Hybrid Models for Effective Design and Optimization of Large-Scale Multiplexing Networks

In this paper, hybrid models for the design and optimization of large-scale multiplexers are presented. First, analysis based on the low-pass prototype filter circuit model proves that the first cavity of a channel filter next to the manifold is critical in the design of multiplexers. A hybrid model is proposed, combining a circuit model and the electromagnetic (EM) analysis of the coupling junction close to the manifold. The fact that only a single junction requires EM simulation facilitates the introduction of neural networks to multiplexer design for the first time. Another format of the hybrid model is then built by replacing the EM simulation with a neural model for the output coupling to the first resonator. The efficiency of the method is demonstrated using a six-channel multiplexer and a 13-channel multiplexer. Responses predicted by the hybrid models agree well with full EM simulations and measurement results. The proposed hybrid models are proven to be able to provide a high degree of accuracy in both in-band and out-of-band frequency ranges and fast, enabling iterative optimization involving all design parameters for large-scale multiplexers.