Compressed sensing kernel design for radar range profiling

Compressive sensing (CS) is a technique for accurate signal reconstruction using lower sampling rates than prescribed by Nyquist/Shannon sampling theory under conditions where the signal has a sparse representation in some basis. However, the random projections usually adopted by CS do not exploit priori knowledge of the sensing task or signal structure (other than sparsity). In this paper, we use a task-specific information-based approach to optimizing sensing kernels for radar range profiling of man-made targets. We assume a MoG prior model for the targets and a Taylor series expansion that enables a closed-form gradient of information with respect to the matrix representation of the sensing kernel. We compare the performance of this optimized sensing matrix to random measurements and to optimum Nyquist performance. Simulation results demonstrate that the proposed technique for sensing kernel design outperforms random projections.