Joint Transmission Mode Selection and Scheduling for AoI Minimization in NOMA-Capable WP-IoT Networks: A Deep Transfer Learning Solution

Abstract

Age of information (AoI) serves as a key metric for characterizing information freshness. In this article, we investigate the AoI minimization of a non-orthogonal multiple access (NOMA)-capable wireless-powered Internet of Things (WP-IoT) network, where a base station (BS) consistently sends radio frequency (RF) signals to power IoT sensors (IoT-Ss), and selected IoT-Ss are scheduled to transmit status update packets to the BS in each time slot. We first formulate a scheduling problem for average AoI (AAoI) minimization with NOMA transmission and solve for a near-optimal scheduling policy with a deep Q-network (DQN)-based solution. Next, we propose a novel joint transmission mode selection and scheduling (JTMSS) design to further minimize the AAoI of the network. Specifically, the system adaptively selects one of three transmission modes: NOMA, orthogonal multiple access (OMA), and no transmission and schedules two, one, or none sensors for transmission, respectively. Considering the discrete hierarchical action space of the JTMSS problem, we formulate a parameterized-action Markov Decision Process (PAMDP) and develop a deep transfer learning (DTL)-based solution with a two-tier DQN framework to find a near-optimal JTMSS policy. Besides, we present a partial tuning approach during online operation to alleviate the effects of environmental changes. Simulation results verify that the JTMSS design with DTL achieves a significant performance gain over the scheduling-only policy for NOMA transmission. Moreover, the online tuning with DTL converges quickly during online operation.