This page provides a MATLAB software implementation for the new algorithms we proposed and evaluated in:
D. Kim, J. P. Haldar. “Greedy Algorithms for Nonnegativity-Constrained Simultaneous Sparse Recovery.” Signal Processing 2016, In Press.
[link]
These algorithms are extensions of classical greedy algorithms for sparse recovery, and offer improved performance in the presence of additional nonnegativity (NN) constraints, simultaneous sparsity (S) constraints, or a combination of nonnegativity and simultaneous sparsity (NNS) constraints.
The software includes NN-, S-, and NNS- variations of the following four greedy algorithms:
- Orthogonal matching pursuit (OMP)
- Subspace pursuit (SP)
- Compressive sampling matching pursuit (CoSaMP)
- Hard thresholding pursuit (HTP)
For comparison, the original (unconstrained) algorithms are also provided.
The software also includes an illustrative example that empirically demonstrates the value of using additional NN, S, and NNS constraints. The example output (shown below) demonstrates that the NNS-based version of SP can offer substantially better perfect recovery performance than previous SP-based algorithms that use less constraints.
The software can be downloaded using the form below: