Design of microstrip radio frequency bandpass filter based on improved BP algorithm

A neural network based method for the design of 1/4 wave length microstrip line radio frequency filter is proposed. The four key parameters of the microstrip filter: line longth L, line width W, slit S as well as frequency are used for the neural network inputs; and the reflection factor S₁₁ and port transmission factor S₂₁ and input terminal voltage standing-wave ratio vswr1 are used as the network outputs. In the paper, the BP algorithm is improved according the thought of using activation function gain to change gradient search direction, and the simulation example is also given. The neural network training results and filter simulation results using electromagnetism simulation software ADS are basically consistent, which shows that the proposed new method for microstrip radio frequency filter design is feasible.