Affiliations: [a] School of Electronics and Information Engineering,
Sichuan University, Chengdu, Sichuan, China | [b] Department of Biological and Enviroment Engeneering,
Cornell University, Ithaca, NY, USA
Correspondence:
[*]
Corresponding author: Yang Yang, School of Electronics and
Information Engineering, Sichuan University, Chengdu 610065, Sichuan, China.
Tel.: +86 28 854 087 79; Fax: +86 28 854 087 79; E-mail: [email protected]
Abstract: A two-port U-shaped Coplanar Waveguide (U-CPW) used for broadband
complex permittivity measurement of high loss solutions is presented. Compared
with the one-port measurement device, this two-port U-CPW coupled with
artificial neural networks algorithm has the ability of reducing the
multiple-solutions. Also, the measured T/R (transmission/reflection)
coefficients were used instead of simulated reflection coefficients for
training the neural networks that enhanced the accuracy of measurement results.
Two real-valued neural networks associated with the U-CPW are used to
reconstruct the two parts of the complex permittivity separately from the
measured T/R coefficients. First, a large number of measured T/R coefficients
of broadband frequency points of ethanol-water mixtures along with their
theoretical complex permittivity are used to train two separate neural
networks. Second, the trained networks are employed to reconstruct the complex
permittivity of other solutions using the measured T/R coefficients. Compared
with the results of Cole-Cole equation, the reconstructed results of
methanol-water mixtures and ethanol-methanol mixtures at different volume
fractions have been obtained with sufficient degree of accuracy in the
frequency range 1–3 GHz at 20°C. For predicted
values using the developed artificial neural network (ANN) architecture and
those obtained from literature, the average relative errors of real and
imaginary parts of complex permittivity are 2.27% and 2.12%, respectively.
