The proliferation of large-scale Internet-of-things (IoT) deployments and the emergence of 6G wireless technologies have created a pressing need for intelligent, energy-aware, and low-latency communication frameworks. In this work, we propose a novel two-phase reinforcement learning (RL)-based architecture designed to minimize the age of information (AoI) in 6G-enabled IoT networks. Our approach integrates (i) a deep deterministic policy gradient (DDPG)-driven backscatter-assisted cognitive radio non-orthogonal multiple access (CR-NOMA) scheme in the uplink, and (ii) a lightweight Q-learning-based power-domain NOMA (PD-NOMA) strategy for the downlink. In the uplink, energy harvesting (EH) sensors employ deep RL to jointly optimize backscatter reflection coefficients and transmission scheduling over shared spectrum using CR-NOMA. This enables energy-efficient communication and reduced AoI under dynamic energy and channel conditions. In the downlink, the edge node serves multiple IoT users simultaneously using PD-NOMA, where a Q-learning agent intelligently decides whether to transmit fresh or cached data to each user based on battery levels, channel quality, and information freshness. Both phases are modeled as Markov decision processes (MDPs), allowing agents to learn independently and converge toward optimal policies that balance information freshness, spectral efficiency (SE), and energy constraints. Extensive simulations demonstrate that the proposed framework effectively reduces AoI across both phases, with consistent convergence even under varying sensor densities and EH conditions. Moreover, by relying on explainable and verifiable learning mechanisms, our model addresses emerging concerns around reliability and trustworthiness in artificial intelligence (AI)-driven 6G-IoT systems. This framework represents a step toward scalable, adaptive, and responsible AI integration for future mission-critical IoT applications.

The proliferation of low-powered devices in the Internet of Things (IoT) necessitates energy-efficient communication paradigms. Backscatter communication (BackCom) combined with cognitive radio-inspired non-orthogonal multiple access (CRNOMA) offers a promising solution. However, optimizing such systems is complex, especially when considering realistic energy harvesting (EH) models. This paper investigates the sum rate optimization of an EH-enabled passive backscatter node (BN) coexisting with primary devices (PDs) in a CR-NOMA network, employing a practical non-linear EH model. We leverage deep reinforcement learning (DRL) to dynamically optimize the BN’s reflection coefficient. Crucially, we conduct a comparative study of several DRL algorithms. These include deep deterministic policy gradient (DDPG), its prioritized replay variants (PERDDPG and CER-DDPG), and the stability-enhanced twin delayed DDPG (TD3). We also evaluate on-policy methods such as proximal policy optimization (PPO), as well as entropy-regularized algorithms like soft actor-critic (SAC) and its recurrent extension (RSAC). Additionally, we include the asynchronous advantage actor-critic (A3C) for comparison. We evaluate their performance in terms of sum rate, reflection coefficient adaptation, and harvested energy, contrasting results under non-linear versus linear EH models. Our findings provide insights into algorithm suitability for optimizing BackCom systems under realistic EH constraints, highlighting performance trade-offs and the impact of EH non-linearity.

With the increasing number of Internet-of-things (IoT) devices, the need for energy-efficient and spectrum-efficient networks that can support resource-constrained devices within existing wireless infrastructures becomes critical. This paper investigates the application of deep reinforcement learning (DRL) algorithms to optimize the energy efficiency (EE) of a secondary device (SD) equipped with radio frequency energy harvesting (RF-EH) antennas. The system models a wireless powered communication network (WPCN) where the SD employs a cognitive-radio non-orthogonal multiple access (CR-NOMA) scheme to transmit data during uplink communications of neighboring primary devices (PDs). Among the DRL approaches evaluated, proximal policy optimization (PPO) emerged as the most effective, achieving the highest EE values and demonstrating its suitability for this problem. Additionally, our results show that equal gain combining (EGC) consistently achieves superior EE compared to other diversity-combining techniques, making it a favorable choice for self-sustaining IoT networks. These findings provide valuable insights into the role of diversity-combining techniques and DRL algorithms in enhancing SD performance in dynamic EH environments.

With the rapid growth of Internet-of-Things (IoT) devices and unmanned agents (UAs), there is a rising need for energy- and spectrum-efficient wireless networks that can support large-scale, resource-constrained deployments. To meet this demand, integration of deep reinforcement learning (DRL), non-orthogonal multiple access (NOMA), and energy harvesting (EH) offers a promising approach to enhance energy efficiency (EE) and spectrum utilization in future sixth-generation (6G) networks, particularly for sustainable Internet of UA (IUA) communications. In this paper, we investigate an IUA network where multiple low-power secondary users (SUs), equipped with radio frequency energy harvesting (RF-EH) antennas, use a cognitive radio NOMA (CR-NOMA) scheme to share uplink channels with nearby primary users (PUs). We formulate a joint transmit power control and EH scheduling problem to maximize the long-term EE of the SUs and spectrum utilization of the network, subject to quality-of-service (QoS) constraints. To address the decentralized nature of the problem, we model the environment as a multi-agent system where each SU independently optimizes its transmission and EH strategies. A range of DRL and non-DRL algorithms is then applied to solve this optimization problem. We also explore different RF-EH diversity combining techniques to further boost system performance. Simulation results highlight the impact of these techniques on EE of SU, offering insights for optimizing performance under dynamic EH conditions.

Age of information (AoI) is a critical metric for ensuring data freshness in carbon-intelligent and energy-efficient industrial Internet-of-things (IIoT) networks. We consider a user-specific sensing framework operating in a time division duplexing (TDD)–based IIoT network, consisting of energy harvesting (EH) sensors, users, and a cache-enabled edge node. The proposed framework aims to minimize AoI during data transmission to users while addressing the sensors’ energy constraints. For the uplink transmission of data from sensors to edge node, a quality-of-service-aware cognitive-radio non-orthogonal multiple access (CR-NOMA) is employed to enhance spectrum efficiency and minimize cache AoI. During the downlink transmission, upon user’s request, the edge node dynamically decides whether to retrieve cached data or request a fresh update from the sensors. This decision-making process is driven by reinforcement learning (RL) using a Q-table-based approach, where the edge node infers sensor battery levels from received updates and prioritizes user requests accordingly. We formulate this problem as a Markov decision process (MDP) and define an optimal policy that strikes a balance between minimizing AoI and adhering to energy constraints. To achieve this, we develop RL-based solutions, including Q-learning and deep Q-networks (DQN). Extensive simulations demonstrate that our approach achieves up to 90% AoI reduction while significantly improving energy efficiency, making it well-suited for real-time, delay-sensitive applications in modern IIoT networks.

Amidst the ongoing debate about limited spectral availability, there remains a persistent demand for the development of spectrally efficient self-sustainable network (SSN) models. This paper addresses this challenge by optimizing spectral efficiency (SE) in uplink transmissions for an energy harvesting (EH)-enabled secondary user (SU) that operates opportunistically among multiple primary users (PUs) in an Internet-of-things (IoT) network. The PUs are assumed to employ a rotational time division multiple access (TDMA) scheme for transmissions, where the signals are divided into time slots for each PU to transmit data in a cyclic manner, while the SU uses an opportunistic non-orthogonal multiple access (NOMA) technique to transmit data without interfering with the PU transmissions, such that, at any given time slot, a PU and a SU share the same frequency band simultaneously. The SE of the system is maximized jointly by employing convex optimization and a deep reinforcement learning (DRL) model, specifically the deep deterministic policy gradient (DDPG) algorithm. Simulations demonstrate that the proposed approach significantly improves the SE of the considered IoT network, highlighting its potential for efficient spectrum management in IoT networks. We present a comprehensive SE analysis of the system, which further underscores the robustness and adaptability of our approach in optimizing SE under diverse operational conditions.

Mobile edge computing (MEC) has emerged as a promising paradigm to enhance the computational capabilities of resource-constrained secondary devices (RCSDs) in proximity to prescheduled primary devices (PDs). In this context, we introduce a novel framework where an energy harvesting (EH)-enabled RCSD efficiently offloads computational tasks to an MEC server, while employing a cognitive radio-inspired non-orthogonal multiple access (CR-NOMA) scheme for efficient data transmission. The RCSD also harvests energy from the ambient radio frequency (RF) signals of the surrounding PDs. We propose a deep reinforcement learning (DRL)-based optimization strategy, specifically the deep deterministic policy gradient (DDPG) algorithm to minimize the service delay of the RCSD and optimize the time-sharing coefficient for harvesting energy when offloading computational tasks to the MEC server. This dynamic resource allocation strategy intelligently determines the duration for which RCSDs transmit data and allocate time for energy harvesting, thereby ensuring an optimal balance between computation offloading and energy sustainability. Simulations demonstrate the effectiveness of the proposed scheme in max-imizing the utility of the RCSDs while minimizing the overall service delay of the RCSD.

In the rapidly evolving landscape of advanced wireless networks, self-sustainable Internet of Things (IoT) networks become pivotal, necessitating to seamlessly accommodate additional resource-limited devices into the existing wireless infrastructures. To this end, this article considers an IoT scenario with a wireless-powered communication network (WPCN) where a resource-constrained secondary node (SN) with energy harvesting (EH) capabilities harvests energy from the ambient radio-frequency (RF) signals to meet its energy requirements. Notably, we introduce RF-EH diversity-combining techniques, such as equal gain combining (EGC), maximum ratio combining (MRC), and selection combining (SC), tailored for linear EH models. To address the spectrum scarcity, the SN employs a Quality of Service (QoS)–aware nonorthogonal multiple access (NOMA) scheme to opportunistically transmit data within the uplink transmissions of the primary devices (PDs) operating around. Aiming to maximize the sum rate of the SN, we jointly optimize the EH time and transmit power of the SN using deep reinforcement learning (DRL). Specifically, we implement a set of DRL and non-DRL algorithms to investigate their robustness in diverse RF-EH diversity-combining environment settings. Simulation results demonstrate the influence of diversity combining techniques on the sum rate performance of the SN, providing valuable insights into their role in optimizing SN performance under dynamic EH environments.

This study presents an analysis of the bit error rate (BER) and system capacity in a multi-user multiple-input multiple-output (MU-MIMO) wireless system that deploys or-thogonal frequency-division multiplexing (OFDM) for wideband communication in industrial Internet-of- Things (IIoT) networks. The focus is on analyzing and evaluating the impact of imperfect channel state information (CSI) on MU-MIMO-OFDM system performance in comparison to perfect CSI within industrial settings. For this, we consider an IIoT network consisting of a base station (BS) and multiple IIoT devices, each equipped with multiple antennas. The CSI is computed using the least squares (LS) estimation technique. Furthermore, we investigate the tradeoff between system capacity and BER performance, considering various MIMO configurations to determine an optimal setup. The simulation results demonstrate that both imperfect CSI and MIMO configurations significantly influence BER performance. The findings of this research could provide valuable insights for the design and optimization of MU-MIMO-OFDM-based IIoT networks.

In today's world of burgeoning IoT and 6G communications, supporting low-powered devices is crucial to fully capitalizing on the Next Generation Internet of Things (NG-IoT) revolution. The proliferation of these devices will unlock unprecedented opportunities for energy efficiency, sustainability, and ubiquitous connectivity. This paper investigates the sum rate optimization of a quality-of-service (QoS)-aware EH-enabled passive IoT device in a cognitive radio inspired non-orthogonal multiple access (CR-NOMA)-assisted backscatter communication network. Our goal is to optimize the sum rate of a secondary passive IoT device while guaranteeing the QoS requirements of the scheduled primary device. The deep deterministic policy gradient (DDPG) algorithm is employed to dynamically adjust the reflection coefficient of the backscatter node, yielding optimal performance. Our results demonstrate significant improvements in the sum rate, highlighting the importance of incorporating advanced machine learning (ML) techniques into IoT and wireless communication domains to address critical challenges and enhance the overall performance of NG-IoT networks.