Application Of The Improved Momentum VLBP Neural Network To Novel Defected Ground Structures

Artificial Neural Network (ANN) which simulates human cerebra is complex non-linear system consists of a great of nerve cells. The training results can be obtained correctly and quickly from the model which has been trained successfully within the range of training. The BP algorithm prevalent applied is inefficient, so it should be improved.Defected ground structures (DGS) are expanded from the photonic bandgap (PBG) structures. On the microwave circuit's ground metallic plane, the defected units are etched artificially in order to change the ground current distribution and influence the frequency properties of the transmission lines. The structures of DGS are generally calculated using electromagnetic field numerical value analysis methods such as finite-difference time-domain (FDTD) method. These methods can precisely analyze the frequency properties of DGS, but they are computationally expensive.In this paper, an improved artificial neural network model of lowpass filter with combinatorial nonperiodic defected ground structures (CNPDGS) and one of a novel CNPDGS with the characteristic of double stop band are developed. The momentum VLBP algorithm used in the two models integrates the momentum with the variable learning rate backpropagation (VLBP), updates the weight and the bias at each sample point, optimizes the training samples before training and is accomplished by C++ language because most of the core of artificial neural network algorithms in the world is accomplished by C, C++ language for more efficiency. The structure size of CNPDGS and the frequency are defined as the input samples of the ANN model, the parameters of transmission coefficient are defined as the output samples. Within the range of training, the parameters of transmission coefficient can be obtained correctly and quickly from the model which has been trained successfully. The result indicates that the momentum VLBP algorithm is more timesaving than FDTD and more efficient than the basic BP algorithm.