Pilot-based decentralized precoding design for UE-centric cell-free massive MIMO

Abstract

This thesis endeavors to propose a fully decentralized precoding design within a UE-centric cell-free massive MIMO system, focusing on minimizing MSE. To adapt to the UE-centric setting, an AP allocation algorithm was devised based on LSF coefficients. The AP selection algorithm comprises two levels: initially establishing optimal UE-AP connections by prioritizing APs with higher LSF coefficients, followed by widening AP clusters around those with maximal LSF values to augment the network performance.To guarantee that the network will always serve each UE regardless of its channel condition, a coordinated AP allocation is also included in the AP allocation algorithm.

Following the configuration of the UE-centric environment, fully decentralized precoder designs are implemented using two algorithms based on the gradient-based (GB) and best-response (BR) methods. These algorithms iteratively optimize precoders through bi-directional training sequences. The OTA approach is proposed to avoid the need for backhaul signaling to exchange UE-specific CSI during the training process. Consequently, the UE precoders must be designed at each AP, varying based on the association. These precoders are designed through MSE optimization. This framework can be extended to time-varying channel conditions, allowing for regular updates to AP selection and the design of precoders at each AP based on the association. Numerical results have demonstrated the superiority of decentralized precoding designs over conventional systems, such as centralized approaches, local MMSE, and local MF. This advantage underscores the efficacy of decentralized strategies in optimizing communication system performance under realistic operational conditions.