A CAD APPROACH BASED ON ARTIFICIAL NEURAL NETWORKS FOR CONDUCTOR- BACKED EDGE COUPLED COPLANAR WAVEGUIDES

摘要

Conventional edge-coupled coplanar waveguide structure was proposed in 1970 to implement a CPW directional coupler. However, the coupling effect is relatively weak due to its edge coupled configuration. To increase the coupling effect, an edge coupled coplanar waveguide structure without the central ground plane was proposed [3]. The coupling coefficient of edge - coupled CPW structure can be enhanced by adding an extra floating conductor on another side of the substrate [2]. The coupling phenomenon associated with this conductor - backed edge coupled CPW structure was characterized by the conventional spectral - domain approach, but the full-wave approach needs a lot of computer resources and the full-wave results were still lack of validation. For design purpose, a fast computational tool is also required to calculate the characteristic impedances of transmission lines, including the CPW structures with or without the backed conductor. In this paper, a new conformal mapping quasi static approximation method based on ANNs is used to calculate accurately the odd-and even-mode characteristic impedances, coupling coefficient and effective permittivities of CB-ECCPWs. ANNs have been recently recognized as a fast and flexible tool for RF and microwave modeling, analysis and design. ANN models are developed from measured or simulated microwave data training process. Resulting ANN models are used in place of CPU- intensive theoretical for fast accurate microwave design, analysis and optimization. The ANN employed in this paper is the MLPNN. Four learning algorithms BR, LM, QN and SCG are used to train the MLPNNs. These learning algorithms are employed to obtain better performance and faster convergence with simpler structure.