In PMU calibration, response time and delay time are measured by feeding to the PMU an AC signal with a step change in magnitude or phase. Estimating accurately the reference step location is an important task in this process. In a paper presented at the 2019 IEEE AMPS, we proposed a Hybrid Step Location Estimator (HSLE), which is based on the Hilbert's transform (HT) analysis of the AC signal to obtain the related instantaneous magnitude and phase, upon which the estimation is carried out. We found that the performance of the HSLE decreased with the SNR dB of the AC signal. To render the HSLE more robust to noise, in this paper, we propose using the so called Polynomial Approximation Total Variation (PATV) decomposition to extract a noiseless step component from the instantaneous functions. The modified estimator (HSLE_PATV) is evaluated against the HSLE. Experimental results show that, for SNR dB > 55 dB, the HSLE_PATV and the HSLE perform similarly. For SNR dB E {30,40} dB, the HSLE_PATV outperforms the HSLE, providing more accurate step location estimates, with higher confidence levels, but at a higher computational cost.