In this article, a high-accuracy blood glucose detection sensor, consisting of a novel tunable bandpass filter (TBPF) and artificial neural networks (ANNs), is presented. The performance of the proposed sensor is based on the changes in the transmission pole (TP) of TBPF for four different dc voltage (VDC) values. For each VDC, a certain value of

${S}_{{11}}$

and resonance frequency are obtained for each blood plasma sample. Next, the data are processed by a new ANN model based on a multilayer perceptron (MLP) and radial basis function (RBF), which is performed using 177 different data points, and the value of blood glucose is predicted with high accuracy. The TBPF is designed using low-impedance resonators and the coupling effects, as well as a lumped elements circuit. The TP of the proposed filter, which also defines the filter’s passband, can be controlled by a lumped elements circuit and covers the frequency range of 1.4–2.2 GHz. The input variables comprise eight features from the measurements in real and are the

${S}$

-parameter attenuations in dB and resonant frequencies at four applied voltages. Furthermore, the output variable is the glucose level of the blood. The maximum values of the mean relative error percentage for the training data for MLP and RBF algorithms are 0.66% and 0.32%, respectively. The sensor has a compact size of

$17.9\times12.3$

mm2 and a high sensitivity of 375 kHz/mgdL.

Saeed Mehdipourbashi

Publications