The coastal plains of the central and eastern United States contain deep sections of unconsolidated to poorly consolidated sediments. These sediments mask deeper crustal and upper-mantle converted phases in teleseismic receiver functions through large amplitude, near-surface reverberations. Thick sediments also amplify ambient noise levels to generally reduce data signal-to-noise ratios. Removing shallow-sediment wave-propagation effects is critical for imaging deep lithospheric structures. A propagator matrix formalism is used to downward-continue the wave field for teleseismic P waves into the midcrust and then to separate the upgoing S-wave field from the total teleseismic response of the P wave to expose deep Sp conversions. This method requires that the Earth model from the surface to the reference depth be known. Teleseismic P-wave data for the Memphis, Tennessee, station (MPH) are analyzed using a reference-station deconvolution technique to produce vertical and radial P-wave transfer functions. These transfer functions are modeled using a simple model parameterization for sediment structure through grid inversion. The inverted Earth model is incorporated into the wave-field continuation and decomposition technique to estimate the upgoing S-wave field at 10 km depth in the crust. Moho and possible deeper Ps conversions are identified with this process.