Learning-based automatic modulation recognition on edge devices is envisioned as a critical enabler of real-time spectrum management in future 6G Internet of Things networks. In addition to the requirement of accurate modulation detection, edge devices pose the challenge of reducing the size of the automatic modulation recognition model. This paper proposes a lightweight hybrid model of multi-channel convolutional neural network and the mobile vision transformer. The proposed model addresses both accuracy improvement and model size reduction by employing weight-based pruning and post-training dynamic range quantization. Performance results in edge computing environments such as the Raspberry Pi and Jetson platforms using the RadioML 2016.10a dataset show that the proposed model achieved up to a 91% reduction in memory usage compared to its original version prior to pruning and quantization and demonstrated up to an 8% improvement in average accuracy compared to the baseline convolutional neural network model.

This paper presents a computationally efficient approach for state-of-health (SOH) estimation. This approach has two novel components: large language model (LLM)-based prompt engineering for feature extraction and a edge-optimized extreme learning machine (TinyELM) for efficient SOH estimation. In our LLM-based feature engineering, we leverage GPT-5 to automatically generate potential health indicators from raw voltage and current data, then select two to four features that exhibit strong correlations with the target SOH. Building upon these features, we develop TinyELM as a streamlined and optimized variant of ELM designed for improved estimation accuracy and portability on resource-constrained edge devices. The performance of TinyELM is compared against several deep learning-based models as well as the conventional ELM model and linear-based models on three edge devices, namely, the STM32, Raspberry Pi 4, and Jetson Nano. Experimental results demonstrate that TinyELM achieves state-of-the-art performance in SOH estimation while delivering up to 28.4× faster inference than the conventional ELM and 3× faster inference than the other linear models. Furthermore, the results highlight the effectiveness of our LLM-based feature engineering, showing approximately 5× lower root mean square error on average compared to traditional and automated feature engineering approaches.

This paper proposes a novel medium access control (MAC) protocol tailored for underwater swarmed AUV networks (USANs). Designed to address the challenges of acoustic communication such as long propagation delays and limited bandwidth, the proposed protocol adopts a contention-free, delay-aware time scheduling mechanism. It introduces a hierarchical frame structure with area-based slot partitioning and propagation delay-aware slot assignment to ensure collision-free and low-latency communication. A deterministic scheduling algorithm assigns transmission offsets based on sorted propagation delays, promoting fairness and scalability across multiple AUV swarms. Simulation results demonstrate that, compared to AB-MAC, AST-TDMA, TDMA, and slotted-ALOHA, the proposed MAC protocol reduces end-to-end delay by over 80%, increases uplink channel utilization up to 5× (exceeding 50%), and maintains a retransmission overhead ratio as low as 1.5 under heavy traffic conditions. Moreover, it achieves a propagation-aware delay factor close to 1.0, indicating near-optimal delay behavior. These results confirm the protocol’s effectiveness for reliable and efficient communication in dense USANs.

Temperature monitoring system (TMS) aims to reduce the infection spread and outbreak of COVID-19 through early detection. Conventional and currently deployed TMS have high implementation cost and require a substantial amount of space. Also, the performance often depends on the accuracy of the thermal camera. To address this, we propose Edge TMS wherein a multi-task cascaded convolutional neural networks (MTCNN)-based TMS is deployed on an edge AI device. To overcome the resource constraints of edge AI devices, an optimization method is applied to compress MTCNN up to 100×. The compressed MTCNN is deployed on the local PC, Jetson Xavier, Jetson TX2, and Jetson Nano which yields a Pruning-Per-Reduction Ratio (PPRR) values of 1.21, 1.63, 1.99, and 2.10, respectively. We proposed the PPRR metric to measure the performance of the compressed model. Low PPRR values indicate an improvement in the hardware performance and computational efficiency of the optimized model. The optimized model deployed in all the Jetson series achieved an average percent power reduced (%R) of 53.18% with a percent difference of 35.9% from the results of the local PC.

Social distance monitoring (SDM) systems are vital in fighting the spread of the coronavirus (COVID-19). Existing SDM systems employ bounding box-method, which imposes inaccurate distance estimation due to the high variance in its output coordinates. To solve this problem, an SDM system based on multitask cascaded convolutional neural networks (MTCNN) is proposed. Instead of using bounding box coordinates, face detection and facial landmarks localization of MTCNN are used to provide fixed coordinates and increase the distance estimation accuracy of SDM. However, while the accuracy issue is solved by using MTCNN, the SDM system suffer from large computational requirements due to the cascaded networks added on top of the distance estimation process. To deal with this challenge, a constrained optimization technique is employed to each stage of MTCNN with the goal of reducing its hardware requirements while keeping the same reliability as the original implementation. Experimental results show that the SDM system based on the optimized MTCNN achieves higher accuracy performance with reduced computational requirements as compared with conventional SDM systems. This allows the proposed SDM system using optimized MTCNN to be deployed efficiently on edge devices.

Smart streetlight is an outdoor infrastructure that uses technologies, such as sensors and actuators, to provide intelligent outdoor lighting and replace the power-consuming traditional streetlights. Although the current smart streetlight systems are employing these technologies for the maintenance, they simply gather data wirelessly on a periodical basis and still have drawbacks to provide real-time monitoring operation. To address these issues, the implementation of an IoT-based low-delay smart streetlight monitoring system is proposed in this article. In addition, a data filtering algorithm is also proposed in this article where redundant data are ignored to avoid overloading and excessive data storage consumption. Implementation results show that the proposed monitoring system and data filtering algorithm are able to provide real-time monitoring of smart streetlights with minimal time execution up to 0.11 ms and greatly reduce data storage usage up to 88.57%, respectively.

Wireless propagation loss modeling has gained significant attention due to its critical importance in forthcoming dynamic wireless technologies. Stochastic and map-based propagation models require more information (elevation extension, statistical scattering characteristics) than required by empirical models (i.e., operating frequency, distance between transceivers, and height of the antennas), but such information is not always available. Thus, empirical models are still widely used to evaluate coverage, link budget, and received signal strength. The drawback of empirical models is inaccuracy in highly dynamic transmitter and receiver environments. To reduce the error caused by the use of a single environment, we divide a geographical terrain to employ a specific propagation model in each segment of the wireless link. We enhance a deep learning (DL) encoder-decoder architecture to extract semantic information from satellite imagery to divide an environment into three classes. Our DL architecture achieved a segmentation accuracy of 89.41%, 86.47%, and 87.37% in urban, suburban, and rural classes, respectively. Simulation results indicate that estimating propagation loss with our multi-environment model reduced the root mean square deviation (RMSD) with respect to two publicly available wireless tracing datasets, CU-WART and Portland MetroFi, by 3.79dB and 4.09dB, respectively.

A new advanced medium access control protocol is described to support 100 km reach access networks, exhibiting the required quality-of-service for next generation passive optical networks based on orthogonal frequency division multiple access. The protocol enables the optical network units (ONUs) to utilize the idle period in each packet transmission time based on their originally granted bandwidth, using the same subcarriers but different time slots to increase the effective transmission bandwidth. The network throughput, end-to-end packet delay and packet loss rate are evaluated by means of both service level agreement and class-of-service differentiation. As a result the packet delay at 80% ONU offered load is less than 3 ms even for the lowest service level ONUs. In addition, the throughput efficiency is 94% of the total network capacity of 40 Gbps, for 100 km long-reach links.

In this paper, we shed some light on the latency and reliability issues of mobile backhaul networks, which have been largely ignored in the past, and examine their impact on LTE-A heterogeneous networks (HetNets). Specifically, we propose a backhaul-aware user association algorithm for fiber-wireless (FiWi) enhanced LTE-A HetNets. The performance limiting factors of state-of-the-art fiber backhaul infrastructures are highlighted and a variety of solutions are described. To mitigate the vulnerability of the backhaul against fiber cuts, we introduce different advanced protection techniques. Accounting for the given conditions of the backhaul in terms of delay and reliability, we present a distributed load-balancing algorithm for user association in FiWi-LTE HetNets. The proposed algorithm is analyzed and evaluated numerically by comparing its performance with state-of-the-art alternative approaches in terms of average delay, blocking probability, average achievable throughput, and service interruption percentage. The obtained results demonstrate that our algorithm outperform its counterparts in terms of delay and service interruption percentage, while its average achievable throughput is the same as that of a backhaul-unaware alternative solution. In addition, the blocking probability of the proposed backhaul-aware load-balancing method is shown to be higher than that of backhaul-unaware ones.