Discrete Sparse Signals: Compressed Sensing by Combining OMP and the Sphere Decoder

Abstract: We study the reconstruction of discrete-valued sparse signals from underdetermined systems of linear equations. On the one hand, classical compressed sensing (CS) is designed to deal with real-valued sparse signals. On the other hand, algorithms known from MIMO communications, especially the sphere decoder (SD), are capable to reconstruct discrete-valued non-sparse signals from well- or overdefined system of linear equations. Hence, a combination of both approaches is required. We discuss strategies to include the knowledge of the discrete nature of the signal in the reconstruction process. For brevity, the exposition is done for combining the orthogonal matching pursuit (OMP) with the SD; design guidelines are derived. It is shown that by suitably combining OMP and SD an efficient low-complexity scheme for the detection of discrete sparse signals is obtained.