Signal Recovery in Perturbed Fourier Compressed Sensing
arXiv:1708.01398
Abstract
In many applications in compressed sensing, the measurement matrix is a Fourier matrix, i.e., it measures the Fourier transform of the underlying signal at some specified `base' frequencies , where is the number of measurements. However due to system calibration errors, the system may measure the Fourier transform at frequencies that are different from the base frequencies and where are unknown. Ignoring perturbations of this nature can lead to major errors in signal recovery. In this paper, we present a simple but effective alternating minimization algorithm to recover the perturbations in the frequencies \emph{in situ} with the signal, which we assume is sparse or compressible in some known basis. In many cases, the perturbations can be expressed in terms of a small number of unique parameters . We demonstrate that in such cases, the method leads to excellent quality results that are several times better than baseline algorithms (which are based on existing off-grid methods in the recent literature on direction of arrival (DOA) estimation, modified to suit the computational problem in this paper). Our results are also robust to noise in the measurement values. We also provide theoretical results for (1) the convergence of our algorithm, and (2) the uniqueness of its solution under some restrictions.
arXiv commentNew theortical results about uniqueness and convergence now included. More challenging experiments now included
References in corpus (9)
- Square-Root Lasso: Pivotal Recovery of Sparse Signals via Conic Programming
- Sparsity-Cognizant Total Least-Squares for Perturbed Compressive Sampling
- General Deviants: An Analysis of Perturbations in Compressed Sensing
- Compressive Phase Retrieval via Generalized Approximate Message Passing
- Self-Calibration and Biconvex Compressive Sensing
- Joint Sparse Recovery Method for Compressed Sensing with Structured Dictionary Mismatches
- Robustly Stable Signal Recovery in Compressed Sensing with Structured Matrix Perturbation
- Convex Optimization Approaches for Blind Sensor Calibration using Sparsity
- A Non-Convex Blind Calibration Method for Randomised Sensing Strategies