Compensation of Misalignment Errors in Solid Dielectrics Characterization With Machine Learning-Enhanced Differential Microwave Sensor

This paper presents a new differential microwave sensor with machine learning algorithm for compensation of the measurement divergences caused by misalignment of the material under test (MUT). The deviation of MUT position on the sensitive area of sensor causes variations in transmission zero (TZ) and transmission pole (TP) values in frequency response of the sensor. This paper presents machine learning algorithm employed for material classifications using corresponding TZ and TP values. A regressor configuration based on frequency response of the proposed sensor has been developed to determine the deviation amount. Dual-mode split ring resonators connected to a microstrip line is proposed as a highly sensitive solution for solid dielectric characterization. Frequency response of the proposed sensor features two TZs and one TP. Establishing the differential frequency as the difference between the TZ and TP allows effective compensation against ambient conditions. The proposed differential architecture also offers compact structure using one single sensing element, reducing the footprint to the half of conventional differential sensors. Training the Bayesian Regularization algorithm on measured data shows 0.998 regression with 0.125 mean square error. The experimental results for the fabricated prototype demonstrate a normalized sensitivity of 5.97%, which is higher than that of existing structures.