One of the essential elements of future battery management systems (BMS) is diagnostics. Electrochemical impedance spectroscopy (EIS) is competent in state estimations and analysis of lithium-ion batteries, but no online techniques are available for the BMS yet. Data-driven approaches are gaining prominence in battery applications where complex characteristics need to be modeled, but the burden of a massive amount of data makes it difficult to advance into deep learning. In this paper, a deep learning framework is proposed to estimate the online EIS of operating batteries. The framework consists of developing an empirical battery model and a neural network. The battery model accelerates the neural network training step by substituting the expensive and time-consuming battery experiment for data acquisition. The neural network estimates the EIS by processing the terminal voltage, current, and temperature data, which are typically measurable on BMS. The experimental validation on 50 Ah battery cells shows the EIS estimation performance with the root mean squared error of less than 45.8 µΩ While this paper validates the framework as a proof-of-concept in a reduced operating condition, it suggests a novel alternative solution for online EIS implementation.