Leveraging CNN and Transfer Learning With EfficientNet for Enhanced Optical Coherence Tomography (OCT)
Arekanti S., Bhushan C.M., Alam I., Koppuravuri P., Shaik F., Hussain M.M., Gazi F.
Article, International Journal of Biomedical Imaging, 2026, DOI Link
View abstract ⏷
Optical coherence tomography (OCT) is an important imaging modality for detecting retinal disorders, including diabetic macular edema (DME), choroidal neovascularization (CNV), and drusen, as well as distinguishing healthy retinal patterns. Nevertheless, visual assessment of OCT scans is labor-intensive, observer-dependent, and dependent on expert clinical knowledge, encouraging reliable automated screening tools. This work introduces a hybrid Deep Transferred EfficientNet and Support Vector Machine (DTE-SVM) approach for multiclass OCT image classification. In contrast to standard CNN models that commonly use a softmax layer for final prediction, the proposed framework employs transfer-learned EfficientNet representations with an SVM decision model to strengthen interclass discrimination and robustness within the extracted feature space. Experiments used a working subset of 8767 OCT images selected from the OCT2017 dataset, representing nearly 10.4% of the complete 84,495-image collection. This smaller yet representative subset was adopted to lower computational cost while maintaining class distribution among CNV, DME, drusen, and normal groups. Five EfficientNet architectures, from B0 through B4, were assessed using stratified 10-fold cross-validation. Among them, DTE-SVM-B0 produced the strongest overall results, obtaining an accuracy of 0.9076, precision of 0.9074, recall of 0.9076, and F1-score of 0.9073. DTE-SVM-B2 showed the smallest fold-level standard deviation, reflecting comparatively stable behavior. Comparative experiments indicated that DTE-SVM-B0 surpassed ResNet50, DenseNet-121, and EfficientNetB0 softmax-based classifiers, whereas ablation testing verified that substituting the softmax layer with SVM improved results for every EfficientNet backbone. Class-level clinical evaluation also revealed strong sensitivity and specificity, with CNV identified most accurately and drusen remaining the hardest category. Overall, these results suggest that compact EfficientNet feature extractors integrated with SVM offer accurate, stable, and computationally practical OCT-driven retinal disease classification for clinical decision-support settings.
TinyML-Based AQI Forecasting Using CNN with Butterworth Low-Pass Filter for Edge Devices
Conference paper, Proceedings - International Conference on Computational Intelligence and Networks, 2026, DOI Link
View abstract ⏷
Air pollution, particularly PM2.5, is a major concern for public health in urban areas. In this paper, we propose a TinyML pipeline for PM2.5 forecasting that integrates a lightweight 1D-CNN with zero-phase Butterworth low-pass denoising to suppress sensor noise without phase distortion. Using hourly CPCB data from Perungudi (Chennai, 2024), the denoised CNN achieves strong accuracy on the normalized scale (MSE 0.000061, MAE 0.00389, RMSE 0.0078, R2=0.9908). The model is converted to TensorFlow Lite and deployed on an ARM Cortex-M4 with a 42 KB footprint, 11.4 ms inference latency, and 4,344 B RAM; measured power is 2,501 μ W. Relative to an unfiltered CNN and Multilayer Perceptron (MLP) / Feed-Forward Neural Network (FFNN) baselines, the denoising front end yields substantial error reductions while also lowering latency and memory. These results indicate that principled signal preprocessing paired with compact CNNs enables accurate, low-power, on-device P M2.5 monitoring suitable for smart-city deployments.
Toward Efficient Deployment of Compressed Neural Networks on MCU for EdgeAI Applications
Conference paper, Lecture Notes in Networks and Systems, 2026, DOI Link
View abstract ⏷
Deploying neural network models onto low-power miniaturized computers such as Micro-Controller Unit (MCU) is becoming increasingly important for edge intelligence (EdgeAI) applications due to its potential to reduce the reliance on cloud connectivity and enhance real-time processing capabilities. However, this process often requires compressing the neural network models to fit the limited memory and computational resources of MCUs while maintaining acceptable performance. In this paper, we explore various techniques for compressing neural network models, including quantization, pruning, and model optimization frameworks like TensorFlow Lite and CMSIS-NN. Additionally, it discusses the challenges associated with deploying compressed models onto MCUs, such as limited memory, computational constraints, and the need for hardware acceleration techniques. By addressing these challenges and leveraging efficient deployment strategies, deploying neural network models onto MCUs can significantly improve edge computing performance, reduce power consumption, and enhance scalability for a wide range of applications. We also consider real-time implantation of sensor nodes to capture images.
Multi-Controller TinyML Architecture for Object Detection with Efficient Communication and Applications
Bhushan C.M., Koppuravuri P., Nomitha Prasanthi B., Gazi F., Hussain M.M.
Conference paper, Proceedings - International Conference on Computational Intelligence and Networks, 2026, DOI Link
View abstract ⏷
Real-time object detection in resource-constrained environments poses significant challenges due to the limited computational power of controller units (CUs). Traditional single-controller approaches struggle to process multiple object attributes simultaneously, leading to increased inference latency and reduced efficiency. To address this, we propose a multi-controller TinyML-based architecture where two CUs operate collaboratively each identifying different parameters. Upon detection, one CU transmits its result to the second CU through UART protocol, consolidating both parameters and displaying the final classification output. The final results are then transmitted via TCP or UDP, depending on network traffic conditions, to ensure efficient data exchange. This paper also explores various real-world applications where multiple CUs collaboratively enhance object detection efficiency in resource-limited environments.
LUMI: An On-Device LLM Framework for Secure and Cloud-Free Smart Home Automation
Bai H.L., Bhushan C.M., Shaik F., Gazi F., Hussain M.
Conference paper, Proceedings - International Conference on Computational Intelligence and Networks, 2026, DOI Link
View abstract ⏷
Currently, existing voice-assisted smart homes are based on cloud-based computation, which gives rise to several concerns regarding data privacy, latency, and dependency on the network. This paper proposes a novel, secure, and efficient smart home system, namely, Localized User-initiated Machine Intelligence (LUMI), which ensures data privacy, ultra-low-latency voice interaction, and real-time device automation through local speech processing, semantic reasoning via Large Language Model (LLM), and IoT device management through a completely local architecture. Unlike existing cloud-based smart homes, in the proposed system, all computation occurs locally, ensuring data privacy, ultra-low-latency voice interaction, and real-time device automation. By employing local computation of the Mistral LLM, the proposed system can efficiently translate voice commands into device management messages through the MQTT protocol. Experimental results validate the proposed system as a novel, secure, and efficient solution, ensuring 96.7 % accuracy and 412 ms latency.
ProTSF: IoT-Based Outdoor Air Pollution Forecasting Using Bayesian Optimization-Based LSTM
Conference paper, Lecture Notes in Networks and Systems, 2026, DOI Link
View abstract ⏷
By integrating IoT devices with outdoor air quality monitoring, smart cities can proactively respond to pollution hotspots, optimize resource allocation, and implement targeted interventions to mitigate the impact of poor air quality on citizens and the environment. In this context, deep learning can empower accurate outdoor air quality forecasting by analyzing complex data patterns, enabling proactive measures for pollution mitigation and citizen well-being in smart cities. To achieve the same, we present a unique air pollution forecasting model ProTSF, which is Bayesian optimization-based deep learning model Long Short-Term Memory (LSTM), to ensure high fidelity for any dataset in this article. The assessment findings show this model’s usefulness, which points to future uses in more smart city forecasting scenarios. We attained a remarkable average accuracy of 95.83 percent through daily forecasts, with an average Root Mean Squared Error (RMSE) value of 0.089 parts per million (ppm), which is the highest as compared to other well-established time series forecasting models (e.g., AR, MA, ARIMA, SARIMA) achieved till date.
Deploying TinyML for energy-efficient object detection and communication in low-power edge AI systems
Bhushan C.M., Koppuravuri P., Prasanthi N., Gazi F., Hussain M.M., Abdussami M., Devi A.A., Faizi J.
Article, Scientific Reports, 2025, DOI Link
View abstract ⏷
Edge Artificial Intelligence (Edge AI) is driving the widespread deployment of neural network models on resource-constrained microcontroller units (MCUs), enabling real-time, on-device data processing. This approach significantly reduces cloud dependency, making it ideal for applications in industrial automation and IoT. However, the deployment of deep learning models on such constrained devices poses significant challenges due to limitations in memory, computational power, and energy capacity. This paper presents a real-time object detection system optimized for energy efficiency and scalability, which integrates well-established model compression techniques, such as quantization, with a low-cost MCU-based platform. The system leverages MobileNetV2, a lightweight neural network, quantized to achieve the best trade-offs between accuracy and resource consumption. The proposed solution integrates a camera and Wi-Fi module for capturing and transmitting image data, utilizing dual-mode TCP/UDP communication to balance reliability and low-latency transmission for IoT applications. We present a comprehensive system-level analysis, exploring the trade-offs between latency, memory, energy consumption, and model size. The Visual Wake Words (VWW) dataset is used for this research, which demonstrates the practical performance and scalability of the system for real-time applications in smart devices, industrial monitoring, and environmental sensing. This work emphasizes the integration of TinyML models with constrained hardware and offers a foundation for scalable, autonomous, energy-efficient Edge AI solutions. Quantitatively, 8-bit post-training quantization achieved 3– storage reduction, yielding deployable flash footprints of 286-536 KB within a 1 MB flash / 256 KB SRAM budget, on-device inference latency ranged from 3.47 to 14.98 ms per frame with energy per inference of 10.6–22.1 J, while quantized MobileNet variants maintained accuracy. In wireless reporting, UDP reduced one-way latency relative to TCP, whereas TCP provided higher delivery reliability, underscoring application-dependent protocol trade-offs for real-time embedded deployments.
Transparent Intrusion Detection Using ML: Integrating Explainable Models for Cybersecurity
Abdussami M., Sri Pranav N., Ramani M.S.R., Singothu K.M., Gurram S.K., Aguru A.D., Hussain M.M., Gazi F.
Conference paper, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC, 2025, DOI Link
View abstract ⏷
In today's world of connected systems, protecting those systems is imperative. Intrusion Detection Systems (IDS) have an important function to perform to protect these systems by detecting and monitoring malicious activity and policy breaches. This project develops an intelligent IDS based on Artificial Intelligence (AI) and Machine Learning (ML) to identify potential security threats online. The system applies sound ML methods to examine the traffic on a system and separate benign and malicious behavior. A variety of ML methods / Algorithms exist to differentiate bad and good data, and among those methods are supervised and unsupervised learning models to enhance known and unknown threat detection. Benchmark data sets are used to train the system so that it can detect a variety of attacks, including distributed denial of service (DDoS), probing, and remote-to-local (R2L) attacks efficiently. There is also use of feature selection and dimensionality reduction methods to achieve maximum computational performance and efficiency. The suggested AI and ML-based IDS provides scalable and flexible security through continuous learning to respond to emerging threats. It is a potent choice for securing networks of today because it supports scalability and real-time analytics.
ConFi: A Confidence-Aware Framework for Ultra-Precise Indoor Localization in 6G Networks Using UAV-Assisted Wi-Fi Fingerprints
Devi M.N.N., Bhushan Ch.M., Kumar S.H.D., Gazi F., Hussain M.M., Devi A.A.
Conference paper, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC, 2025, DOI Link
View abstract ⏷
Indoor localization is essential for various modern applications, particularly in environments where traditional GPS-based positioning is unreliable. This paper introduces the ConFi framework, a confidence-aware system that combines UAV-assisted Wi-Fi fingerprinting with machine learning models to achieve ultra-precise indoor localization. By leveraging the high-speed, low-latency capabilities of 6G networks, ConFi ensures real-time, scalable, and adaptive positioning even in highly dynamic environments. The system utilizes UAVs equipped with Wi-Fi modules to dynamically collect accurate signal data, which is processed through a convolutional neural network (CNN) optimized for local environmental variations. Reinforcement learning is integrated into the system to enable continuous adaptation, enhancing the robustness of the framework. Experimental results show that ConFi achieves a test accuracy of 92%, outperforming traditional models like LSTM and LoRaWAN. The ConFi framework is suitable for real-world applications in complex indoor environments such as hospitals, airports, and multi-storey buildings, where precise and reliable indoor localization is required.
A UAV-Assisted Architecture for Fault-Tolerant LoRa-Based V2I Communication
Siripalli H.D.K., Shaik F., Devi M.N.N., Bhushan C.M., Gazi F., Hussain M.M.
Conference paper, 2025 IEEE Future Networks World Forum: Beyond Connectivity: 6g for a Sustainable and Intelligent Future, FNWF 2025, 2025, DOI Link
View abstract ⏷
Reliable data exchange between vehicles and road-side units (RSUs) is critical for intelligent transportation systems, yet communication is often disrupted by low signal strength, intermittent connectivity, and missing acknowledgments. This paper proposes a deterministic, UAV-assisted architecture for fault-tolerant V2I communication using LoRa and IEEE 802.11 backhaul links. In the proposed system, UAVs are triggered when RSUs detect degraded RSSI or ACK loss, thereby providing a redundant relay path to maintain session continuity. All logic is implemented on off-the-shelf embedded hardware without reliance on cloud or simulation platforms. Experimental results from real hardware deployments demonstrate a 14% improvement in message delivery ratio, steady latency of approximately 34 ms, and enhanced SNR performance in low-coverage areas. The architecture is lightweight, interpretable, and immediately deployable for real-world ITS applications.
A Blockchain-based DDoS Attack Mitigation Framework for Mission-critical IIoT Environments
Conference paper, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC, 2025, DOI Link
View abstract ⏷
The Mission-critical IIoT (MC-IIoT) networks are highly interconnected and distributed to ensure real-time monitoring and control, where security, reliability, and availability are paramount. These features make MC-IIoT networks vulnerable to Distributed Denial of Service (DDoS) attacks that must be mitigated while combating single-point of failure of attack detection systems, limited scalability of industrial control systems, high latency and low reliability. In this paper, we propose a Blockchain-based DDoS attack mitigation framework for MC-IIoT networks which persuades these requirements. Initially, a machine learning-based cascaded model consisting of an autoencoder-enhanced decision tree model and a lightweight isolation forest algorithm performs attack detection and trust score evaluation of IIoT devices. A consortium Blockchain network of IIoT gateways on which this cascaded model is installed performs attack mitigation using a chaincode based on previously evaluated trust scores. The proposed model is evaluated on Hyperledger Fabric and the performance is analyzed with Hyperledger caliper benchmarking. The comparative analysis of the proposed framework with state-of-the-art proves its effectiveness in terms of attack detection accuracy (99.93%), CPU utilization (30%), decreased latency (12.5 ms), increased throughput (21%), chaincode execution time (3 ms) and attack detection rate (99.6%).
Dependency-Aware Microservice Placement in Dynamic Vehicular Edge Computing
Conference paper, 2025 IEEE Future Networks World Forum: Beyond Connectivity: 6g for a Sustainable and Intelligent Future, FNWF 2025, 2025, DOI Link
View abstract ⏷
Vehicular Edge Computing (VEC) is a promising paradigm to support latency-sensitive and computation-intensive vehicular applications such as object detection, trajectory prediction, and hazard alerts. Microservice architectures enable scalability and modularity, but placing dependent microservices across heterogeneous edge service providers (ESPs) remains a critical challenge. Existing placement strategies often treat services independently, ignoring inter-service dependencies, which leads to increased inter-node communication delays and degraded performance. In this paper, we propose a dependency-aware microservice placement framework for VEC using deep reinforcement learning. Our approach leverages a Proximal Policy Optimization (PPO) based actor-critic agent to learn placement decisions that jointly minimize end-to-end service latency and maximize resource utilization while respecting capacity and SLA constraints. Experiments on real vehicular mobility traces from Luxembourg City demonstrate that the proposed method significantly reduces service latency and SLA violations compared to random and greedy baselines, while efficiently utilizing edge resources.
Towards Efficient IoUT Vision: Simulated Compressed Image Transmission and Detection
Conference paper, 2025 IEEE Future Networks World Forum: Beyond Connectivity: 6g for a Sustainable and Intelligent Future, FNWF 2025, 2025, DOI Link
View abstract ⏷
Underwater communication is constrained by limited bandwidth, high latency, and signal attenuation, particularly in acoustic transmission scenarios. To address these challenges, this paper investigates a simulation-based framework for underwater image transmission using discrete cosine transform (DCT) and quantization-based compression. Grayscale images are compressed at varying quantization levels (3-bit to 8-bit) and transmitted over a simulated underwater acoustic channel with relay-based attenuation and noise. The impact of compression on image fidelity is evaluated using Peak Signal-to-Noise Ratio and Structural Similarity Index (PSNR, SSIM), while object detection robustness is assessed via classification accuracy using a MobileNetV2 model. Experimental results demonstrate a clear trade-off between compression efficiency and detection accuracy, offering insights into optimal quantization strategies for bandwidth-limited underwater environments. This simulation framework, developed entirely in MATLAB, provides a scalable testbed for evaluating vision-centric Internet of Underwater Things (IoUT) applications prior to real-world deployment on resource-constrained devices.
OUL-HMT: Optimized AAV Localization Using Hybrid Metaheuristic Techniques
Article, IEEE Journal on Indoor and Seamless Positioning and Navigation, 2025, DOI Link
View abstract ⏷
Achieving an exact localization is a complex and essential issue for autonomous aerial vehicles (AAVs) due to their three-directional high-speed mobility. Identifying the accurate flying position of AAVs for resource management and task reallocation is still challenging. In these scenarios, the position of the AAVs must be identifiable in a timely and precise manner. A bioinspired metaheuristic hybrid model was proposed to overcome the shortcomings of inaccurate altitude and improve the AAVs' flying positional coordinates. The proposed model incorporates the particle swarm optimization (PSO) with a fuzzy logic technique. PSO is used to find the optimal or near-optimal positions for the AAVs by minimizing localization error across a wide search space. Once the PSO has determined a feasible solution, fuzzy logic is applied for fine tuning the position based on real-time environmental factors (e.g., signal strength, sensor data, or global positioning system errors). This combination achieved both global efficiency (via PSO) and local precision (via fuzzy logic), ensuring robust localization even in noisy or dynamic conditions during AAVs flight operations. The model, compared to the state-of-the-art model, shows more accuracy in AAV localization with real-time operational data.
Integrated Underwater Data Transmission and Object Detection System Using TinyML and Multi-Hop Networks
Bhushan C.M., Laskar S.H., Garikapati J.S., Sai Srihitha P., Hemanth Durga Kumar S., Gazi F., Hussain M.M.
Conference paper, 2025 21st International Conference on Intelligent Environments, IE 2025, 2025, DOI Link
View abstract ⏷
Underwater communication faces challenges like high attenuation, limited bandwidth, and energy constraints. This paper presents an underwater communication system using ultrasonic sensors for image transmission and TinyML for efficient object detection. The architecture comprises a Raspberry Pi-based transmitter node, intermediate repeater nodes and a receiver node. Images are processed using Discrete Cosine Transform (DCT), transmitted as text files, reconstructed via Inverse DCT (IDCT), and analyzed using a lightweight MobileNetV2 model for real-time object detection. The integration of TinyML enables energy-efficient on-device inference, addressing resource constraints in edge devices. The system demonstrates effective data transmission and accurate detection, with applications in underwater surveillance, marine monitoring, and aquaculture. This work underscores TinyML's role in advancing AI-driven Internet of Underwater Things (IoUT) technologies.
Leveraging Edge Resources for Indoor Localization for Improved Accuracy
Conference paper, 2025 21st International Conference on Intelligent Environments, IE 2025, 2025, DOI Link
View abstract ⏷
This paper presents a novel approach to zone prediction through Wi-Fi fingerprinting combined with machine learning, leveraging data collected by a UAV and a robotic car across distinct zones. Among the evaluated models, the Confidence-Aware Framework (ConFi) outperformed state-of-the-art methods, achieving a test accuracy of 91%, compared to Long Short-Term Memory (LSTM) at 73% and Long Range Wide Area Network (LoRaWAN) at 85%. This superior accuracy underscores the ConFi model's capability to effectively manage the complexities of real-world environments. By leveraging the ConFi framework, the proposed system enhances precision, scalability, and adaptability over traditional methods. This study represents a significant advancement in indoor localization, offering a deployable and efficient solution for GPS-denied environments with promising applications in healthcare, logistics, and disaster management.
Resource management in fog computing: Overview and mathematical foundation
Book chapter, Swarm Intelligence: Theory and Applications in Fog Computing, Beyond 5G Networks, and Information Security, 2025,
Evolutionary Algorithms for Edge Server Placement in Vehicular Edge Computing
Article, IEEE Access, 2025, DOI Link
View abstract ⏷
Vehicular Edge Computing (VEC) is a critical enabler for intelligent transportation systems (ITS). It provides low-latency and energy-efficient services by offloading computation to the network edge. Effective edge server placement is essential for optimizing system performance, particularly in dynamic vehicular environments characterized by mobility and variability. The Edge Server Placement Problem (ESPP) addresses the challenge of minimizing latency and energy consumption while ensuring scalability and adaptability in real-world scenarios. This paper proposes a framework to solve the ESPP using real-world vehicular mobility traces to simulate realistic conditions. To achieve optimal server placement, we evaluate the effectiveness of several advanced evolutionary algorithms. These include the Genetic Algorithm (GA), Non-dominated Sorting Genetic Algorithm II (NSGA-II), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), and Teaching-Learning-Based Optimization (TLBO). Each algorithm is analyzed for its ability to optimize multiple objectives under varying network conditions. Our results show that ACO performs the best, producing well-distributed pareto-optimal solutions and balancing trade-offs effectively. GA and PSO exhibit faster convergence and better energy efficiency, making them suitable for scenarios requiring rapid decisions. The proposed framework is validated through extensive simulations and compared with state-of-the-art methods. It consistently outperforms them in reducing latency and energy consumption. This study provides actionable insights into algorithm selection and deployment strategies for VEC, addressing mobility, scalability, and resource optimization challenges. The findings contribute to the development of robust, scalable VEC infrastructures, enabling the efficient implementation of next-generation ITS applications.
Facility Location in 6G-aware Vehicular Edge Computing
Conference paper, International Symposium on Advanced Networks and Telecommunication Systems, ANTS, 2024, DOI Link
View abstract ⏷
This paper tackles the facility location problem in 6G -enabled Vehicular Edge Computing (VEC) systems, focusing on the optimal placement of Roadside Units (RSUs) and Unmanned Aerial Vehicles (UAVs). The goal is to minimize Quality of Service (QoS) degradation by addressing challenges like dynamic vehicle mobility, traffic variations, and real-time task offloading. A mathematical optimization model is proposed, considering latency, energy consumption, packet loss, and handover costs. To solve this complex problem, heuristic algorithms such as Hill Climbing, Tabu Search, Simulated Annealing, and A∗ search are introduced. Extensive simulations evaluate their performance on energy efficiency and cumulative latency across various traffic and network conditions. The results reveal the strengths of each algorithm, offering valuable insights for their application in VEC scenarios. These findings contribute to scalable, energy-efficient solutions for 6G-aware VEC networks, particularly in dynamic vehicular environments, advancing research in edge computing and network optimization.
Design and Fabrication of an IoT based Air Purifier using HEPA Filter
Choudhary A., Saini L., Ahmad A., Banerjee H., Gazi F.
Conference paper, 11th International Conference on Internet of Everything, Microwave Engineering, Communication and Networks, IEMECON 2023, 2023, DOI Link
View abstract ⏷
Air pollution has become one of the major problems for human health across the globe. Indoor air pollution poses more risk than outdoor air pollution, since the human body is more exposed to the inside. It refers to the contamination of indoor air that causes harmful health problems. To purify the polluted air, the air purifier is a main necessity. In this Internet of Things (IoT) project, we have created an air purifier that purifies the indoor air by passing it through the High Efficiency Particulate Air (HEPA) filter and Ultraviolet light. The microcontroller used is the NodeMCU based on the ESP8266 WiFi enabled chip. Two MQ135 air quality sensors used to monitor the change and calculate the efficiency of the purified air. IoT customization eases the manual work of changing the purifier manually every time by providing a platform that operates on the mobile phone via Wi-Fi where the user can operate it accordingly. It provides an equivalent grade of services at a very moderate cost that is affordable even for the average person and can provide real time analytics. The proposed system is capable for providing hands-on practical operations, which enables more exciting possibilities in IoT based automation systems.
RE-MAC: A Hybrid MAC Protocol for Underwater Multimedia Communication System
Gazi F., Ahmed N., Misra S.
Article, IEEE Systems Journal, 2023, DOI Link
View abstract ⏷
In this article, we propose a reliable and efficient medium access control (MAC) protocol, RE-MAC, for resource-constraint underwater acoustic multimedia communication systems. RE-MAC achieves scalability by utilizing a hybrid MAC protocol smartly, which takes advantage of contention, reservation, and null data packet (NDP) based channel access mechanisms. The proposed scheme achieves multihop communication and synchronization for video streaming using a transmission opportunity (TXOP) based efficient channel access mechanism at the access point (AP) and relay nodes. TXOP time has been utilized by relay and AP for sending stream of frames with low overhead. To improve the efficiency in multihop communication, RE-MAC finds an optimal frame duration for relays and AP. RE-MAC efficiently places the NDP mechanism replacing traditional ACK-based control packet exchange. We incorporate the TXOP in AP and relay nodes for multimedia streaming and sleep mode for stations. Further, the video compression technique gives the efficiency of our system to improve the transmission of multimedia data. Our proposed schemes help communicate multimedia data without compromising the quality of service. We have demonstrated our proposed scheme experimentally in a prototype. After that, a simulation-based detailed performance analysis is provided to see the scalability of the proposed scheme. Compared to the existing state-of-the-art solutions, RE-MAC significantly delivers the throughput, energy consumption, delay, and packet loss.
AquaStream: Multihop Multimedia Streaming over Acoustic Channel in Severely Resource-Constrained IoT Networks
Mukherjee A., Gazi F., Pathak N., Misra S.
Article, IEEE Internet of Things Journal, 2022, DOI Link
View abstract ⏷
Robust and reliable communication systems, whether based on electromagnetic (EM) waves or light, fail to perform under water due to very high attenuation and changing visibility conditions. The present generation of systems designed for underwater communications relies mostly on acoustic waves, typically in the ultrasonic frequency range. In this work, we develop and evaluate a means of implementing underwater acoustic channel-based severely constrained IoT networks using low-cost, off-The-shelf, open hardware electronics and transducers, which can support direct communication between two nodes at a data rate of 2.4 kbps for over 65 m. These nodes can be deployed over much longer distances through multihop relay topologies. Furthermore, we evaluate the efficacy of our system toward supporting multimedia data transmission and even attempt multimedia streaming through our deployed underwater IoT network using video compression and reduced sampling of the video frames. We observe that the system successfully supports multihop network topologies and undertakes multimedia transmission by compromising the quality of the data. The system has a clear tradeoff between data quality, transmission range, and transmission delays.
ProStream: Programmable Underwater IoT Network for Multimedia Streaming
Gazi F., Ahmed N., Misra S., Tiwari M.K.
Article, IEEE Internet of Things Journal, 2022, DOI Link
View abstract ⏷
Underwater Internet of Things (IoT) faces challenges, such as low data rate, high node mobility, high error probability, and propagation delay. Furthermore, underwater multimedia communication is more challenging, as it requires a high data rate and reliable delivery. The next-generation networking paradigm - software-defined network (SDN) - improves flexibility and virtualization through programmability. SDN increases resource utilization, simplifies management, and reduces operating costs. This article proposes ProStream, a reconfigurable SDN-based underwater multihop network for improved topology management, association control, and multimedia transmission. The proposed scheme considers a network with multiple access points (APs) and mobile relays and stations. The SDN controller places flow rules in the relays, and APs are placed as per the location of the nodes. Moreover, we present a programmable station for reducing Tx/Rx complexity in underwater networks. The stations or relays are triggered through programmability to notify their sleep time and current relays and AP for association and data transmission. We evaluate the proposed scheme first through simulation-based experimentation. We also develop a small-scale real testbed prototype of the proposed SDN. The performance results show significant improvement of performance in terms of latency, throughput, and packet delivery ratio.
Reinforcement Learning-Based MAC Protocol for Underwater Multimedia Sensor Networks
Gazi F., Ahmed N., Misra S., Wei W.
Article, ACM Transactions on Sensor Networks, 2022, DOI Link
View abstract ⏷
High propagation delay, high error probability, floating node mobility, and low data rates are the key challenges for Underwater Wireless Multimedia Sensor Networks (UMWSNs). In this article, we propose RL-MAC, a Reinforcement Learning (RL)-based Medium Access Control (MAC) protocol for multimedia sensing in an Underwater Acoustic Network (UAN) environment. The proposed scheme uses Transmission Opportunity (TXOP) for relay nodes in a multi-hop network for improved efficiency concerning the mobility of the relays and sensor nodes. The access point (AP) and relay nodes calculate traffic demands from the initial contention of the sensor nodes. Our solution uses Q-learning to enhance the contention mechanism at the initial phase of multimedia transmission. Based on the traffic demands, RL-MAC allocates TXOP duration for the uplink multimedia reception. Further, the Structural Similarity Index Measure (SSIM) and compression techniques are used for calculating the image quality at the receiver end and reducing the image at the destination, respectively. We implement a prototype of the proposed scheme over an off-the-shelf, low-cost hardware setup. Moreover, extensive simulation over NS-3 shows a significant packet delivery ratio and throughput compared with the existing state-of-the-art.
Measuring Real-Time Road Traffic Queue Length: A Reliable Approach Using Ultrasonic Sensor
Mandal A., Sadhukhan P., Gaji F., Sharma P.
Conference paper, Lecture Notes in Electrical Engineering, 2020, DOI Link
View abstract ⏷
Traffic congestion not only lengthens the travel time of a commuter but also increases the fuel consumption as well as air pollution. Thus, traffic congestion management has drawn significant research interest over the past few years. On the other hand, a reliable estimation of the congestion density via traffic queue length measurement at any segment of the road in real time is highly desirable to control the traffic congestion in the urban areas. Thus in this paper, we focus on how to estimate the congestion density in a reliable way and propose a real-time traffic queue length measuring approach for the signalized intersections using ultrasonic sensor node (USN). The experimental results demonstrate that our proposed approach achieves almost 100 success in vehicle detection if the vehicle queue is formed within a distance of 2.5 m from the position of USN. Moreover, our proposed approach also attains high success rate in the detection of vehicle queue even when USN is placed 2 m away from the vehicle queue and it lies in between two vehicles by having the intervehicle distance falls within the range of 30–80 cm.
UnRest: Underwater reliable acoustic communication for multimedia streaming
Gazi F., Misra S., Ahmed N., Mukherjee A., Kumar N.
Conference paper, Proceedings - IEEE Global Communications Conference, GLOBECOM, 2020, DOI Link
View abstract ⏷
Due to the low data-rate, high propagation delay, floating node mobility, and high error probability, underwater multimedia communication is still challenging. In this paper, we propose an acoustic-based reliable streaming network for resource-constrained underwater communication. The proposed protocol uses a Null Data Packet (NDP)-based contention and acknowledgment mechanism to reduce control overhead and improve reliability and energy efficiency. With the use of a lightweight Traffic Indication Map (TIM) and video compression technique, our system's efficiency for multimedia transmission is further improved. The proposed schemes provide multimedia communication without compromising on the quality of the data transmitted. We experimentally demonstrate the proposed scheme in a hydrodynamic water-tank facility on our campus. The testbed we have built in this facility is capable of running real-time video streaming. The system's performance, which is evaluated using parameters such as coverage, range, latency, and energy consumption, was found to prove the proposed solution's validity.
An IoT based intelligent traffic congestion control system for road crossings
Conference paper, Proceedings of the 2018 International Conference On Communication, Computing and Internet of Things, IC3IoT 2018, 2018, DOI Link
View abstract ⏷
Traffic congestion is one of the major issues with the public transportation system in recent time. The traffic congestion has a negative impact on the productivity, competitiveness and economic growth of a country. Hence traffic congestion control has become an important area of research and significant number of solutions to this problem came out of various research efforts in the said field over the past few decades. Among these, vehicle-to-vehicle (V2V) communication based approaches cannot accurately estimate the density of traffic congestion. On the other hand, the traffic signaling systems having predetermined fixed operation time cannot manage the traffic volume changing over time and thus, long traffic queues are generated at the road crossings. To address the above mentioned issue, this paper proposes an internet-of-things (IoT) based intelligent traffic congestion control system that dynamically sets the signal operation time based on the measured values of traffic congestion density. Moreover, a novel technique of measuring the density of traffic congestion created at the road crossings is also presented in this paper.
Towards smart city: Sensing air quality in city based on opportunistic crowd-sensing
Dutta J., Chowdhury C., Roy S., Middya A.I., Gazi F.
Conference paper, ACM International Conference Proceeding Series, 2017, DOI Link
View abstract ⏷
Cities are expanding and more and more citizens are exposed to air pollutants both indoors and outdoors. This may have adverse effects on citizens' health. In this paper, we present AirSense, an opportunistic crowd-sensing based air quality monitoring system, aimed at collecting and aggregating sensor data to monitor air pollution in the vicinity (building/neighbourhood) and the city. We introduce a light weight, low power and low cost air quality monitoring device (AQMD) and demonstrate how AQMD and smartphones in a crowd collaboratively gather and share data of interest to the cloud. In cloud, collected data are analyzed and an aggregate view is generated from data collected from various sensors and from different users for providing an air pollution heat map of the city. Unlike previous works, both micro and macro level air quality monitoring is possible with Airsense. End user can view his/her pollution footprint for the whole day, the neighborhood (local) air quality and AQImap (air quality index map) of the city on his/her smartphone. The system is implemented and the prototype is also evaluated.
AirSense: Opportunistic crowd-sensing based air quality monitoring system for smart city
Dutta J., Gazi F., Roy S., Chowdhury C.
Conference paper, Proceedings of IEEE Sensors, 2016, DOI Link
View abstract ⏷
Citizens are exposed to air pollutants both indoors and outdoors due to their activities, which may result in a variety of health effects. In this paper, we present AirSense, an opportunistic crowd-sensing based air quality monitoring system, aimed at collecting and aggregating sensor data to monitor air pollution in the vicinity and the city. We introduce a light weight, low power and low cost air quality monitoring device (AQMD) and demonstrate how AQMD and smart phone devices in a crowd collaboratively (through offloading) gather and share data of interest to the cloud. In cloud, collected data will be analyzed and an aggregated view will be generated for providing an air pollution heat map of the city. End user can view both the neighborhood (local) air quality and AQImap (air quality index map) of the city on his/her smartphone.
