Federated Learning–enhanced Weighted Fair Queuing for Adaptive Resource Allocation in Wireless Sensor Networks International Journal of Intelligent Engineering and Systems, 2026 Wireless sensor nodes operate with very limited memory, battery power, and processing capability, which makes it difficult for them to handle complex tasks independently.Edge computing helps overcome this limitation by offloading computation to nearby edge servers with higher storage, bandwidth, and processing capacity.However, efficiently managing and allocating these edge resources is challenging due to dynamic traffic patterns, topology changes, and node mobility in wireless networks.Fixed allocation methods often fail under such conditions.To address this, a dynamic queue allocation mechanism based on federated learning (FL) is proposed.FL enables multiple edge nodes to collaboratively train a global model using local traffic information without sharing raw data.This allows the system to adaptively predict and allocate packet queues based on real-time network conditions.Experimental results show that FL-WFQ improves efficiency by about 11% over classical WFQ and fairness by about 5%, while achieving significantly higher gains compared to WRR and FIFO baselines.
Mobility-Aware Resource Allocation in Edge Networks for Smart Neighborhoods Sravya Pallantla, D. Haritha, Shanti Chilukuri Ssrg International Journal of Electronics and Communication Engineering, 2025 Wireless Sensor Networks (WSNs) comprise sensor and actuator nodes that either generate or consume data. These end devices typically have limited memory, energy, and computational capacity, making them ill-suited for intensive processing tasks. Edge computing addresses this limitation by offloading data collection, processing, and forwarding to edge nodes with greater resources. This architecture aligns well with WSNs, but efficient allocation of edge resources remains a significant challenge, particularly under node mobility and fluctuating traffic. The resource allocation problem at the edge is NP-hard, and while static solutions exist, they often fail under dynamic network conditions. Memory management for packet queues is especially critical. The scheduling and dropping policies at edge nodes directly influence Quality of Service (QoS). Weighted Fair Queuing (WFQ), a popular scheduling method, assigns different weights to traffic classes, affecting uplink bandwidth distribution. However, bursty data, varying packet sizes, and node mobility complicate fair and efficient weight assignment. This study proposes a federated learning-based framework for dynamic queue and bandwidth allocation in mobile WSNs. The model adapts to real-time changes in traffic and topology while reducing communication overhead. Simulation outcomes demonstrate improved resource utilization and network performance, validating the effectiveness of the proposed approach in dynamic WSN environments.
SmartHART: A Priority-aware Scheduling and Routing Scheme for IIoT Networks using Deep Reinforcement Learning Shanti Chilukuri, Aditya Gupta, Hemanth Sri Sai Pulamolu 2023 15th International Conference on Communication Systems and Networks Comsnets 2023, 2023 The aims of the Industrial Internet of Things (IIoT) are improved efficiency of production facilities and operations in factories and better logistics and supply chain management. Communication of data in the IIoT can be challenging due to diverse and critical demands on the Quality of Service (QoS). Medium Access Control (MAC) in such networks is typically using Time Division Multiple Access as recommended by stan-dards such as the WirelessHART. In this paper, we propose and evaluate a Deep Reinforcement Learning (DRL)-based scheduling and routing scheme (called SmartHART) for WirelessHART networks. SmartHART leverages on redundancy in the network and chooses schedules and routes that minimize the end-to-end delay of data, taking priority of data into consideration. The reward design of the SmartHART agent also minimizes the maximum packet queue length in a network, making it suitable for memory-constrained field devices. Simulation results show that SmartHART can reduce the end-to-end delay by as much as 18% and the maximum packet queue size by up to 60%.
Deadline-Aware TDMA Scheduling for Multihop Networks Using Reinforcement Learning Shanti Chilukuri, Guangyuan Piao, Diego Lugones, Dirk Pesch 2021 IFIP Networking Conference IFIP Networking 2021, 2021 Time division multiple access (TDMA) is the medium access control strategy of choice for multihop networks with deterministic delay guarantee requirements. As such, many Internet of Things applications use protocols based on time division multiple access. Optimal slot assignment in such networks is NP-hard when there are strict deadline requirements and is generally done using heuristics that give suboptimal transmission schedules in linear time. However, existing heuristics make a scheduling decision at each time slot based on the same criterion without considering its effect on subsequent network states or scheduling actions. Here, we first identify a set of node features that capture the information necessary for network state representation to aid building schedules using Reinforcement Learning (RL). We then propose three different centralized approaches to RL-based TDMA scheduling that vary in training and network representation methods. Using RL allows applying diverse criteria at different time slots while considering the effect of a scheduling action on meeting the scheduling objective for the entire TDMA frame, resulting in better schedules. We compare the three proposed schemes in terms of how well they meet the scheduling objectives and their applicability to networks with memory and time constraints. One of the schemes proposed is RLSchedule, which is particularly suited to constrained networks. Simulation results for a variety of network scenarios show that RLSchedule reduces the percentage of packets missing deadlines by up to 60% compared to the best available baseline heuristic.
NimbleCache - Low cost, dynamic cache allocation in constrained edge environments Shanti Chilukuri, Dirk Pesch IEEE Wireless Communications and Networking Conference Wcnc, 2021 Edge computing and caching of data in the Internet of Things (IoT) has several benefits such as reduced energy consumption by IoT end devices and increased availability of data and Quality of Service (QoS). In typical IoT scenarios, edge nodes (gateways) support several end devices, each of which may produce data in different patterns. In addition, data generated by different types of end devices varies in the application QoS requirements while also widely varying in the data access patterns by IoT services. Managing the data storage resources at edge nodes in such scenarios is a difficult task, especially since the edge nodes themselves may have limited computation capability and storage space. In this paper, we propose a dynamic, differentiated edge cache allocation strategy called NimbleCache that has low computational requirements and performs efficient cache allocation at edge nodes. Based on a Mixture Density Network (MDN), NimbleCache allocates varying portions of the edge cache to traffic of different IoT applications to achieve cache hit ratios very close to the target hit ratio. Simulation results show that NimbleCache achieves good average cache hit ratio with low cache space requirement and small computational overhead.
RECCE: Deep Reinforcement Learning for Joint Routing and Scheduling in Time-Constrained Wireless Networks Shanti Chilukuri, Dirk Pesch IEEE Access, 2021 Time Division Multiple Access-based Medium Access Control protocols tend to be the choice for wireless networks that require deterministic delay guarantees, as is the case in many Industrial Internet of Things (IIoT) applications. As the optimal joint scheduling and routing problem for multi-hop wireless networks is NP-hard, heuristics are generally used for building schedules. However, heuristics normally result in sub-optimal schedules, which may result in packets missing their deadlines. In this paper, we present RECCE, a deep REinforcement learning method for joint routing and sCheduling in time-ConstrainEd networks with centralised control. During training, RECCE considers multiple routes and criteria for scheduling in any given time slot and channel in a multi-channel, multi-hop wireless network. This allows RECCE to explore and learn routes and schedules to deliver more packets within the deadline. Simulation results show that RECCE can reduce the number of packets missing the deadline by as much as 55% and increase schedulability by up to 30%, both relative to the best baseline heuristic. RECCE can deal well with dynamic network conditions, performing better than the best baseline heuristic in up to 74% of the scenarios in the training set and in up to 64% of scenarios not in the training set.
On the convergence and optimality of the firefly algorithm for opportunistic spectrum access Shanti Chilukuri, Sireesha Rodda, Lakshmana Rao Kalabarige International Journal of Advanced Intelligence Paradigms, 2021 Meta-heuristic algorithms have been proven to be efficient for engineering optimisation. However, the convergence and accuracy of such algorithms depends on the objective function and also on several choices made during algorithm design. In this paper, we focus on the firefly algorithm for optimal channel allocation in cognitive radio networks. We study the effect of various probability distributions including the Levy alpha stable distribution for randomisation of firefly movement. We also explore various functions for converting firefly positions from the continuous space to the discrete space, as is necessary in the spectrum allocation problem. Simulation results show that in most cases, Levy flight gives better convergence time and results for common optimisation problems such as maximising the overall channel utilisation, maximising the channel allocation for the bottleneck user and maximising proportional fairness. We also note that no single discretisation function gives both good convergence and optimality.
Achieving Optimal Cache Utility in Constrained Wireless Networks through Federated Learning Shanti Chilukuri, Dirk Pesch Proceedings 21st IEEE International Symposium on A World of Wireless Mobile and Multimedia Networks Wowmom 2020, 2020 Edge computing allows constrained end devices in wireless networks to offioad heavy computing tasks or data storage when local resources are insufficient. Edge nodes can provide resources such as the bandwidth, storage and innetwork compute power. For example, edge nodes can provide data caches to which constrained end devices can off-load their data and from where user can access data more effectively. However, fair allocation of these resources to competing end devices and data classes while providing good Quality of Service is a challenging task, due to frequently changing network topology and/or traffic conditions. In this paper, we present Federated learning-based dynamic Cache allocation (FedCache) for edge caches in dynamic, constrained networks. FedCache uses federated learning to learn the benefit of a particular cache allocation with low communication overhead. Edge nodes learn locally to adapt to different network conditions and collaboratively share this knowledge so as to avoid having to transmit all data to a single location. Through this federated learning approach, nodes can find resource allocations that result in maximum fairness or efficiency in terms of the cache hit ratio for a given network state. Simulation results show that cache resource allocation using FedCache results in optimal fairness or efficiency of utility for different classes of data when compared to proportional allocation, while incurring low communication overhead.
Adaptive Differentiated Edge Caching with Machine Learning for V2X Communication Vinayaka Shashank Varanasi, Shanti Chilukuri 2019 11th International Conference on Communication Systems and Networks Comsnets 2019, 2019 Connected vehicles that communicate with the traffic network around them have several uses in providing road safety and infotainment. Such applications leverage on Vehicle-to-Anything (V2X) communication, which is challenging because of rapidly changing topology and traffic patterns. We propose a differentiated edge caching scheme called FlexiCache for such networks. In FlexiCache, the cache is split into sections to hold data of different classes with suitable replacement policies. Further, FlexiCache uses kernel ridge regression (KRR) to predict the proportion of cache to be allocated to each traffic type, for a desired quality of service(QoS) parameter. It then uses a self-learning mechanism that adapts cache allocation to the network conditions. Simulation results show that FlexiCache performs better than undifferentiated caching and also that the predictions by KRR result in QoS which is very close to the target value.
T-move: A light-weight protocol for improved QoS in content-centric networks with producer mobility Swaroopa Korla, Shanti Chilukuri Future Internet, 2019 Recent interest in applications where content is of primary interest has triggered the exploration of a variety of protocols and algorithms. For such networks that are information-centric, architectures such as the Content-Centric Networking have been proven to result in good network performance. However, such architectures are still evolving to cater for application-specific requirements. This paper proposes T-Move, a light-weight solution for producer mobility and caching at the edge that is especially suitable for content-centric networks with mobile content producers. T-Move introduces a novel concept called trendiness of data for Content-Centric Networking (CCN)/Named Data Networking (NDN)-based networks. It enhances network performance and quality of service (QoS) using two strategies—cache replacement and proactive content-pushing for handling producer mobility—both based on trendiness. It uses simple operations and smaller control message overhead and is suitable for networks where the response needs to be quick. Simulation results using ndnSIM show reduced traffic, content retrieval time, and increased cache hit ratio with T-Move, when compared to MAP-Me and plain NDN for networks of different sizes and mobility rates.
The cloudlet with a silver lining Sainath Kommineni, Aneesh De, Sashank Alladi, Shanti Chilukuri 2014 6th International Conference on Communication Systems and Networks Comsnets 2014, 2014
