Joint Parameterized Resource Allocation for IAB Networks with Deep Reinforcement Learning

Abstract

The integrated access and backhaul (IAB) demonstrates significant advantages in terms of economy and stability over traditional fiber-like links. Although promising, with the high difficulty of interference management, appropriately allocating the spectrum bandwidth and power resources in IAB system to each communication link is still challenging. In this paper, we investigate the joint optimization problem of spectrum allocation and power management in an IAB network operating in the mm-Wave, which aims to improve the total date rate of the IAB system while satisfying the users' quality of service (QoS) requirements and to ensure fairness among users with the same QoS requirements. With these goals, we formulate the joint optimization as a mixed integer nonlinear programming (MINLP) problem. To solve this problem, we propose to use a state-of-the-art actor-critic based reinforcement learning algorithm in the parameterized action space, namely parameterized action-space deep deterministic policy gradient (PADDPG), to obtain an efficient optimization strategy on bandwidth and power allocation, called parameterized resource allocation (PRA). The proposed scheme can achieve model-free, real-time resources allocation in the absence of perfect CSI. Numerical experiments demonstrate the superiority of our scheme over benchmark algorithms in terms of data rates, QoS satisfaction and user fairness.