With the emergence of various Internet of Things (IoT) technologies, energy-saving schemes for IoT devices have been rapidly developed. To enhance the energy efficiency of IoT devices in crowded environments with multiple overlapping cells, the selection of access points (APs) for IoT devices should consider energy conservation by reducing unnecessary packet transmission activities caused by collisions. Therefore, in this paper, we present a novel energy-efficient AP selection scheme using reinforcement learning to address the problem of unbalanced load that arises from biased AP connections. Our proposed method utilizes the Energy and Latency Reinforcement Learning (EL-RL) model for energy-efficient AP selection that takes into account the average energy consumption and the average latency of IoT devices. In the EL-RL model, we analyze the collision probability in Wi-Fi networks to reduce the number of retransmissions that induces more energy consumption and higher latency. According to the simulation, the proposed method achieves a maximum improvement of $53\%$ in energy efficiency, $50\%$ in uplink latency, and a $2.1$-times longer expected lifespan of IoT devices compared to the conventional AP selection scheme.

To support dramatically increasing services from internet of thing (IoT) devices with the sporadic and fluctuated generation of short packet traffic, this paper investigates joint dynamic time division duplexing (TDD) and radio resource control (RRC) connection management in a single-cell massive IoT network. Specifically, under the grant-free transmission incurring packet collision, this study models the factors affecting the time resource utilization (TRU) and energy consumption of IoT devices as a comprehensive system utility and further formulates the problem as a decision-making process aiming for balancing the long-term average TRU and energy consumption. To address the formulated problem, based on the deep reinforcement learning framework, this paper designs an experience-driven joint dynamic TDD and RRC connection management scheme that intelligently i) determines the TDD configuration based on the most recent downlink (DL)/uplink (UL) traffic demands and ii) adjusts the RRC state of each IoT device to control the maximum number of transmitting IoT devices. Finally, trace-driven simulation results demonstrate that the proposed scheme outperforms existing benchmarks, such as Static TDD and Dynamic TDD , in terms of transmission success ratio difference (TSRD) (up to 89% reduction), time resource utilization (TRU) (up to 17× increase), and energy consumption (up to 70% reduction) of IoT devices.

In 5G networks, the legacy paging mechanism for device connections faces challenges due to limited resources and increasing demands from human-type communication (HTC) and massive machine-type communications (mMTC). This results in higher connection setup time (CST) under high traffic. Group paging, developed for mMTC, allows multiple devices to respond to a single paging message, but increases energy consumption for HTC due to irregular data transmission patterns. To address this, we propose the deep reinforcement learning-based group configuration control (DRL-GCC) mechanism. DRL-GCC combines legacy and group paging methods to optimize paging for HTC by dynamically adjusting group configurations, using the proximal policy optimization algorithm to reduce energy consumption and satisfy the required CST. Simulation results demonstrate that DRL-GCC significantly reduces energy consumption in the random access channel (RACH) procedure and the overall energy usage of devices while ensuring compliance with the CST constraint compared to its alternative methods. Notably, DRL-GCC achieves up to a 43% reduction in energy used during the RACH procedure and a 9% decrease in total energy consumption compared to the existing benchmarks. These improvements imply DRL-GCC’s ability to balance energy efficiency with operational performance, providing a robust solution for efficient HTC-focused service provisioning in 5G networks.

Group Paging (GP) is essential for scalable device wake-up in Fifth-Generation (5G) Internet of Things (IoT) networks. However, grouping devices with heterogeneous traffic characteristics can lead to energy inefficiency and degraded access reliability. This paper presents Deep Learning-based Group Paging (DL-GP), a unified Deep Learning (DL) framework that learns energy-optimal group assignments while satisfying Quality of Service (QoS) constraints. We formulate GP as an energy-aware optimization problem and solve it using a differentiable model that integrates: (i) a Sinkhorn layer to approximate discrete group labels, (ii) a Log-Sum-Exponential (LSE) surrogate to stably optimize Transmission Frequency (TF) and Discontinuous Reception (DRX) cycles, and (iii) an adaptive retransmission controller to enforce Access Success Rate (ASR) requirements. Trained on synthetic and LoRaWAN traces, DL-GP reduces average device energy consumption by 35% and 17% compared to Greedy and Genetic baselines in large-scale simulations, and by 66% and 64% in a real LoRaWAN deployment, respectively. It maintains low Paging Overhead (PO), satisfies ASR requirements, and achieves sub-millisecond processing latency. These results demonstrate that DL-GP is a practical real-time solution for energy-efficient and QoS-compliant paging in large-scale 5G IoT systems.