Receiver function techniques are widely used in imaging crustal and mantle structure beneath a seismic station. The weak P-to-S conversions at deep seismic structures are usually masked by strong shallow reverberations when unconsolidated sediments are present below the station, making it nearly impossible to utilize receiver function techniques. We develop a method to estimate sediment and crustal structures beneath a seismic station based on wavefield downward continuation and decomposition method. The method parametrizes velocity structure beneath the station with a stack of constant velocity layers overlying a homogeneous half-space, and approximates the teleseismic P wave and its coda by the structural response to an incoming plane P wave. Our method is based on the principle that the upgoing S wavefield is absent in the half-space, and searches for the optimum velocity and thickness of the layers that give the minimum S-wave energy flux from the half-space to the layers. An iterative grid-search algorithm from the top to the bottom layers is employed to implement the search. In this study, we only use models comprising either only one crust layer or two layers (sediment+crust) with a half-space mantle, although models with more layers are also implementable. The method is especially useful in resolving seismic structure beneath a station sitting on unconsolidated sediments. It not only can be used to determine the sediment thickness and velocity structure, but also provides an effective way to generate subsurface receiver functions, which are formed by deconvolving the upgoing P wavefield from the upgoing S waves at the top of hardrock crust, and thus are free from shallow reverberations. The technique is applied to various synthetic data generated with different types of velocity model and noise levels, and appears to have good capability in recovering the input models. We further applied this method to teleseismic data recorded at a station inside the Songliao Basin in northeast China. The estimated sediment thickness and velocity agrees well with the results of previous activesource studies. The subsurface receiver functions also show a superior power in exposing the Moho Ps conversions, resulting in a well-defined peak in the H-? domain, which are absent in the regular receiver function data.