Iterative Hard Thresholding Using Least Squares Initialization

An improvement of the typical iterative hard thresholding (IHT) algorithm is presented by means of least squares initialization. Unlike the initialization that assigns all zeros to the initial estimate, the linearized least squares estimate is instead adopted herein. Numerical examples are conducted under a sparse signal recovery problem. Performance of the proposed approach in terms of root-mean-square relative error (RMSRE), computational time, and memory consumption is compared to the typical IHT algorithm. It is shown in numerical example that a good initialization can deliver lower RMSRE, less computational time, and less memory requirement.