—In this work, a novel method for non-invasive, label-free, multi-variable volumetric concentration analysis of mixtures is presented based on wide-band microwave resonators. The proposed microwave structure is a resonator with multiple splits for improving the overall sensitivity. The wide-band analysis of the sensor’s spectrum provides many resonances which shift in any of them could be translated to the dielectric permittivity at that frequency. Therefore, by considering frequency shifts in as many as 30 resonance frequencies over a high-frequency span from 100 MHz to 6.5 GHz, the sensor provides invaluable information of the mixture components. This is because of the non-linear and unique variations of the dielectric permittivity of different materials resulting in a unique frequency-dependent shift in the resonance frequencies. Seven liquids soluble in each other are mixed with different percentages with a fixed overall mixture volume in this work. According to the extreme complexity of the problem, an artificial neural network is employed with the shifts in the mentioned resonance frequencies due to introducing each sample to the sensor are considered as the input features and the volumetric concentration of each of the seven components are calculated as the outputs. The network is trained with 750 samples and tested with the different 150 other samples with the overall average MSE of as small as 0.057%.
