The deployment of large-scale LoRaWAN networks requires jointly optimizing conflicting metrics like Packet Delivery Ratio (PDR) and Energy Efficiency (EE) by dynamically allocating transmission parameters, including Carrier Frequency, Spreading Factor, and Transmission Power. Existing algorithms often ignore the complexity of multi-objective dynamic adaptation to oversimplify this challenge, focusing on a single metric or lacking the adaptability needed for dynamic channel environments, leading to suboptimal performance. To address this, we propose two online learning-based resource allocation frameworks that intelligently navigate the PDR-EE trade-off. Our foundational proposal, D-LoRa, is a fully distributed framework that models the problem as a Combinatorial Multi-Armed Bandit. By decomposing the joint parameter selection and employing specialized, disaggregated reward functions, D-LoRa dramatically reduces learning complexity and enables nodes to autonomously adapt to network dynamics. To further enhance performance in LoRaWAN networks, we introduce CD-LoRa, a hybrid framework that integrates a lightweight, centralized initialization phase to perform a one-time, quasi-optimal channel assignment and action space pruning, thereby accelerating subsequent distributed learning. Extensive simulations and real-world field experiments demonstrate the superiority of our frameworks, showing that D-LoRa excels in nonstationary environments while CD-LoRa achieves the fastest convergence. In physical deployments, our algorithms outperform state-of-the-art baselines, improving PDR by up to 10.8% and EE by 26.1%, demonstrating their practical effectiveness. Moreover, extensive simulations with up to 250 nodes confirm the scalability and efficiency of the proposed frameworks in large-scale LoRaWAN networks.

The deployment of LoRa networks necessitates joint performance optimization, including packet delivery rate, energy efficiency, and throughput, by dynamically configuring multiple LoRa parameters for packet transmission across varying channel environments. Due to the complexity of modeling channel features and the coupling relationship between LoRa parameters and metrics, existing works have sacrificed adaptability by focusing on specific aspects rather than the whole. Therefore, we propose D-LoRa, a distributed parameter adaptation scheme, based on reinforcement learning towards network performance. We first build a comprehensive analytical model for the LoRa network that considers complex channel features, including path loss, quasiorthogonality of spreading factor, and packet collision. Then, we formulate the joint optimization problem as a combinatorial Multi-Armed Bandit (CMAB) problem and devise metric factors to handle the trade-off among different performance metrics. Experimental results show that our scheme can increase the packet delivery rate by up to 18.5% and demonstrate superior adaptability across different performance metrics.

Using multiple unmanned aerial vehicles (UAVs) with backscatter communication to collect data from Internet of Things (IoT) devices has emerged as a promising solution. However, many existing UAVs path planning schemes for data collection suffer from performance degradation due to their limited consideration of the full collaboration of UAVs and dynamic stochastic environments. Therefore, we propose a path planning scheme for the data collection task in multi-UAV IoT based on multi-agent reinforcement learning (MARL) to minimize the task completion time. Due to the inherent asynchronous decision making among the agents, we model the path planning problem as a macro-action decentralized partially observable Markov decision process. Furthermore, we design an action mask mechanism to enhance data efficiency, which accelerates the training speed. Simulation results show that our scheme reduces the average task completion time by 15 %.

Wireless sensor networks (WSNs) enable intelligent collaborative perceptions in the Internet of Things. However, devices in WSNs are battery-powered with limited energy resources. During transmission, routing policies significantly affect the energy efficiency in terms of both energy consumption and energy balance among nodes, and further impact the network lifetime. Previous works mostly used heuristic fixed strategies to make routing decisions based on incomplete information in a distributed manner for lower control costs and faster calculation when facing numerous devices in WSNs, which easily lead to performance limitations and routing loops. To this end, we model the network lifetime maximization problem as a decentralized partially observable Markov decision process and propose a new scheme MeFi based on Mean Field Reinforcement Learning to perform real-time energy-efficient routing policies for WSNs. The utilization of Mean Field Theory effectively simplifies the intractable interactions among numerous agents and guides the policy training. Additionally, a prioritized-sampling loop-free algorithm is developed to eliminate routing loops and avoid routing policies with significant energy consumption. Experimental results show that our scheme outperforms several algorithms by up to 50%, significantly enhancing energy efficiency and extending WSN lifetime under different circumstances.

As one of the key concepts in the 5G network, MEC can support the latency-sensitive and compute-intensive services by widely deploying computing and storage capacity to the base stations at the network edge. Because these services are sensitive to latency, the joint optimization problem of task offloading and resource allocation needs to be solved in a short time. In this paper, we propose a Fast AI-assistant Solution for Task Offloading and Resource Allocation in MEC (FAST-RAM), which can directly solve the joint optimization problem leveraging a deep neural network. FAST-RAM can produce the offloading policy and resource allocation scheme in milliseconds. Meantime, our solution has near-optimal performance and sufficient feasibility under different network environments.

More and more applications learn from the data collected by the edge devices. Conventional learning methods, such as gathering all the raw data to train an ultimate model in a centralized way, or training a target model in a distributed manner under the parameter server framework, suffer a high communication cost. In this paper, we design Select Neighbors and Parameters (SNAP), a communication efficient distributed machine learning framework, to mitigate the communication cost. A distinct feature of SNAP is that the edge servers act as peers to each other. Specifically, in SNAP, every edge server hosts a copy of the global model, trains it with the local data, and periodically updates the local parameters based on the weighted sum of the parameters from its neighbors (i.e., peers) only (i.e., without pulling the parameters from all other edge servers). Different from most of the previous works on consensus optimization in which the weight matrix to update parameter values is predefined, we propose a scheme to optimize the weight matrix based on the network topology, and hence the convergence rate can be improved. Another key idea in SNAP is that only the parameters which have been changed significantly since the last iteration will be sent to the neighbors. Both theoretical analysis and simulations show that SNAP can achieve the same accuracy performance as the centralized training method. Compared to the state-of-the-art communication-aware distributed learning scheme TernGrad, SNAP incurs a significantly lower (99.6% lower) communication cost.

Video streaming service has been consuming a massive amount of Internet traffic during recent years. Even though Dynamic Adaptive Streaming over HTTP (DASH) has become the mainstream technology for improving users' Quality of Experience (QoE), the competing of multiple independent DASH streams could degrade the QoE and make unfair resource allocation. With the support of Software Defined Networking (SDN), it is possible to jointly optimizing resource allocation and bitrate adaptation in this competing scenario. In this paper, we propose FAIR-AREA, a fast Artificial Intelligence based joint optimization of rate adaptation and resource allocation of DASH service. With FAIR-AREA, we can solve this complex optimization problem only in milliseconds and achieve near optimal performance at the same time.

Mobile edge computing (MEC) is an emerging paradigm which has drawn much attention from the academy and industry. Leveraging 5G technique, MEC provisions the ability to support the latency-sensitive and compute-intensive services by deploying computing and storage capacity at the network edge. As one of the critical problems in MEC, resource allocation problem needs to be solved within a very short time to satisfy the low latency requirement of services. In this paper, we propose an Accelerator for Resource allocation in MEC (ARM), which can directly solve the resource allocation problem based on deep neural network. With the aid of our scale-free representation scheme and feasible guarantee algorithm, ARM can solve the problem in milliseconds. Meanwhile, our algorithm achieves near 2-factor approximation to the optimal solution.

Dynamic Adaptive Streaming over HTTP (DASH) has been broadly applied within most of mainstream video delivery services. At the same time, the lack of Quality of Experience (QoE) and the competition among several DASH streams have attracted significant attention. Aiming to supply better QoE for DASH streams in terms of video quality as well as starvation-free playing back, and achieve fairness among clients, we try to formulate the joint optimization of resource allocation and rate adaptation (JRA2) problem in this paper based on the information of streams' playout buffers and available network bandwidth. We first analyze the buffer behavior of a DASH stream using a modified version of M/D/1 queue and formulate the JRA2problem based on the analysis. To solve this problem, we propose an algorithm based on Generalize Benders Decomposition (GBD) to get optimal solution (JRA2-G) and devise one heuristic algorithm (JRA2-A) for acceleration. Based on the sufficient demonstrating results, the proposed algorithms can supply high-quality and smooth playing back for clients.

Hybrid Software-Defined Networking (HSDN) is a transitional networking form of SDN where SDN elements are partially deployed in traditional networks. Previous researches show that redirecting every flow of source-destination pair through at least one SDN switch can obtain flow manageability, e.g., access control and traffic measurement. Intuitively, the selection of SDN switch as the waypoint for every flow has a significant effect on the Traffic Engineering (TE) performance, such as maximum link utilization and routing efficiency. And it is worth noting that SDN switch can split traffic to the outgoing links to exactly profit the TE performance. In this paper, from the perspective of TE performance, we propose a flow routing and splitting (FRS) algorithm whereby we jointly determine an appropriate SDN switch for every flow as the waypoint, as well as optimizing the traffic splitting fractions for every SDN switch among its outgoing links to minimize the maximum link utilization. We conduct simulations with different SDN deployment rate. The results indicate that, when 20% of the SDN switches are deployed, the proposed FRS algorithm can obtain a lower maximum link utilization compared with other state-of-art works. Not only that, FRS algorithm can also generate a little longer paths for every flow on the average, which has a limited influence on the routing efficiency.

A constant-g m input stage, which features both constant small-signal and large-signal behavior over the entire input common-mode range, is proposed in this paper. The technique is based on a single differential pair, avoiding the issues associated with complimentary pair techniques, such as the degradation of common-mode rejection ratio and signal-dependent input referred offset. A novel capacitive level shifter keeps the effective input common-mode voltage constant, without attenuating the differential-mode voltage. The overall technique is independent on the operation regions of input transistors; it also does not rely on the quadratic characteristic of input MOS transistors. A prototype chip was designed for 0.35- μ m CMOS technology of 3-V supply voltage. The simulation results showed constant-gm (±0.2% variations) and constant slew rate over the entire input common-mode range.

A low power, high bandwidth continuous-time delta-sigma modulator is proposed in this paper. In contrast to traditional continuous-time delta-sigma modulators, this design utilizes passive networks, consisting of only resistors and capacitors, to perform part of the functions of loop filters. Passive networks do not consume power, introduce no distortions. For similar performance, considerable power can be saved. Based on the proposed technique, a 1.1 MHz delta-sigma modulator is designed for 0.25mum CMOS technology. Simulation results show that this modulator can reach 14b performance with only about 15mWpower consumptions