Smith chart modeling by artificial neural networks

In microwave engineering design, many different circuit designing phase needs transmission line frequently, for example in input and output matching circuit of an amplifier, quarter-wave transfomer etc. In this work, MLP and RBF neural networks are used for the one-to-one mapping and universal approximators between the port impedances for analysis and synthesis of the transmission line segment, which is called as the ?unit element?. Impedance transformation from the input port to the output port in terms of the unit element parameters ? , Z0 is defined as the analysis problem which is assumed to be the problem in the forward direction. Synthesis of the impedance throughout all the operation bandwidth B with the circuit parameters ?, Z0 is defined as the problem in the reverse direction. With this pupose, the forward and reverse problems can be defined by means of the two blackboxes: (i) the forward problem: Black-box in analysis ; (ii) the reverse problem: Blackbox in synthesis. The manual analysis and design of microwave circuits are generally tedious and error prone. The Smith chart provides a very useful graphical tool to these problems. A great deal of knowledge can be acquired from a Smith chart e.g. standing wave ratio, single and double stub tunings and much more. Although Smith charts are valuable and contain significant amount of information, inaccurate observations can lead to erroneous results and frustration. In this work, it is aimed to achieve an ANN model of the Smith chart. In this model, the two bilinear transformations between the rectangular empedance-plane and polar reflection coefficient-plane are employed for the training data. In the current work, the feed-forward MLP type of neural network is utilized with four layers. The source impedance, physical length and the operation frequency are inputted to this network, while the output reflection coefficient, the output impedance, single-stub and double-stub matching parameters, standing wave positions are taken as the outputs. In the last stage, instead of dealing with differenting to the complex equations, moduler ANN model of the Smith Chart is used together with the numerical sensitivity analysis. It should be noted that all novel improved ANN models used in this work are verified to converge target values in higher and enough accuracy. Keywords: Distributed-parameter circuit analysis and synthesis, Smith chart, Single and double-stub impedance matching, Artificial neural networks, Sensitivity analysis, Blackbox model, Neural unit element, Neural smith chart.