Unified Theory for Recovery of Sparse Signals in a General Transform Domain

Compressed sensing provided a new sampling paradigm for sparse signals. Remarkably, it has been shown that practical algorithms provide robust recovery from noisy linear measurements acquired at a near optimal sample rate. In real-world applications, a signal of interest is typically sparse not in the canonical basis but in a certain transform domain, such as the wavelet or the finite difference domain. The theory of compressed sensing was extended to this analysis sparsity model. However, existing theoretic results for the analysis sparsity model have limitations in the following senses: the derivation relies on particular choices of sensing matrix and sparsifying transform; sample complexity is significantly suboptimal compared to the analogous result in compressed sensing with the canonical sparsity model. In this paper, we propose a novel unified theory for robust recovery of sparse signals in a general transform domain using convex relaxation methods that overcomes the aforementioned limitations. Our results do not rely on any specific choice of sensing matrix and sparsifying transform and provide near optimal sample complexity when the transforms satisfy certain conditions. In particular, our theory extends existing near optimal sample complexity results to a wider class of sensing matrix and sparsifying transform. We also provide extensions of our results to the scenarios where the atoms in the transform has varying incoherence parameters and the unknown signal exhibits a structured sparsity pattern. Numerical results show that the proposed variable density sampling provides superior recovery performance over previous works, which is aligned with the presented theory.