Fully Decentralized Cell-Free Massive MIMO Networks

Cell-free massive multiple-input multiple-output (CF-mMIMO) has been cast as a technological pillar for sixth generation (6G) wireless networks to suppress inter-cell interference and enhance spectrum efficiency. As the conventional centralized CF-mMIMO network needs to connect all the base stations (BSs) to the central processing unit (CPU) for baseband signal processing, the backhaul signaling overhead and computational complexity become challenging with the increasing number of BSs and the channel dimension. To this end, we propose a fully decentralized CF-mMIMO network without the CPU where each BS only exchanges limited signaling information with its adjacent neighbors and carries on local computation. Firstly, the alternating direction method of multipliers (ADMM) algorithm is utilized to convert the weighted sum-rate (WSR) maximization problem into a decentralized consensus optimization problem that each BS locally updates beamformers and combiners until a consensual solution is reached. Then, we employ the block gradient descent (BCD) algorithm to optimize the hybrid beamforming for BSs and adopt the manifold optimization (MO) to design the combining for users in a decentralized manner, respectively. Additionally, the convergence, the complexity and the backhaul signaling overhead are analyzed in detail. Finally, numerical results demonstrate that the WSR performance of the decentralized CF-mMIMO network basically approaches the centralized CF-mMIMO network while the communication-computation efficiency outperforms the centralized counterpart.