International Journal of Digital Technologies

DeTR-OCR: Combining DETR and Transformer-Based OCR for Real-Time Video Caption Extraction

Abhishek Kumar Mishra; Kapil Kumar; Sanjeev Bhardwaj — 2026-10-16

The rapid production of video content across digital platforms requires strong systems for extracting textualslogans in real time video.This paper represents DeTR-OCRi.e a novel framework that integrates DetectionTransformer(DETR) and Transformerbased Optical Character Recognition (TrOCR) to focuses on these difficulties. utilizing DETR’s set forecast pa-radigm and TrOCR’s recognition ability to DeTR-OCR accomplishes end-to-end caption extraction from video streams. ICDAR 2017 Multi-Lingual Scene Text (MLT) and News VideosText datasets are used to test accuracy of proposed model. The system yields a mean Average Precision (mAP) of 0.82 and Character Error Rate (CER) of 0.09 on ICDARtogether with 28 frames per second (FPS) for realtime achievement. Provisional analyses expose superiority over baselines like as Efficient and Accurate Scene Text Detector (EAST) and Tesseract,while destruction studies underscore the effectiveness of flexible frame sampling. Regardless of challenges with lowcontrast text, DeTR-OCR suggestions a promising outcome for live sub-titling and accessibility with concrete way for future improvements in video text processing.

A Comparative Study of Indoor Kitchen Air Quality and Outdoor Ambient Air Quality

2026-10-16

The Cooking activity makes a substantial contribution to indoor pollution. Indoor air pollution (IAQ) has become a significant environmental and public health concern in recent years in countries like India, where 48.5% of the Indian population i.e. women cooks daily. In this paper an Internet of Things (IoT) based model is developed to compare the air quality outside and inside the kitchen while cooking. The proposed model measures the level ofkey harmful gases like Carbon Monoxide (CO), Carbon Dioxide (CO) insidethe kitchen while cooking and outside the building.The proposed model comprises ESP8266 microcontroller, MQ series gas sensors, Liquid Crystal Display (LCD) and jumper wires. The collected data reveals that there is a significant impact of cooking on the key pollutants gases, and there is a spike in the level of gas concentration while cooking inside the kitchen.

Reinforcement Learning-Based Framework for Cybersecurity in Online Examination Systems

Anurag Shah;, Manoj Kumar; Sarika Jain — 2026-10-16

As cyber security threats are becoming more and more complex, sophisticated and persistent, the methods of tradi-tional defence mechanism are becoming obsolete and are struggling hard in keeping pace cost-effectively to preventing the black swan [1] threats of the todays cyberworld, that are poised with the advent of Artificial Intelligence, hence need is being felt for the adaptive capabilities of the cybersecurity tools along with the dynamic decision-making abilities to mitigate the ever evolving cybersecurity risks associated not only in the cyberworld but also with the online examination systems being used today in every step of the learning process as well as the shortlisting process of any institute/organisation. Machine Learning (ML) must be used to counter the attacks on the system today, as the same are increasing exponentially day by day. For this, Reinforcement Learning (RL), which is a powerful subset of ML, must be effectively used to create strategies in dy-namic environments with varied hardware and software in different scenarios to effectively detect and control the ever-evolving cybersecurity risks and threats while effectively addressing the fatigue in manual processes and providing real-time decision-making for the stakeholders. In this paper, we are investigating extensively the application of RL techniques to pro-vide adaptive decision-making processes by enabling systems with proactive decision-making in real-time changing threat scenarios. This paper evaluates the present state-of-the-art RL-based methods, while highlighting their strengths and weak-nesses, and it also identifies the areas that require further improvements to propose a novel RL-based cybersecurity frame-work tailored to the specific demands of online examination platforms.

UPI Fraud Detection Using Ensemble Approach

Rebeca K. Yeptho; Bendangmenla; Monita Wahengbam; L Lanuwabang — 2026-10-16

Unified Payment Interface (UPI) has paved the way for convenient digital transaction and has enhanced the speed of digital transaction. Users can simply use the technology to make online payments at an accurate speed through mobile number, Virtual Payment Address (VPA) or QR code. But with the evolving technologies and advanced system, there has been cases where fraudulent activities are encountered leading to financial risk and losses. In order to tackle this problem, this research develops an ensemble RNN and XGBoost system for detecting UPI fraud by examining the patterns to recognize and alert on potentially fraudulent activities. It also encompasses data cleaning and preparation, and training of predictive models. Dataset from open source has been trained and tested against the Ensemble RNNBoost (Ensemble RNN and XGBoost) model and has been compared with various ML models. The model’s performance was assessed using various evaluation metrics, including F1-score, precision, recall, and accuracy. The main reason to combine XGBoost and RNN is to exploit sequential patterns and tabular feature interactions simultaneously, leading to overall performance and more robust predictions. As a result, RNNBoost model produce the maximum accuracy rate of 96.44% in comparison with other traditional models like DT (95.00%), RF (95.88%), LR (80.25%), KNN (83.00%), SVM (83.00%), NB (83.00%) and RNN (81.25%).

Deep Learning Alternatives to Error Backpropagation: A Review

Jasbir Singh Muneer Ahmad Dar — 2026-10-16

Error Back Propagation have become the dominant training method for deep neural networks, but it is counter-intuitive to the way biological mind works (biological implausibility) and requires substantial compute resources, so much so that the ownership of adequately trained large language models have become the domain of a few resource rich organiza-tions. This review examines the current trends in the research into alternatives to deep learning, such as biologically-inspired approaches, local learning rules, and novel optimization strategies. The author surveys key advancements mostly from 2020 onwards, discussing the promising approaches such as Forward-Forward algorithms, Hebbian Learning, Equilibrium Propa-gation, and Predictive Coding Frameworks. The error Back-Propagation (eBP) is currently the most successful approach to train deep networks and the alternatives are only chasing accuracies achieved by this method. All major AI models use Deep Learning with eBP and this more-is-more approach has effected alternative forms to AI research [1]. Nevertheless, if the al-ternative methods begin to approach benchmarks set by eBP, a few percentage points drop in accuracy may be acceptable provided the training is possible at a fraction of the compute cost of eBP. Such reduction in requirement for resources is most desirable and, in addition to being beneficial for developers, users and environment, shall democratise the Artificial Intelli-gence ecosystem.

The Rise of Quantum IoT in Smart Healthcare: A Technical Review of Frameworks and Future Directions

Poulomi Deb, Akshay Reang, Prantosh Debbarma; Kumar Amitabh, Debadrita Pal — 2026-10-16

The rapid digitalization of healthcare through Internet-connected medical devices has enabled remote monitor-ing and data-driven clinical decision-making. However, classical Internet of Things (IoT)-based healthcare systems face significant limitations in terms of computational capability, scalability, and cybersecurity. The Quantum Internet of Things (QIoT) introduces a transformative approach by integrating quantum sensing, quantum communication, quantum computing, and quantum imaging into healthcare ecosystems. This paper presents a comprehensive technical review of QIoT applications in smart healthcare, focusing on architectural frameworks, enabling technologies, and security me-chanisms. A unified six-layer QIoT architecture is analyzed, incorporating device, network, edge, cloud, application, and management layers. Recent advancements (2021–2025) in quantum key distribution, quantum machine learning, and post-quantum cryptography are critically examined. Furthermore, practical deployment challenges such as cost, hardware limitations, and integration with existing healthcare systems are discussed. The paper also outlines future research direc-tions toward the realization of Healthcare 5.0.

Detection and Mitigation of Deep fakes and AI-Generated Media

2026-04-02

The rapid advancement of generative artificial intelligence (AI) has resulted in an unprecedented increase in the use of “deep fakes,” highly realistic manipulation of images, videos, and audio content generated by techniques such as Genera-tive Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Diffusion Models. While deepfakes have the po-tential to bring numerous benefits to areas such as entertainment, education, accessibility, and health care, there are many po-tential risks associated with them, including, but not limited to, spreading misinformation or propaganda, identity theft, sabo-taging elections and other government processes, creating unwanted pornography without consent, and reducing the public's faith in digital evidence. This literature review synthesizes the findings from 39 studies on various aspects of deepfakes, in-cluding survey research, experiments, forensic methods, and sociotechnical analyses of both the creation and detection of deepfakes. The main five categories of strategies for detecting deepfakes are: Physiological signal analysis (e.g. eye blinking, inconsistencies in the head pose, lip sync errors etc.). Image and Video Forensic Analysis (e.g. distortions caused by warping, residual noise, and background clues) Deep Learning-Based Detection Techniques (e.g. Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Optical Flow Based Networks and Hybrid Architectures). Hybrid Frameworks and Tool Kits. Human-Centered Detection Techniques, emphasizing the limitations of Human Perception. This review evalu-ates several benchmark data sets used to develop detection research, including DFDC, Celebr-DF, Face Forensics++, DF-Wild, and Deepfake TIMIT. Additionally, this review addresses topics of importance to developing resilience against the in-creasing threat of deep fakes, including, but not limited to, the challenges of achieving Adversarial Robustness, Computa-tional Efficiency, Generalization Gaps, and Dataset Bias. Lastly, the review identifies some of the remaining open problems and possible avenues for future research in multimodal detection, Explainable AI, Lightweight Real-Time Systems, and Regulatory Frameworks.

A Comparative Review of Long Short-Term Memory Based Encoder-Decoder, Transformer, and Hybrid Architectures for Text Sum-marization

Tulsipriya Hanglem, Takhellambam Sonamani Singh; Khundrakpam Veeshel Singh — 2026-10-16

Natural language processing is a recent phenomenon that has been growing in recent years owing to the availa-bility of text documents online. With these massive repositories of text information, researchers have focused on automat-ically and efficiently extracting information by using text summarization systems. Their application is found in so-cial network analysis, market intelligence, educational systems, and financial analysis. In this paper, the recent progress in automatic summarization is comprehensively described. This paper concerns the different optimization methods in text summarization. We present a taxonomy that illustrates the use of different strategies in automatic text summarization in situations where there is single or multi-document automatic text summarization. We present the summaries of some representative state-of-the-art automated text summarization methods and compare them in benchmark datasets. Possible applications, research problems, open issues, and future directions of the automatic text summarization sphere are also addressed in the paper.Extractive text summarization techniques are gaining in-creased popularity because they are capable of generating an informative summary with minimal linguistic re-sources. The next step after converting a sentence to a bag of words will be to first select important sentences in the text, which we shall summarize. Due to the strong data dependence on the performance of the methods, the effect of the parameter on different datasets is examined. Furthermore, we also consider the methods of assessment such as ROUGE metrics and more recent semantic-based ones to assess quality of the summary and the competitiveness of the facts.

IoT-Driven Smart Waste Management with Integrated Security for Urban Environments-(BinBot)

2026-10-16

The rapid growth of the urban population has been challenging, as the waste generated is not efficiently managed. The inefficient methods are not able to coordinate between community needs and the management processes. This failure of management leads to overflowing bins, unpleasant odors and an unhygienic environment, which spreads many diseases, making human beings feel unsafe and uncomfortable in their own surroundings. The role is not only of management, but the first step is to be taken by the people, using bins properly and not throwing the waste on roads or in water bodies. This paper focuses on waste management once waste is thrown into the bins properly. BinBot an IOT-based waste management system designed to assist municipal department and support smart city operations by monitoring bins and taking real-time actions once the bins get full. In this project, sensors microcontrollers and cloud technology are integrated to provide real-time monitoring, data collection and real time alerts along with the location of bin reducing time wastage and ensuring efficient and hygienic waste management.

FPGA Implementation of A 64-Tap Fir Filter Using Radix-8 Booth Multiplier In Verilog

Nallapati Sowmya, Sreejeesh S G; Nanditha N Varma — 2026-10-16

In this paper, the FPGA implementation of a 64-tap FIR filter is realized by using a Radix-8 Booth multiplier for efficient multiplication and high-speed performance. Designed using Verilog HDL, and implemented on a Xilinx Artix-7 FPGA (xc7a100tcsg324-1), the prototype was implemented on a Nexys A7-100T development board, with synthesis and implementation performed in Vivado 2021.2. Considering the computational accuracy and hardware complexity, the Q1.15 fixed-point format is adopted in this design. In this work, the architecture is analyzed based on the resource utilization and operating frequency, making it appropriate for real-time FPGA-based digital signal processing applications.

Design of Analog Power Line Filter for Sensitive Equipment

Meesala Joshita, Maddula Sreeja;, Anand N, Jayaraj U. Kidav — 2026-10-16

For precise and dependable operation, modern electronic systems depend heavily on a clean and steady power source. However, undesired disturbances like noise, harmonics, and electromagnetic interference are frequently present in practical power lines, which can impair the functionality of delicate equipment. In order to reduce such disruptions and enhance power quality, an analog power line filter is designed and implemented in this work. Resistors, capacitors, and operational amplifiers are among the basic analog components used to implement the suggested filter. It works by maintaining the integrity of the basic power signal while attenuating unwanted frequency components, especially those resulting from fluctuations, harmonics, and EMI. The design places a strong emphasis on practicality, affordability, and simplicity.

FPGA Implementation of Vehicle Emission Monitoring and Alert System

Vishnu Hari, Nanditha N Varma; Sreejeesh S G — 2026-10-16

This paper presents the design, synthesis, and full hardware validation of a real-time vehicle emission monitoring and alert system implemented in Verilog HDL on a Xilinx Artix-7 FPGA (XC7A100TCSG324-1) using the Arty A7-35T development board. The system simultaneously monitors four critical vehicular pollutants — Carbon Monoxide (CO), Carbon Dioxide (CO2), Sulfur Dioxide (SO2), and Particulate Matter (PM2.5) — by processing 10-bit Analog-to-Digital Converter (ADC) in-puts through a fully synthesisable fixed-point arithmetic pipeline. Pollutant thresholds are aligned with WHO Air Quality Guidelines and Bharat Stage 6 (BS6) emission norms: CO at 20%, CO2 at 15%, SO2 at 10%, and PM2.5 at 100 μg/m3. When any threshold is exceeded, a threshold com-parison flag is asserted within a single clock cycle (10 ns at 100 MHz), and a ―Needs Repair‖ alert is displayed on a 16×2 LCD with rotating multi-gas messages. End-to-end alert latency — from ADC value presentation to LCD character update — is analytically derived and experimentally confirmed to be no greater than 30 clock cycles, or 300 ns at 100 MHz, which is more than four orders of magnitude faster than comparable IoT-based SMS alert systems (10–20 seconds). Post-implementation results show a total on-chip power consumption of 0.077 W, 9.86% LUT utilisation, 6.69% flip-flop utilisation, and a worst-case timing slack of +0.1 ns. Hardware validation is performed using the Vivado Integrated Logic Analyser (ILA) for internal waveform capture and the Virtual In-put/Output (VIO) core for real-time ADC stimulus injection. The prototype establishes a scalable, low-power, connectivity-independent plat-form for automotive emission diagnostics and sets the stage for physical sensor integration (MQ-7, MQ-135, MQ-136, PMS7003), IoT connectivity, and on-chip machine learning inference.

Dragonfly: A Lightweight Offline Antivirus System for Executable Malware Detection Using Machine Learning

HeisnamBidyaluxmiDev, Sarungbam Usharani Dev, ThongamBobo, shamSanjotaChanu — 2026-10-16

Antivirus products are often depend on signature-based and heuristic-based methods, which are often unable to manage the ever-increasing malware variants. Dragonfly is a Windows-focused, fully offline antivirus system designed for practical end-users without network connectivity. The detection models are implemented in Python and operate with an em-bedded inference pipeline for deploying the models with an Electron-based inference and a file watcher for real-time scan-ning on end-user machines, it also utilizes a custom static feature engineering and vectorization pipeline based on raw Porta-ble Executable (PE) measurements. The representation includes byte-level distributions, entropy structure, string semantics, PE structural features, and hashed API interaction patterns. The detection models are trained on the EMBER 2027 benchmark and evaluated on the EMBER 2028 dataset, which allows generalization under concept drift. We evaluate Logistic Regres-sion, Support Vector Machines, Random Forests, xgboost, and lightgbm using cross-validation and temporal testing. While Gradient boosting methods achieve the strongest overall detection performance, scalability and deployment constraints at the EMBER scale motivate the selection of lightgbm as a balanced trade-off between accuracy, inference latency, and resource usage, with an accuracy of 89.93% and an f1-score of 89.48%. Inference cost, memory consumption, and confusion matrix analysis also support this selection. Model inference is integrated through an embedded Python managed by a native C++ helper, enabling low-latency inter-process communication with the user interface .Experimental results show that Dragonfly delivers a reliable static malware antivirus that can achieve practical throughput through careful feature engineering, prin-cipled evaluation, and modular system desig

Predicting Student Depression and Stress Using Multimodal Daily Behavioral Data

Lalhminghlua R.Dengchhunga — 2026-10-16

This study investigates the possibility of forecasting depression, stress and mood in college students, through the examination of their activities. The dataset comprises. Subsequent depression assessments (PHQ-9) daily step counts, sleep habits, screen time social media engagement, communication records, movement data and students self-reported mood and stress states. A Random Forest Regression technique was applied to estimate PHQ-9 scores while Random Forest Classification categorized students into Mild, Moderate and Severe depression groups. Addi-tional models were created to forecast stress levels (1-5) mood ratings (0-10) and total risk (Low/High).We can have resulted to show that factors like duration of sleeping, screen timing, motion and social well-being, communication frequency are very important indicator of mental health changes. This approach can be useful, non-intrusive way to monitor students mental health and well-being for providing support when needed.

Hybrid Log Intrusion Detection: Combining Anomaly Detection with Snort Signatures for Host Security

Prashant Kumar Tripathi, Abhinav Mishra;, Ajay Verma, D.K. Mishra — 2026-10-16

In this paper, we describe the design of a hybrid log-based IDS using anomaly-based and signature-based techniques for host security in Python. The aim is to take advantage of the complementary strengths of anomaly detection for novel threats and signature detection for known threats in order to increase both the coverage and accuracy of detection. The system leverages machine learning algorithms to model normal log behavior and detects anomalies, and integrates the Snort (open-source IDS) signature engine to catch known attacks. The synergy between anomaly detection and signature analysis seeks to cover zero-day attacks and known exploits so as to mitigate the limitations of each approach on its own..We have validated the performance of the hybrid IDS using some public intrusion detection benchmark datasets such as the KDD Cup 99 dataset. The experimental results are quite impressive-the combined approach shows higher detection rates and a lower false-positive rate compared to standalone anomaly or signature-based. These findings have hence established that our hybrid IDS, written in Python, can substantially enhance threat detection and reduce false alarms. This has been achieved by successfully fusing machine learning-driven anomaly detection with Snort-based signature analysis for host intrusion detection.

Plant Diseases Detection for Sustainable Agriculture

2026-10-16

Agriculture plays a crucial role in the economy of any country and agriculture, providing food, raw materials, and employment opportunities. However, plant diseases continue to be a significant threat to crop yield and food security. Early and reliable identification of such diseases is vital to mitigate their impact. Conventional identification methods are often dependent on expert analysis, which may be inaccessible in rural or remote regions and are prone to subjectivity and delays. To address these challenges, this work presents an automated system for detecting plant diseases using Convolutional Neural Networks (CNNs) based on the ResNet-34 model. ResNet-34 is selected due to its efficient handling of deep network training through residual learning, which enable better accuracy. The system is trained on the publicly available Plant Village dataset, comparing over 54303 leaf images categorized into 38 different plant disease classes. The approach includes preprocessing the images, applying transfer learning using ImageNet pretrained weights, and fine-tuning the model for better generalization. The result shows high performance on validation and testing datasets, showing the model works effectively in real-time plant disease detection. This solution is a practical and scalable tool aimed to helps the farmers with technological support, promoting healthy crops and sustainable agricultural practices and reduces the crops and financial losses due to disease in plants.

A Comprehensive Review of FPGA-Based power Efficient Arithmetic Logic Units using Carry-save Arithmetic synthesis, analysis and comparative Insights

Vaishali Bharati; Yogesh Kakasaheb Shejwal; Anand N; Jayraj U. Kidav — 2026-10-07

The unremitting requirement on high-performance and energy-efficient computing in embedded systems, digital signal processing (DSP), and cryptographic applications has fuelled the need of optimized Arithmetic Logic Unit (ALUs). Field-Programmable Gate Arrays (FPGAs) provide a programmable platform that is suitable to deploy application specific ALU structure. Carry-Save Arithmetic (CSA) of various optimization approaches has become one of the key approaches to optimizing the speed and power usage of arithmetic circuits, especially in multiple-operand operations such as accumulation and multiplication. In this paper, the authors provide a review of the power-efficient ALU designs that are based on carry-save arithmetic and use FPGA. We perform an analytical methodology of synthesis methodologies, architecture variants and optimization strategies that have been used to include CSA in the ALU datapaths. An in-depth analysis of power, performance and area (PPA) trade-offs with particular attention to dynamic reduction of power by minimizing glitching and critical path delays is discussed in the review. Moreover, we provide comparative data about modern-day designs, comparing them to traditional binary arithmetic designs in the throughput, latency, logic usage, and on-chip power dissipation. Important issues like design complexity, verification costs and how to interface CSA units with standard logic are addressed extremely. Lastly, the paper provides a note of future research, such as the synergy between CSA and approximate computing, CSA use as a low-voltage operation in FPGAs, and CSA-based datapath generation using automated high-level synthesis tools. The goal of this survey is to be a reference point to a researcher and engineer as they think of the next generation of power-constrained, and yet, arithmetic systems on reconfigurable hardware.

Study on Steganography using DNA

Fred Ropekkima Darlong, Poulomi Deb, Kumar Amitabh, Indrajit Tripura — 2026-10-16

Steganography is the practice of hiding data in plain text or other types of media. It is crucial to protect or make it safe sensitive information from unauthorized access or tampering. Traditional digital storage methods have issues with capacity, security, and traceability. In contrast, DNA-based steganography uses the large storage capacity and complexity of nucleotide sequences. This paper examines the encoding and decoding schemes, security models, synthesis processes, and storage methods of DNA steganography. It also explores trends in encryption patterns, data hiding techniques, and system weaknesses. Conquering the issues of scalability, cost, and standardization remains in spite of tremen-dous progress in areas such as error correction, biochemical stability, and data security. This review emphasizes the need for multidisciplinary re-search integrating information theory, cryptography, and biotechnology in order to overcome the existing limitations. This paper presents future research directions to achieve secure and optimal data hiding using bio-logical sequences by analyzing the benefit and limitation of DNA-based steganography.

A Multi-LSTM Ensemble and PPO-Driven Trading Agent with Integrated FinBERT Sentiment Intelligence

Moodu Venkatesh; Prasoon Kumar K G — 2026-10-16

Financial market is a non-linear dynamical system which is very complicated, which is conditioned by a great number of factors, starting with historical price, tendencies towards abrupt changes of world opinion [1]. Traditional algorithmic trading tends to fail since it is based on fixed rules that are not adaptive to the stochastic character of the mar-ket. [4] [7] To solve this, we would suggest FinSynapse and FinBERT Sentiment Intelligence, an improved trading agent that combines Deep Reinforcement Learning. Natural Language Processing (NLP) (DRA) [7] [12]. Our hybrid architecture which works on by implementing group of parallel LSTM Networks to dis-extract time characteristics like momentum, trend and volatility which is generated through a Proximal Policy Optimization (PPO) agent [11] and it is used for decision making. Moreover, we combine FinBERT to quantify financial news market sentiment, which can be adjusted by the agent for decision making. Backtesting on Dow Jones data indicates that this mixed strategy works with a Accumulative Return of 97.44 and Sharpe Ratio of 2.21, significantly becoming superior to conventional baseline models.

Performance Enhancement of Optimized Dynamic Circuit for Low-Power and High- Speed Applications

2026-10-16

In this work the design of a high-speed, low-power dynamic comparator for analog-to-digital converter (ADC) applications is proposed and analyzed. The design incorporates a differential amplifier-based pre-amplifier and a regenerative latch made up the two-stage design of the proposed circuit. In order to improve performance, more transistors are added to act as control switches, isolating the pre-amplifier stage once the latch is engaged. Without compromising comparison accuracy, this method significantly lowers kickback noise and dynamic power dissipation. To achieve a power-delay trade-off, an improved version of the proposed design is also developed, offering flexibility for various design objectives. The Cadence Virtuoso tool is used to construct and simulate the comparator in 45nm CMOS technology. The proposed circuits key parameters such as the power consumption, offset voltage, propagation delay, and power-delay product and offset voltage for which the simulation results are significantly superior to those of conventional dynamic comparators. Because of its enhanced speed and energy efficiency, the recommended comparator is perfect for high-performance, battery-operated, low- voltage ADC systems

FPGA Implementation and Synthesis of Rate-1/2 Convolutional Encoder and Deterministic Viterbi Decoder on Basys3 (Artix-7)

Ayushi ; Sreeja Maddula; Shruti Shakya Anand N; Dr. Jayaraj U Kidav — 2026-10-16

Robust forward error correction (FEC) techniques are necessary for reliable data transmission in noisy communication channels. Because of its simplicity, deterministic behavior, and suitability for digital circuits, convolutional encoding in conjunction with Viterbi decoding has long been recognized as one of the most effective error-correcting schemes for real-time hardware environments. This study uses Verilog HDL to present the full design, simulation, hardware implementation, and experimental validation of a rate 1/2 convolutional encoder coupled with a deterministic Viterbi decoder. The Basys3 Artix-7 FPGA development board is used to implement both encoder and decoder architectures. The decoder uses state transitions and branch metric comparison to implement the Viterbi algorithm, while the convolutional encoder uses a two-bit shift-register- based state machine that generates two parity bits for each input bit. The accuracy and robustness of the implemented design are confirmed by the entire system, which includes simulation on Vivado, testbench-based random bitstream testing, and real-time FPGA verification. Because of its resource efficiency, the suggested design can be used for low-power embedded hardware, digital communication system prototypes, and academic learning. Advanced FEC schemes like turbo codes, LDPC codes, soft-decision Viterbi decoders, and SDR-based implementations use the work as an experimental basis.

Bridging the Linguistic Divide: A Comprehensive Review of Methods and Challenges in Low-Resource Machine Translation

Sunil Angom, Pougailung Gangmei; Moirangthem Daniel Singh — 2026-10-16

Large language models (LLMs) have become useful in high-resource languages due to the rapid advancement of natural language processing (NLP), yet the vast majority of the world’s languages remain underserved because of a severe lack of data. This paper develops a complete taxonomy of machine translation (MT) methods for low-resource languages (LRLs) and critically evaluates their construction. Rule-based (RBMT) and statistical machine translation (SMT) remain essential for digitally dark languages such as Tulu and Kangri, while transformer based frameworks and parameter efficient fine tuning (PEFT) demonstrate fluency in mid resource scenarios. We also evaluate data-centric interventions, balancing the advantages of parallel corpus mining against the risks of synthetic data, which can distort semantic polarity in sensitive domains. Our synthesis identifies crossover points where statistical methods outperform neural approaches and integrates findings from dynamic language identification, hybrid transfer learning, and retrieval augmented generation (RAG). The paper concludes with a resource-tier recommendation table and proposes future directions including neuro symbolic hybrid architectures, unsupervised corpus mining, and culturally grounded evaluation metrics.

Numerical Modelling of Efficient Lead-Free Inorganic Perovskite Solar Cell Using SCAPS-1D

Vaibhava Srivastava, Ghanshyam Shivhare; R.K. Chauhan, Sachin Singh — 2026-10-16

Perovskite solar cells (PSCs) have become one of most emerging thin-film photovoltaic technologies. However, traditional lead-based perovskites face major limitations due to their toxicity and limited long-term stability, creating a strong need for environmentally friendly, lead-free alternatives. This work present, SCAPS-1D software is utilized to simulate a PSC based on a stable and eco-friendly inorganic absorber. To enhance carrier mobility, electron transport layer (ETL) is incorporated, and the commonly used PCBM is replaced with graphene oxide (GO), which also contributes to lowering the device cost. The key design parameters, such as thickness, acceptor concentration and defect density of active layer, metal work function and operating temperature are thoroughly examined. With GO functioning as the ETL, optimized device performance achieved includes a PCE of 25.27%, Voc of 1.12 V, Jsc of 28.51 mA/cm2, and an FF of 79.0%. The proposed lead-free PSC structure offers a promising way to addressing stability and toxicity concerns while improving overall device performance.

Evaluating AI Models for Mental Health Monitoring and Diagnosis: A Comprehensive Systematic Review across Text, Speech and Physiological Signals

Gaurav Sharma Gaurav Gupta — 2026-10-16

The scale of the global mental health crisis — affecting roughly 970 million people worldwide stands in sharp con-trast to the tools we currently have for detecting and monitoring it. Questionnaires are subjective. Clinical interviews are re-sourceintensive. And in much of the world, there simply are not enough trained professionals to meet demand. Over the past decade, artificial intelligence has been proposed as a way to address some of these gaps, and a substantial body of research has accumulated around three main data sources: physiological signals, text, and speech. This paper reports findings from a PRISMA-guided systematic review in which we searched eight major databases including PubMed, IEEE Xplore, ACM Dig-ital Library, and ScienceDirect for peer-reviewed studies published between 2015 and 2025. After two independent review-ers screened 120 records, 63 met our inclusion criteria. We grouped these into four modality categories: physiological signal studies (EEG, HRV, ECG, electrodermal activity), NLP-based text analysis, speech and acoustic analysis, and multimodal fusion approaches. EEG-based deep learning models reported the highest accuracy figures sometimes above 97%, though we argue these numbers are difficult to interpret without knowing the sample size and experimental conditions behind them, which in most cases were quite limited. Transformer models such as BERT and RoBERTa did well on depression and suici-dality detection tasks, speech models showed useful diagnostic signals via acoustic and prosodic features, and multimodal systems outperformed single-source approaches across the board. What the review also surfaces, though, is a consistent set of problems: small and unrepresentative samples, almost no external validation, very little attention to fairness or interpretabili-ty, and reporting practices that make replication difficult. These gaps matter. They are what currently stands between techni-cally interesting research and tools that could actually be used in clinical practice.

Non-Western Logic Traditions for Explainable AI: A Survey and Research Agenda

Abhinaba Basu Vikas Mittal — 2026-10-16

Explainable AI (XAI) research predominantly relies on Western logic traditions-Aristotelian syllogisms, Toulmin’s argumentation schemes, and formal symbolic logic. However, non-Western cultures have developed rich epistemological frameworks over millennia that offer complementary perspectives for interpretable AI systems, particularly for modern machine learning models and large language models. We present, to our knowledge, the first dedicated cross-tradition survey of non-Western logic traditions explicitly oriented toward XAI applications, covering traditions from India (Nyaya, Buddhist pramana, Jain logic, Yoga epistemology), East Asia (Neo-Confucian li-qi, Japanese Bud-dhist innen, Chinese Mohism), Africa (Ubuntu philosophy, Ethiopian rationalism), and Islamic scholarship (Kalam, Falsafa, Sufi epistemology). For each tradition, we identify (1) core epistemological principles, (2) structural reasoning patterns, (3) potential XAI applications, and (4) any existing computational implementations.

AgroBot: A Universal AI-Based Agricultural Assistant Integrating CNN Disease Recognition and Multilingual Chatbot Support

Mohd Iliyas, Sandeep Kumar Mishra; Priyanshu Pathak — 2026-10-16

Plant diseases significantly affect crop productivity, especially in developing regions where timely access to expert agricultural assistance is limited. Recent advancements in deep learning offer promising solutions for automated disease de-tection using plant leaf images. However, existing systems are generally limited by monolingual interfaces, lack of explaina-bility, absence of administrative control, and poor field applicability. This paper presents AgroBot, a universal AI-based agricultural assistant that integrates a custom convolutional neural network (CNN) for image-based plant disease recognition with a multilingual knowledge-driven agricultural chatbot. The system supports 38 crop-disease classes with a training image resolution of 128×128 pixels and additional text-based symp-tom–treatment metadata for eight languages. The CNN model achieved a 95.07% validation accuracy with a validation loss of 0.1969, demonstrating strong classification performance with minimal overfitting. The multilingual CNN-powered chatbot is driven by a structured base of knowledge based on the ideas of a JSON, which allows the domain-consistent agricultural advice. AgroBot also supports a role-based authentication of farmers and administrators, live dataset editing, user management, and a deployment interface that is powered by Streamlit. Findings demonstrate that AgroBot is a viable, effective, and scala-ble solution to AI-assistance in agriculture to enhance the ease and dependability of farmers in rural areas. There are future directions that involve mobile deployment, multilingual chatbot support through speech and the possibility of implementing it with IoT crop monitoring.

The Future of Cloud Security: Integrating Quantum-Resistant AI-Powered Solutions

Nandan Sharma; Ritul Agrawal — 2026-10-16

The fast merging of quantum computing and artificial intelligence is bring- ing significant changes to cloud security.
We must create new, quantum-secure approaches because quantum computers are set to crack the present encryption
standards. At the same time, AI makes it easier to detect and manage threats, analyze unusual activity, and organize security
in today’s advanced cloud sys- tems. In this paper, we present a way to implement cloud security by pairing post-quantum
cryptography with the help of AI systems. It examines machine learning happening in predictive analysis, automated reaction,
and secure sys- tem construction, together with methods to preserve data integrity in the future using quantum-safe encryption.
The study demonstrates that hybridization could significantly improve enterprise and governmental systems as we
prepare for a situation after quantum computing is uncovered

Design, Validation and Characterization Grain-128 AEADv2 Lightweight Crypto IP Core

Fathimathu Ruksana K K; Nandakumar R — 2026-10-16

The need for low resource devices has significantly increased as a result of the development of Internet-of-things ap-plications. Various light weighted stream cipher algorithms have emerged as promising solutions for securing IoT communica-tions. They are used on hardware because of its simplicity, speed, limited resources and processing power. Lightweight crypto-graphy overcomes these constraints by optimum utilization of the resources available, while maintaining security, Because of their economical implementations, Grain128AEAD ciphers stand out among other stream ciphers. There are different variants of the Grain cipher, and Grain-128AEADv2 is one of the ciphers. Grain 128 AEADv2cipher with 64 bits plaintext and 128 bits key 96 bit nonce(IV),64 bits associated data forming a 64 bits ciphertext and 64 bits Tag. The Grain 128AEAD cipher is small and quite efficient. In this paper, Design validation and characterization of grain128AEADv2 cipher IP core. The hardware platform used the Xilinx Artix-7 FPGA board and Vivado is used for simulations, synthesis, and implementations utilizing the Verilog hardware description language. The performance parameters results were observed with the help of Xilinx Vivado.

Comparative Analysis of Different Oxides for MoS2 Transistor

Shivangi Srivastava, Md. Ekrama Arshad; Sajal Agarwal — 2026-10-16

The purpose of this study is to evaluate the performance of MoS2 transistor with different gate oxides. This paper systematically evaluates various performance parameters of transistors such as threshold voltage (Vth), on/off ratio, maximum drain current (Id), transconductance (gm), and output conductance (gd) . This comprehensive analysis is done using six different oxides namely silicon dioxide (SiO2), hafnium oxide (HfO2), lanthanum oxide (La2O3), strontium zirconium oxide (SrZrOx), composite of PMMA:SrZrOx, and titanium dioxide (TiO2) for a back-gate device. Results of the paper suggest that SrZrOx exhibits better performance with lower Vth is -0.4 V, along with on/off ratio, maximum Id,and gm, are 1013, 662.57 μA, and 8.12 μS respectively. Device with a relatively higher on/off ratio, high drive current, high gm and lower Vthdemonstrates superior switching and improved performance. With the achieved performance of the above device offers considerable potential for real-time, energy-efficient applications, particularly where power consumption is a big concern. Device with SrZrOx oxide employed for various real-time applications including wearable electronics, medical and industrial applications.

The Convergence of Artificial Intelligence and CRISPR-Cas9: A Review of Innovations, Challenges, and Future Vistas in Genome Editing

Muskan Kirar;, Palak Sobti; Akash Sharan; Abhinav Mishra;,Shaweta Sharma — 2026-10-16

CRISPR-Cas9 has become an effective genome-editing methodology, but the practical deployment of this approach is limited due to challenges such as unintended off-target effects, inconsistent editing efficiency, and lower predictability in complex biological systems. Simultaneously, the development of artificial intelligence (AI) has made it possible to model genomic processes based on data and help us to increase the accuracy and reliability of CRISPR-based treatment. This review is a critical assessment of recent applications of AI combined with CRISPR-Cas9, and specifically guides RNA optimisation, off-target prediction, data enhancement methods, and identification of synthetic lethal interactions applicable to cancer therapy. Instead of a chronological summary of existing methods, the paper includes a comparative analysis of AI-CRISPR-based techniques, along with methodologies, strengths, limitations, and readiness to translate into clinical practice. To conclude, a structured comparative table is highlighted to outline the unresolved gaps that should be bridged in order to translate clinical knowledge to practice. Lastly, recent issues, ethical aspects, and future research are addressed in the direction of responsible AI-driven genome editing.

AdvancesinAI-EnabledPredictiveMaintenance:Models,Industrial Applications, and Research Frontiers

Prafull V. Dhole;, Ankit Manderna; Shraddha Wangujare — 2026-10-16

Predictive Maintenance (PdM), a paradigm enabled by the integration of Artificial Intelligence (AI), the Industrial Internet of Things (IIoT),and advanced data analytics, is presently navigating a crucial shift in the industrial landscape from reactive to preventative maintenance tactics. This change makes it easier to switch from fail-and-fix approaches to predict-and-prevent tactics, which maximizes asset avail-ability, lowers operatingcosts, and improves safety. With a thorough technical examination of deep learning architectures such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, Autoencoders, and the new Transformer models, this paper investigates the state-of-the-art in AI-enabled PdM. With particular focus on energy systems, thermal power generation, and rotating machinery—areas where thermodynamic efficiency and energy analysis are crucial—it critically investigates the implementation of these models across important sectors. Additionally, the study addresses ongoing issues with data scarcity, model interpretability, and cybersecurity in safety-critical situations by identify-ing and discussing emerging research horizons including Explainable AI (XAI), Digital Twins, and Phys-ics-Informed Machine Learning (PIML)

Hardware Implementation of a High-Precision Fixed-Point MAC for Biomedical Application

Himanshu Verma;, Anand N Yogesh Kakasaheb Shejwal;, Jayraja U. Kidav — 2026-10-16

Biomedical signal processing for implantable and wearable technology imposes a crucial design conundrum: operating within strict power and area constraints while obtaining the numerical preci-sion required for clinical validity. With an emphasis on applications like ECG, EMG, and EEG processing, this paper offers a thorough analysis of high-precision Fixed-Point Multiply-Accumulate (MAC) architectures as the key solution to this problem. First, we determine the precision-quantization trade-off by examining the signal-to-noise ratio (SNR) and dynamic range requirements of important biomedical algorithms, such as wavelet transforms, adaptive filters, and light-weight neural networks. Although floating-point units offer inherent dynamic range, optimised fixed-point alternatives are required due to their excessive resource costs. Three axes are used in our review to systematically classify and evaluate MAC architectures: accumulation strategies (guard bits, extended-precision adders), computational core (modified Booth encoding, Wallace/Dadda trees), and system-level optimisation (bit-serial processing, precision-scalable units). We assess their implementation effectiveness and suitability for both ASIC (for ultimate power/area optimisation) and FPGA (for rapid prototyp-ing) targets. With the help of case studies that quantify the effects of MAC architecture decisions on diagnostic metrics like detection sen-sitivity and feature extraction accuracy in real-time biomedical sys-tems, the paper ends with a synthesis of co-design approaches that connect algorithm development and hardware implementation. The goal of this review is to give researchers and engineers a unified frame-work for designing effective, high-fidelity signal processing cores for the upcoming generation of autonomous medical devices.

Performance Improvement of Field oriented Controlled Permanent Magnet Syn-chronous Motor Based Electric Vehicle Drives by ANN Controller

Atul Kumar Asha Rani — 2026-10-16

The present work represents the design and testing of artificial neural network controllers for permanent magnet synchronous motor drives applied in electric vehicles along with the conventional controller. This control methodology in-tends to leverage the power of adaptive learning offered by artificial neural networks (ANN) to overcome the challenges of electric vehicle propulsion, such as varied operating conditions, time-varying parameters, and nonlinear dynamics of systems. Simulation studies indicated that ANN controllers outperformed conventional PI-based field-oriented control methods for all the aspects of speed regulation, ripple reduction in torque, and robustness against perturbation. Real-time modifications of the neural network model to changes in load and outside noise improve performance without the need for complicated decoupl-ing or tedious tuning. These results demonstrate that the ANN technique for the control of PMSMs is quite suitable for elec-tric vehicles and easily integrated into them. It can help in driving by improving efficiency, making runs smoother, and could adapt to various needs of multimodal transportation with further improvements.

Arduino Uno Based NFC-Enabled Smart Labels for Instant Text-to-Speech Access in Public Spaces

Rathlavath Chandru, V Shiva Prasad Nayak; Tahaseen Syed — 2026-10-16

In the modern world, technology plays an essential role in improving accessibility for people with disabilities. This paper deals with Text-to-Voice Converter System for the Visually Impaired Near Field Communication (NFC) Technology based on Arduino Uno. This helps blind people to identify objects using audio output. This paper addresses the lack of accessible information that supports visually impaired individuals in various settings, including pub-lic spaces, hospitals, bus stations, government offices, and educational institutions. To fill this gap, we proposed an Arduino Uno-based Enabled Smart Label System that offers instant text-to-speech (TTS) access. An NFC label encoded with essential information about different products or objects for which we need to be tagged. Once the user taps an NFC card or tag on the NFC reader module PN532, NFC is used to detect NFC tags via the I2C interface. The reader is interfaced with a microcontroller, typically Arduino Uno, and the system identifies the previously stored text. The DF Player Mini, which is a mini-MP3 module, is used to play MP3 files stored on a microSD card, and a Micro (4/8GB) SD Card is used to store audio files (mp3), and it converts them into speech using a dedi-cated TTS module. A buzzer system and LED are used to enhance user feedback and confirm successful reading. This system operates at low power, supports multi-language audio output, and is mainly suitable for use in crowded or high-noise environments. The proposed method also provides low-cost, fast, and reli-able assistive technology that promotes independence and accessibility for visu-ally impaired individuals. The results in this paper demonstrate high response accuracy, and interaction with the user is simple and scalable for various public applications.