Comparison of Damping Approaches for AMP

Approximate message passing (AMP) is a lowcomplexity iterative reconstruction algorithm for compressed sensing. The disadvantage of this algorithm is its restriction to sensing matrices with independent and identically distributed (i.i.d.) entries. This can be avoided by utilizing other AMP-like algorithms, such as vector approximate message passing (VAMP), convolutional approximate message passing (CAMP) or memory approximate message passing (MAMP). However, all of these algorithms are either a lot more intricate due to a higher number of parameters, which have to be adjusted adequately, or have a higher computational complexity. A popular strategy to avoid divergence of AMP is damping, which is a simple and very low-complex solution. However, the usage of damping is not consistent throughout the literature and the approaches are not well motivated. This paper discusses the main damping strategies that are available in the literature and compares them to the other AMP-based algorithms.