Deep Reinforcement Learning for QoS and Energy Optimization in 5G Network Slicing

Minah-Eeba, Winner, Bakare, Bodunrin Isa — Thu, 30 Jul 2026 00:00:00 +0000

Network slicing is an important feature of fifth-generation (5G) standalone systems, supporting the utilization of heterogeneous services like enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC) and massive machine-type communications (mMTC) over a common setup. However, conventional static and heuristic-based slicing methods are unable to effectively adapt to dynamic traffic situations, resulting in suboptimal quality of service (QoS) and energy consumption. To address these problems, this paper suggests an AI-enabled dynamic network slicing structure based on deep reinforcement learning (DRL). A Deep Q-Network (DQN) agent is established to perform real-time resource allocation by discerning network states and optimizing a reward function that jointly reflects QoS parameters and energy efficiency. Simulation results show that the proposed method increases system throughput by up to 25%, decreases URLLC latency to below 10ms, improves packet delivery ratio to 97%, and declines overall energy consumption by approximately 12–17% when compared with conventional approaches. These findings ratify the effectiveness of reinforcement learning for adaptive and energy-efficient network control and highlight its potential to allow autonomous, scalable, and efficient resource orchestration in 5G and upcoming AI-native wireless networks.

ROMEO: Raw Onboard Multispectral Earth Observation-An Edge-AI Framework for Latency- and Bandwidth-Efficient Satellite Sensing

Charan M. H., Abhishek G., Haridarshan K. P., Mallikarjuna, K. B. Janaki — Wed, 14 Oct 2026 00:00:00 +0000

Whether an orbital observation proves useful often depends less on what the instrument recorded than on how long the finding took to reach somebody able to act on it. Under present practice, that interval is dominated by two serial obligations, neither involving sensing: everything the instrument produces must cross a narrow channel, and a lengthy sequence of radiometric and geometric adjustments must then be completed before interpretation may begin. This study proposes ROMEO, which discharges the interpretive obligation in flight and applies it to measurements that have received no correction. The design comprises four stages: instrument interface, a minimal conditioning layer, a divided inference engine, and a controller governing what leaves the vehicle. Division of the inference engine distinguishes the proposal — a shallow admission network, the Gatekeeper, rejects tiles showing cloud, open water, or unremarkable ground, and only survivors reach a deeper segmentation network, the Mapper. Bands are brought into register by Coarse Spatial Coregistration, which replaces runtime correspondence search with constants tabulated beforehand, finishing a granule in roughly 16.63 ms on a CPU and 1.65 ms on a GPU at sub-pixel residual displacement. Supervision derives from THRawS and SegTHRawS. Because conclusions rather than imagery are released, the quantity requiring transmission falls from gigabytes to kilobytes.

International Journal of Satellite-Based Communication and Wireless Networks System

Deep Reinforcement Learning for QoS and Energy Optimization in 5G Network Slicing

ROMEO: Raw Onboard Multispectral Earth Observation-An Edge-AI Framework for Latency- and Bandwidth-Efficient Satellite Sensing