PABSO-DRL: Power and Beam Self-Optimization Scheme for Multiple Slices in MU-MISO Systems

Recently, the concept of zero-touch networks (ZTNs) relying heavily on pervasive artificial intelligence (AI) algorithms for the full automation of future networks have emerged. ZTNs, empowered by AI, will play a significant role in reducing the management complexity of beyond fifth-generation (B5G) networks. One enabling technology of ZTNs is network slicing (NS), which is the cornerstone of B5G networks. However, NS faces challenges, particularly in terms of radio resource allocation management. Therefore, this paper presents an efficient self-optimization scheme for a multi-user multiple-input single-output (MU-MISO) system across different slices; named PABSO-DRL. This scheme dynamically and jointly manages power allocation and beam direction based on a deep reinforcement learning (DRL) framework, considering imperfect channel state information (CSI) and the time-varying dynamics of the NS environment. PABSO-DRL ensures high data rates for enhanced mobile broadband (eMBB) while guaranteeing high reliability for ultra-reliable low latency communications (uRLLC). The problem is formulated as a multi-objective optimization problem and solved by designing multiple deep Q-learning (DQL) agents. The proposed scheme is extensively evaluated under two scenarios: perfect and imperfect CSI, comparing its performance with four traditional benchmark algorithms, and state-of-the-art schemes. Results demonstrate the superiority of the proposed DRL scheme, even under imperfect CSI, highlighting its adaptability to various network slicing conditions.