MT method is widely used in exploration surveys worldwide but hardly applicable in urban areas because of the large amount of artificial electromagnetic noise. The MT time-series data at night is much quieter than in the daytime because most of the electric equipments are shut down for several hours during nighttime. Therefore, we focused on calculating the MT impedance using the quiet time-series data. In this research, the data observed by the Phoenix System was used. The data contain continuous and discontinuous time-series data. As an alternative way, we introduced a robust impedance estimator based on the Hilbert-Huang transform (RMHHT). We indicated that this technique needs 4-hour data to get a reliable impedance up to 1,000-second in the numerical simulation. This short measurement time makes it possible to carry out MT surveys in urban areas with strong noises. This paper demonstrated a strategy to process the broadband MT time-series data properly by the RMHHT estimator. Finally, a successful case study, using the nighttime data to get reliable impedance tensor in the areas contaminated by heavy noises, was demonstrated.