The 3D S-wave velocity of shallow structure, especially the Quaternary sediments at 0-1 km near the surface, is an important issue of concern in urban planning and construction for the requirements of seismic hazard assessment and disaster mitigation. Due to the facility and less dependence on the site environment, noise-based technique is an ideal way to acquire the fine structure of urban sedimentary basin. Based on the dense array composed of more than 900 stations deployed in Tongzhou at a local scale of 20 × 40 km², we proved the lateral variation of the phase velocity of multi-mode surface waves can be estimated directly with adequate accuracy by beamforming seismic noise with moving subarray, without tomography. Rayleigh wave phase velocity maps, at frequencies between 0.3 and 2.5 Hz for the fundamental mode as well as 0.8 and 3.0 Hz for the first overtone, are obtained. The 3D S-wave velocity model at 0-1 km depth with lateral resolution of 1 km is then established by inverting phase velocity maps of two modes. The thickness of the sediments is delineated by the impedance interface given by microtremor H/V (horizontal-to-vertical) spectral ratio. The model is in good agreement with tectonic unit. The sedimentary thickness of Daxing high and two sags located around Gantang and Xiadian are respectively 100-400 m and 400-600 m, which correlates well with the isosurface of S-wave velocity at 1 km/s. The model also presents some evidence on the extension of Daxing fault along NE direction.