Recent observation reveals a stunning fact that the coastal tides are experiencing a rapid change in the last century in several places in the world. To achieve a wide and refined understanding of the phenomenon, high-accuracy tide level data is needed more than ever. In-situ measurements – the traditional and main data source to support tidal harmonic analysis – are often sparse and limited to fixed positions, insufficient to provide information about the spatiotemporal variability of tidal processes beyond the tidal gauges. Satellite altimetry may fundamentally change the situation. This technology measures water level with increased spatial coverage and resolutions. However, satellite altimetry has not been used in tidal analysis due to two major limitations in the harmonic analysis: a) a minimum length of sampled observed data is required to recognize a sufficient number of tidal constituents according to the Rayleigh criterion and b) data sampling/acquisition frequency must be at least two times the major tidal frequencies to avoid the aliasing issue dictated by Nyquist theorem. To address these issues, a novel compressed-sensing approach is proposed to break the limitations. In this method, the prior information of the regional tides (e.g., a reference tidal station near the location of interest) is collected to support a stepwise optimization process to obtain the amplitudes and phase terms of the tide signal for data series with different lengths and time intervals. Instead of least-square-fit approach, stochastic gradient decent is employed using Pytorch. A preliminary study shows that the proposed method can generate the tidal amplitudes and phases with a sampling interval of 16 days and a total data length of 30 days with an acceptable error. The results of this study can be useful to determine an optimum frequency and length for tidal data acquisition for the upcoming SWOT (Surface Water Ocean Topography) satellite, which is supposed to be launched in November 2022 to measure sea and terrestrial water level around the globe for three years and with average revisit time of 11 days.