Decentralized Learning Algorithm for Active User and Data Detection in mMTC CF mMIMO Systems

We consider the problem of detecting active user equipments (UEs) and their data, in a cell-free massive multiple-input multiple-output (CF-mMIMO) massive machine type communication (mMTC) systems. Due to large number of UEs in a mMTC system, data detection at each access point (AP) becomes a sparse recovery problem. By leveraging the sporadic user activity, we develop a decentralized sparse Bayesian learning algorithm, which first detects data locally for each AP, and then improves the detection performance by suitably combining information from other APs. This is done by optimizing the global objective via a local objective corresponding to each AP. We show that our decentralized Bayesian learning algorithm has a lower i) complexity and runtime than its centralized counterparts. Its complexity also remains invariant to the number of APs; and ii) bit error rate (BER) than other existing algorithms.