The real-time broadband properties of impedance in power line communication (PLC) systems are one of the essential characteristics of future smart grids, which would enable the smart online systems to implement fault detection/forecast or the PLC channel health monitoring in the grid. In the current paper, a novel technique was proposed to track impedance only by the channel frequency response (CFR). The CFR can be treated as a known quantity and is normally calculated by the channel estimation algorithms in PLC devices for communication purposes. The relationship between CFR and impedance behavior was first studied in detail, and it was found that the variations in certain key factors, such as the frequency characteristics and the values of peak-valley difference, of the CFR curves could be used to the track real-time impedance. Then, the proposed impedance estimation algorithm harnessed the variational mode decomposition (VMD) as a feature extraction method to obtain useful frequency properties. The machine learning (ML)-based impedance model was also synthesized in the proposed approach. The performance of the proposed impedance tracking method was examined under two different scenarios, and the obtained simulation results demonstrated the efficiencies of the formulated algorithms.