network (ANN) technology is one of the important methods to address the modeling of microwave devices. [3][4][5] Nowadays, devices are operating at higher and higher frequencies, and heterojunction bipolar transistors (HBTs) have the advantages of short transition time, high-cutoff frequency, and large current gain. Therefore, HBTs have been more widely used, [6][7][8][9] showing good prospects for applications in large-scale integrated circuits, fiber optic communication, and power sensors.…”

CNN‐based politics public opinion analysis of undergraduates: A case study with CDN deployment

“…With the fast development of Neural Network (NN) technique, 1 some data analysis based on sample training has been addressed well. Different from others, Convolutional NN (CNN) 2 has an additional convolution layer and it leverages some convolution layers to enhance the performance of sample training and further improve the strong data analysis ability.…”