Joint Detection for Massive Grant-free Access via BiGAMP

With the explosive growth of the number of devices accessing the network, massive access has been a challenging task for the fifth generation (5G) and beyond. The uplink massive access scenario has the feature that only a subset of potential devices is active in each coherence time. In this paper, to serve more devices with lower overhead in massive access scenarios, we adopt a grant-free scheme where devices are allocated unique non-orthogonal pilots. We also design a kind of bilinear generalized approximate message passing (BiGAMP) algorithm which utilizes the row sparse channel matrix structure due to sporadic device access. In this work, the proposed algorithm jointly detects device activities, estimates channels, and detects signals in massive multiple-input multiple-output (MIMO) systems with one phase. The signal observation affords additional information to improve performance compared to other algorithms. In addition, we further analyze state evolution (SE) to characterize the performance of the BiGAMP algorithm. The numerical results demonstrate that the proposed algorithm performs better than baselines in device activity detection, channel estimation, and signal detection. And the numerical results of SE can describe the performance of the proposed algorithm.