Tera-Hertz sub-layer object identification using MCA and Dictionary learning

We address the problem of identification of spherical objects, which are inside a layered structure by wireless sensing. Here, the backscatter response is used to identify the objects. It is a task, which is improved as the resolution gets higher. Therefore, to have a better resolution we consider the THz band. However, in the THz surface scattering expands the backscatter response of the layered structure in time. Thus, the backscatter response of the objects and the layered structure superimpose each other. This becomes even more severe due to the limited bandwidth at the receiver. In consequence, the exploitation of sophisticated signal separation methods for improved object identification is required. Therefore, we propose morphological component analysis (MCA), semi-blind MCA (MCA-SB) and singular value decomposition (SVD) methods. The MCA and MCA-SB rely on the sparse representation of signals. Therefore, dictionaries need to be identified to represent each signal components sparsely, while SVD relies on statistical properties of the signals. The simulation results demonstrate that the surface scattering of the layered structure makes the recovery process of the backscatter response of the objects difficult. However, this can be improved by sparse signal processing with suitable dictionary selection even for a weak objects response.