International Journal of Communication Networks and Distributed Systems (13 papers in press)

Regular Issues

• A QoS-aware task scheduling strategy for heterogeneous cloud environments via hybrid parallel PSO and deep reinforcement learning  
  by Santanu Dam, Gopa Mandal, Abir Chattopadhyay, Kousik Dasgupta 
  Abstract: Cloud computing offers a range of virtualised services to end users via its service providers or CSPs. Cloud Service Providers must ensure uninterrupted service. Simultaneous tasks arising from the Cloud of Things (CoT), Internet of Things (IoT), edge, and fog devices must be managed in real-time and within a designated timescale. A proficient scheduling strategy is essential to coordinate multiple tasks within the allotted timeframe to address this problem. The implementation of deep reinforcement learning (DRL) has significant promise and underscores its effectiveness in developing an optimised scheduling approach. This work presents a job scheduling strategy utilising a deep reinforcement learning algorithm alongside a modified particle swarm optimisation technique. This establishes a new standard for intelligent, energy-efficient, and high-performance scheduling of cloud tasks. It addresses the difficulty of attaining an ideal equilibrium between workload and enhanced accuracy while maintaining a high standard of service quality.
  Keywords: quality of service; task scheduling; cloud computing; DRL; deep reinforcement learning; CoT; Cloud of Things.
  DOI: 10.1504/IJCNDS.2027.10077676
   
  
• Hybrid dropped-weight LSTM-DQN for adaptive FQAM modulation in MIMO-filtered OFDM systems  
  by G. Shyam Kishore, P. Chandra Sekhar 
  Abstract: The proposed paper presents another AM method to multiple input multiple output filtered orthogonal frequency division multiplexing (MIMO-F-OFDM) systems, which uses the hybrid approach utilising a combination of dropped weight long short-term memory (WLSTM) and deep Q-network (DQN). The hybrid WLSTM-DQN neural architecture can select robust and dynamic modulation with a high degree of accuracy in determining signal-noise ratio (SNR) in both cases of uncertain channel conditions. To enhance further the system adaptability, a multi-layer Sparse Bayesian learning (MSBL) method, is adopted to determine imprecise information of the channel, whereas the improved pelican optimisation algorithm (IPOA) is used to select the optimal modulation mode. The different frequency and quadrature amplitude modulation (FQAM) schemes are used to evaluate the proposed approach. It is revealed through performance analysis that the performance of the hybrid model is enormously improved in terms of spectral efficiency and low bit error rate (BER), thus confirming that the hybrid model is efficient in multifaceted communications.
  Keywords: deep learning; DQN; deep Q-network; adaptive modulation; filtered orthogonal frequency division multiplexing; channel estimation; mode selection; multi input multi output.
  DOI: 10.1504/IJCNDS.2027.10078441
   
  
• A comprehensive review of clustering and routing protocols in wireless sensor networks: challenges, trends, and future directions  
  by Md. Kamaruzzaman, Abhijit Chandra 
  Abstract: Wireless sensor networks (WSNs) have emerged as a crucial technology in diverse application domains including environmental monitoring, healthcare and military surveillance etc. Due to the inherent constraints in energy and computational capacity of sensor nodes (SNs), the design of energy efficient communication protocols remains a significant research challenge. Clustering and routing algorithms play a vital role in enhancing the scalability, energy efficiency and lifetime of WSNs. This paper highlights a comprehensive review on clustering and routing algorithms. Clustering techniques are classified based on parameters such as network architecture and energy efficiency; while routing protocols are categorised into data-centric and hierarchical schemes. The paper critically evaluates the strengths, limitations of various algorithms under different network scenarios. In addition, emerging trends such as AI-ML driven approaches and mobility-aware routing have also been discussed. This paper also identifies current research gaps and outlines future directions which provide a valuable reference for researchers.
  Keywords: blockchain integrated routing; energy-aware clustering; hierarchical routing; metaheuristic optimisation; QoS routing; software defined wireless sensor networks; swarm intelligence; wireless rechargeable sensor networks.
  DOI: 10.1504/IJCNDS.2027.10079260
   
  
• Energy efficient fault-tolerant routing framework for wireless sensor networks using modified weighted practical Byzantine fault tolerance algorithm with slot scheduling technique  
  by G. Radhika , N. Radhika  
  Abstract: Heterogeneous wireless sensor networks (HWSNs) are essential for Internet of Things (IoT) applications, enabling real-time environmental monitoring through resource-constrained sensor nodes. However, limited energy resources, inefficient routing, and a lack of adequate fault tolerance compromise network performance and reliability. Hence, this study proposes a multi-stage framework encompassing clustering, routing, scheduling, and fault-tolerance mechanisms. Initially, cluster heads are selected using attribute based clustering to enhance energy efficiency. A Q-learning-based routing algorithm (QLRA) is employed for optimal path selection, while an adaptive code optimisation algorithm (ACOA) identifies suitable forwarder nodes to prevent network isolation. To extend the network’s lifespan, a Slot Scheduling Technique (SST) manages the operational states of the nodes. Furthermore, a modified weighted practical byzantine fault tolerance (MWPBFT) protocol integrated with nature-inspired fringe lizard (NIFL) optimised leader selection enhances reliability by detecting Byzantine and malicious faults. Simulation results reveal that our proposed approach attained the significant improvements compared to existing methods.
  Keywords: fault tolerance; WSNs; wireless sensor networks; MWPBFT; modified weighted practical Byzantine fault tolerance algorithm; NIFL; nature-inspired frilled lizard optimisation; SST; slot scheduling technique; duty cycle mechanism.
  DOI: 10.1504/IJCNDS.2027.10079391
   
  
• Load balancing in cloud computing: challenges, constraints, algorithms, and future research directions  
  by Ravi Gugulothu, Suneetha Bulla, Vijaya Saradhi Thommandru 
  Abstract: This survey critically reviews and classifies existing load-balancing techniques in cloud environments based on their underlying strategies, algorithmic approaches, and optimisation objectives. Furthermore, the strengths and limitations of various algorithms, including round robin, Honeybee foraging, ant colony optimisation (ACO), genetic algorithms, and machine learning-based models are discussed in terms of scalability, adaptability, and convergence behaviour. By identifying the research gaps in current techniques and emphasising the lack of standardised formulations for objective functions and constraints, this survey provides the groundwork for future research in implementing adaptive, robust, and efficient load-balancing methods in cloud computing. In the Meta reinforcement learning-based load balancing approach, the makespan is highly decreased by 19.51%, balanced CPU utilisation by 21.98% and energy consumption by 22.75%. The results effectively proves that the integration of machine learning with optimisation strategies provides more satisfactory outcomes over load balancing in cloud environment.
  Keywords: load balancing; cloud computing; resource allocation; scheduling strategies; performance metrics; optimisation algorithms.
  DOI: 10.1504/IJCNDS.2027.10079607
   
  
• Enhancing cloud data security with multi-block partitioning and hybrid encryption techniques  
  by M. P. Madhura Yadav, Sanjeev Kulkarni 
  Abstract: Cloud computing offers scalable storage for large-scale data; however, protecting sensitive information from unauthorised access in multi-tenant public cloud environments remains a major challenge. This study proposes a hybrid security model integrating multi-block partitioning (MBP) with advanced encryption standard (AES) and Blowfish algorithms to combine AES’s strong security with Blowfish’s processing efficiency. Data are segmented into smaller blocks, each sequentially encrypted by both algorithms, reducing exposure to breaches. The approach is evaluated in a local NetBeans environment using comma-separated values (CSV) datasets ranging from 10 MB to 40 MB. Key metrics, encryption/decryption time, Shannon entropy, and Index of Coincidence, are analysed. Results reveal a favourable balance between performance and randomness, achieving a Shannon entropy of 6.81 and an Index of Coincidence of 0.00246, indicating highly unpredictable ciphertext and enhanced data protection compared to conventional techniques.
  Keywords: hybrid encryption; cloud security; MBP; multi-block partitioning; AES; advanced encryption standard; blowfish algorithm.
  DOI: 10.1504/IJCNDS.2027.10079610
   
  
• Identifying the optimal size of a detector set for the recovery of sensor failure  
  by Rui Teng 
  Abstract: Identifying and recovering from query malfunctions among sensor nodes is crucial for maintaining successful data collection through a wireless sensor network (WSN). A set of supporting nodes for a sensor node aids the sensor node in query-loss recovery. This set of is called a detector set. This paper examines the optimal size of the detector set to maximise the probability of successful query recovery, taking into account of packet-loss factors at both Physical (PHY) and media access control (MAC) layers in a WSN. The optimal size of a detector set impacts both successful recovery of query loss and resources efficiency in a WSN. This study proposes a dynamic identification method to find the near-optimal size of a detector set. The evaluation results demonstrate that the proposed method can rapidly find the optimal size of the detector set.
  Keywords: sensor failure; detector set; recovery.
  DOI: 10.1504/IJCNDS.2027.10080462
   
  
• A lightweight blockchain-based key management scheme for secure VANET communication  
  by Rani Singh, Anand Rajavat 
  Abstract: Modern intelligent transportation systems (ITS) rely on vehicular ad-hoc networks (VANETs) to enable applications such as traffic monitoring, collision avoidance, and emergency alerts. Blockchain has emerged as a promising solution to improve VANET security and reliability, but conventional blockchain frameworks often introduce high computational and communication overhead. To address this, a lightweight blockchain-based key management scheme is proposed that stores only cryptographic keys, employs a simplified proof-of-work (PoW) mechanism with round-robin miner selection, and integrates Byzantine fault tolerance (BFT) to detect the compromised nodes. The scheme is evaluated in NS3 under varying vehicular scenarios by comparing a secured network using the proposed BFT-based revocation mechanism against a network with 20% malicious nodes. Simulation results show up to 94.21% throughput improvement and 23.67% packet delivery ratio (PDR) gain, along with nearly 29-30% reductions in delay and jitter, with only a slight increase in the energy consumption in some scenarios.
  Keywords: VANET; vehicular ad-hoc network; RSUs; roadside units; blockchain; key management; communication security.
  DOI: 10.1504/IJCNDS.2027.10080463
   
  
• Optimised cross-attention based convolutional recurrent neural network model for detection of cyber-attacks  
  by Bilalkhan Raufkhan Pathan, Pratik K. Patel, Hiren Raithatha 
  Abstract: The rapid progress of network-based applications requires accurate and effectual intrusion detection system to ensure network security. In this research, an optimised cross-attention based convolutional recurrent neural network (OCA_CRNN) is introduced for detecting cyber-attacks from network traffic. The model presents a hybrid semi-skipping layered gated recurrent autoencoder (HSL-GRAE), which includes a gated recurrent unit (GRU) module and an autoencoder module for extracting hidden and complex temporal features. These outcome features are collected by a novel OCA_CRNN model, which includes cross-attention, a convolutional layer, and an RNN-based bidirectional model. The improved dung Kepler optimisation is used for tuning hyperparameter details of OCA_CRNN, which improves convergence rate and performance analysis. In experimental analysis, the proposed model uses two public accessible datasets, namely NSL-KDD 99 and CIC-DDoS 2019 dataset. The proposed model achieves accuracy of 99.96% and 99.78%, which demonstrates their model capability and effectiveness.
  Keywords: IDS; intrusion detection system; cyber-attacks; cross-attention; RNN; recurrent neural network; semi-skipping layer; autoencoder; Kepler optimisation.
  DOI: 10.1504/IJCNDS.2027.10080875
   
  
• Efficient cache allocation strategy for video streaming services using hybrid-information centric networking  
  by Supratik Banerjee, Sanjay Kumar Biswash 
  Abstract: In this paper, we propose a cache allocation strategy for video streaming services that optimises the hit rate by allocating optimal storage space to each server, given a fixed storage budget at the edge nodes. Furthermore, since the video files require large storage capacity, we propose caching only the initial t seconds of each video. Thus, the client receives the initial content of the requested video from the edge node provided that the content is cached, and the rest of the content is provided by the server. Finally, less initial latency is experienced by the end-users. This leads to improved quality of experience (QoE).
  Keywords: cache allocation; HLS; HTTP-live-streaming; hybrid-ICN; hICN; ICN; information-centric networking; video streaming.
  DOI: 10.1504/IJCNDS.2027.10081038
   
  
• An improved symbiotic organisms search algorithm for conflict-free scheduling of multi-load automated guided vehicles  
  by Yongchuang Zhang, Changpu Ma, Congming Wu 
  Abstract: This study addresses conflict-free scheduling of multi-load automated guided vehicles (AGVs) in smart factories. A mixed-integer programming (MIP) model is formulated with capacity, time-window, and path-conflict constraints. To solve this NP-hard problem, an improved symbiotic organisms search algorithm (ISOSA) is proposed, integrating matrix encoding with hybrid largest rank value—earliest due date (LRV-EDD) decoding, a simulated-annealing acceptance criterion, three neighbourhood search operators, and a conflict detection and real-time resolution module ensuring feasibility throughout evolution. Experiments on 54 benchmark instances show that ISOSA achieves optimality gaps of 1.04-%3.99% vs. Gurobi on small-scale problems and significantly outperforms IPSO, IGA, and standard SOSA on medium- and large-scale instances in solution quality and stability, with convergence about 74% faster than SOSA. A real-world case study confirms practical effectiveness. This research provides a theoretical foundation and decision-support tool for conflict-free AGV scheduling in smart manufacturing logistics.
  Keywords: multi-load AGV scheduling; conflict-free routing; SOSA; symbiotic organisms search algorithm; mixed-integer programming; smart factory logistics.
  DOI: 10.1504/IJCNDS.2027.10081907
   
  
• Optimised hybrid LSTM/SVR/IOO model for accurate traffic prediction in power information communication networks with distribution technology integration  
  by Weidong Xie, Mingyang Li, Mingdong Zhang, Yan Hu, Chaochao Gao 
  Abstract: Efficient traffic prediction approaches are essential due to the increasing reliance on power information communication networks (PICNs) for electric power services. However, the diverse characteristics of accessing businesses lead to various types of traffic flows within PICNs. Traditional distributed PICNs lack flexibility and suffer from poor accuracy in traffic prediction algorithms. To overcome these challenges, an innovative approach has been proposed that combines software defined networking (SDN) architecture with hybrid support vector regression (SVR) and long short-term memory (LSTM) models, optimised by an improved variant of the osprey optimiser (IOO). Our approach utilises the SDN controller to obtain comprehensive data on the PICN within all time slots, guiding the data sampling procedure. The hybrid optimised LSTM/SVR/IOO model effectively analyses complex nonlinear relationships in large-scale network traffic, resulting in accurate traffic identification. The suggested method has been assessed employing the Mininet and POX platforms from the literature. The simulation findings show that the proposed LSTM/SVR/IOO prediction model effectively predicts the fluctuation pattern of every traffic stream, outperforming current approaches with a reduced margin of error.
  Keywords: SDN; software defined networking; PICNs; power information communication network; traffic prediction; SVR; support vector regression; LSTM; long short-term memory; IOO; improved osprey optimiser; data sampling; nonlinear relationships.
  DOI: 10.1504/IJCNDS.2027.10081952
   
  
• A lightweight hybrid CNN-LSTM framework with hybrid feature selection for cross-dataset IoT botnet detection  
  by Mir Aman Sheheryar, Sparsh Sharma 
  Abstract: With the rapid adoption of the Internet of Things (IoT), Artificial Intelligence (AI)-based cybersecurity mechanisms have become increasingly important for ensuring secure and reliable communication, as AI offers learning based approach to mitigate threats and counter attacks in IoT infrastructures. However, existing approaches frequently suffer from limited generalisation across heterogeneous datasets because of high feature redundancy and lack of modelling of spatial and temporal traffic features, resulting in suboptimal detection performance. To address these limitations, this work presents a lightweight hybrid deep learning architectural framework that incorporates convolutional neural networks (CNN) and long short-term memory (LSTM) networks for botnet detection in IoT environments based on hybrid feature selection involving information gain ratio (IGR) and Mutual Information (MI). The experimental evaluation on the UNSW-NB15 dataset achieved 93.87% accuracy for binary classification and 73.05% for multi-class classification. Cross-dataset validation on the ToN-IoT dataset demonstrated strong generalisation, achieving 99.91% binary classification accuracy and 86.84% multiclass classification accuracy in IoT botnet detection.
  Keywords: artificial intelligence; anomaly-based IDS; botnets; botnet detection; cross-dataset; IoTN; Internet of Things networks; intrusion detection.
  DOI: 10.1504/IJCNDS.2027.10081997