Design of a Compact Power Splitter With Improved Performance for Wireless Applications Using Recurrent and Feed Forward Neural Networks Inverted Models

The power splitter, also known as power divider, is a microstrip component that typically has one input and two or more outputs. The initial design of Wilkinson power splitter for use in modern circuits faces challenges, such as large dimensions, harmonic generation issues, high overall cost, and limited bandwidth and frequency range. In this paper, artificial neural networks (ANNs) including feed forward neural network (FNN) and recurrent neural networks (RNN) inverted models are presented to design and optimize the performance of the resonators incorporated in the proposed power splitter. Innovative method of ANN inverted models is incorporated to ease the complex resonator design procedures and improve its performance. The designed device is analyzed, simulated, and fabricated, which the measured results have verified the simulation and analyses results. The proposed power splitter also utilizes coupling resonators, meander lines, and open stubs in main structure of the power splitter, achieving a wide bandwidth with fractional bandwidth (FBW) of 49%, effective harmonic suppression (removing second to fifth harmonics with values of 27.6 dB, 33.2 dB, 45.5 dB, 21.4 dB, respectively), and excellent miniaturization (65% smaller compared to the conventional model with dimensions of
$0.074~\lambda $
g
$\times 0.075~\lambda $
g
$= 0.00555~\lambda $
g2). Considering the main frequency of 1.4 GHz, the return losses for Port 1, Port 2, isolation, and insertion losses are obtained –17.8 dB, –22 dB, –20.3 dB, and –3.2 dB, respectively. Alongside the acceptable characteristic size, these features make it a promising design.