Sparse Bayesian learning-based massive multi-user detection algorithm

Abstract

Aiming at the problem that most existing algorithms were based on the Gaussian inverse gamma prior model (GIG-SBL), which ignored the sparsity of the support set vector within the sparse solution, a sparse Bayesian learning framework based on the Bernoulli Gaussian inverse gamma prior model (BGIG-SBL) was proposed.By introducing a binary vector of Bernoulli prior, the BGIG-SBL-SMV algorithm based on single measurement vector (SMV) was designed, which utilized the sparsity of the support set vector to improve reconstruction performance.Then the proposed algorithm was extended to multiple measurement vector (MMV) models by sharing the same hyperparameters.The BGIG-SBL-MMV algorithm was developed based on the joint sparsity of MMV.The experimental results show that the proposed BGIG-SBL-SMV can achieve a performance gain of 2 dB in mMTC over the traditional GIG-SBL-SMV.Moreover, BGIG-SBL-MMV has a performance gain of 4 dB as compared with its SMV counterparts, which demonstrates the advantages of the proposed schemes.

Keywords

• sparse Bayesian learning;compressive sensing;multi-user detection;mMTC