Broadband seismic waveform data are stacked to investigate the mantle discontinuities beneath a station. A polarized filter is devised to remove pseudo-signals in the stacked traces, which might be otherwise misinterpreted as a discontinuity. The depth of a mantle discontinuity determined in previous studies depends on the reference model. We suggest the use of data sets which have a range of epicentral distances to the investigated station. The observed travel time of the P-to-S converted phases as a function of epicentral distance can be used to constrain the proper reference model. When the technique is applied to real data, we can determine the absolute depth of a discontinuity with an accuracy of approximately ±10 km. The method is applied to the broadband data of the CDSN stations. There is no significant depression observed for any of the stations except BJI, implying that the lateral scale of the trough in the '660-km' discontinuity under northeast China is smaller than suggested in previous SS precursors studies. Beneath station BJI, the '410-km' and '660-km' discontinuities are elevated 10 km and depressed 30 km, respectively, resulting in an extremely thick transition zone. This may be attributed to the cold pacific plate that exists in the transition zone of the same region. Meanwhile, at station MDJ, where the subducted pacific plate is also found in the mantle transition zone, a multiple-discontinuity structure is observed rather than a depressed '660-km' discontinuity. At station SSE, there is no depression of the '660-km' discontinuity, suggesting that there is no significant difference of temperature at depths around 660 km between SSE and the average mantle.