Doped expectation propagation for low-complexity message passing based detection

摘要

Expectation propagation (EP) has been used for Gaussian approximation of discrete-valued symbols to achieve low-complexity message passing based detection. It has been shown that EP-based Gaussian approximation significantly outperforms the direct Gaussian approximation. A doped EP (DEP) approach to Gaussian approximation is proposed, where the EP message is mixed with the Gaussian message obtained through direct approximation. It is demonstrated that the DEP approach delivers remarkably better performance than the EP one.