http://hdl.handle.net/11189/1913 Wed, 12 Aug 2026 16:18:50 GMT 2026-08-12T16:18:50Z http://hdl.handle.net/11189/10626 Title: Intelligent reflecting surface-aided wireless networks: deep learning-based channel estimation using ResNet+UNet Authors: sakhshra monga, sakhshra monga; Pathania, Aditya; Saluja, Nitin; Gupta, Gunjan; Sharma, Ashutosh Abstract: Accurate channel estimation is essential for optimising intelligent reflecting surface-assisted multi-user communication systems, particularly in dynamic indoor environments. Conventional techniques such as least squares (LS), linear minimum mean square error (LMMSE), and orthogonal matching pursuit (OMP) suffer from noise sensitivity and fail to effectively capture spatial dependencies in high-dimensional intelligent reflecting surface (IRS)-assisted channels. To overcome these limitations, this work proposes a deep learning-driven ResNet+UNet framework that refines initial LS estimates using residual learning and multiscale feature reconstruction. While UNet enhances channel estimation through hierarchical processing, efficiently decreasing noise and enhancing estimate accuracy, ResNet gathers spatial features. Simulation results show that the proposed method significantly outperforms existing methods across various performance metrics.InNMSEversussignal-to-noiseratioassessments, the proposed approach surpasses convolutional deep residual network (CDRN) by 59%, OMP by 81%, LMMSE by 114%, and LS by 115%. When IRS elements are modified, it overcomes CDRN by 60%, OMP by 78%, LS by 107%, and LMMSE by 110%. Along with this, recommended structure performs more effectively than CDRN by 39%, OMP by 44%, LS by 122%, and LMMSE by 129% across various antenna configurations. The proposed approach is particularly beneficial for augmented reality (AR) applications, where real-time, high-precision channel estimation ensures seamless data streaming and ultra-low latency, enhancing immersive experiences in AR-based communication and interactive environments. These results illustrate the proposed method’s scalability and resilience, making it a suitable choice for next-generation IRS-assisted wireless communication networks. Wed, 01 Jan 2025 00:00:00 GMT http://hdl.handle.net/11189/10626 2025-01-01T00:00:00Z http://hdl.handle.net/11189/10625 Title: Real time vehicle classification using deep learning-smart traffic management Authors: Maurya, Tejasva; Kumar, Saurabh; Rai, Mritunjay; Saxena, Abhishek Kumar; Goel, Neha; Gupta, Gunjan Abstract: As global urbanization continues to expand, the challenges associated with traffic congestion and road safety have become more pronounced. Traffic accidents remain a major global concern, with road crashes resulting in approximately 1.19 million deaths annually, as reported by the WHO. In response to this critical issue, this research presents a novel deep learning-based approach to vehicle classification aimed at enhancing traffic management systems and road safety. The study introduces a real-time vehicle classification model that categorizes vehicles into seven distinct classes: Bus, Car, Truck, Van or Mini-Truck, Two-Wheeler, Three-Wheeler, and Special Vehicles. A custom dataset was created with images taken in varying traffic conditions, including different times of day and locations, ensuring accurate representation of real-world traffic scenarios. To optimize performance, the model leverages the YOLOv8 deep learning framework, known for its speed and precision in object detection. By using transfer learning with pre-trained YOLOv8 weights, the model improves accuracy and efficiency, particularly in low-resource environments. Themodel’s performance wasrigorously evaluated using key metrics such as precision, recall, and mean average precision (mAP). Themodelachieved aprecision of 84.6%, recall of 82.2%, mAP50 of 89.7%, and mAP50–95 of 61.3%, highlighting its effectiveness in detecting and classifying multiple vehicle types in real-time. Furthermore, the research discusses the deployment of this model in low-and middle-income countries where access to high-end traffic management infrastructure is limited, making this approach highly valuable in improving traffic flow and safety. The potential integration of this system into intelligent traffic management solutionscould significantlyreduceaccidents,improveroadusage,andprovidereal-timetrafficcontrol.Futurework includes enhancing the model’s robustness in challenging weather conditions such as rain, fog, and snow, integrating additional sensor data (e.g., LiDAR and radar), and applying the system in autonomous vehicles to improve decision-making in complex traffic environments. Wed, 01 Jan 2025 00:00:00 GMT http://hdl.handle.net/11189/10625 2025-01-01T00:00:00Z http://hdl.handle.net/11189/10575 Title: Analysing the ML-based algorithms for GNSS data bit synchronization and decoding Authors: Madonsela, Bheki; Mukubwa, Emmanuel; Davidson, Innocent Ewean Abstract: The position, velocity, and time global navigation satellite systems are vulnerable to signal interference, distortion, jamming, and multipath, which could potentially render the entire system inoperable due to the generally weak signal strength in these conditions. Due to these problems, the Global Navigation Satellite Systems receiver is rendered inoperable by an exceptionally strong navigation frequency band signal along the satellite path. Since global navigation satellite systems are currently widely used, there is a significant increase in the risks of interference, distortion, and jamming. Multipath concerns have been the subject of extensive research with a variety of approaches. But, first, the level of the Global Navigation Satellite Systems multipath must be estimated using a sample of the total signal that the navigation space satellite emits. The satellite constellation and environmental errors have a significant impact on the navigation system. The maximum likelihood estimation technique is presented in this paper along with an evaluation of its consistency and reliability when the Global Navigation Satellite Systems signal multipath is present. As this paper discusses, multipath signals can be numerically discriminated using maximum likelihood estimation techniques based on receiver measurements without the need for additional devices. The measurements and output data derived from the Global Navigation Satellite Systems receiver configuration parameters are used in the maximum likelihood estimation. It was found that the overall performance of the space data synchronization is determined by the number of data bit transitions rather than the total number of bits. An observed state-space representation, lower signal C/N0, and greater Doppler frequency inaccuracy require more data bits for estimation and computation. It was also observed that, in the majority of cases, if not all of them, the bit evolution occurs with a probability equal to 60% upper. With C/N0 in marginal power estimated to 20 dB-Hz without Doppler error, it is probably going to reach the 95%–100% range. If the Doppler error is less than 6 dB-Hz, the signal attenuation caused by the Doppler inaccuracy is insignificant, and the maximum tolerance of the Doppler inaccuracy is 30 dB-Hz. Wed, 01 Jan 2025 00:00:00 GMT http://hdl.handle.net/11189/10575 2025-01-01T00:00:00Z http://hdl.handle.net/11189/10552 Title: The computational modeling of grid-connected double-chamber microbial fuel cell (DCMFC) bioenergy utilizing MATLAB Simulink software Authors: Shabangu, Khaya Pearlman; Mthembu, Nhlanhla; Chetty, Manimagalay; Bakare, Babatunde Femi Abstract: Owing to the depletion of fossil fuels, rising energy costs, and environmental pollution, there has been a growing focus on exploring and exploiting renewable energy sources. This study aims to demonstrate the modeling of a grid-connected double-chamber microbial fuel cell (DCMFC) biomass energy system via MATLAB Simulink software. The experimental measurements obtained from DCMFC outputs served as the basis for developing the inverter-grid connection model and simulation output in MATLAB Simulink software. Briefly, the DCMFC DC boost voltage was connected to the positive and negative buses of the three-phase DC‒AC inverter circuit, with switching patterns controlled by gate pulses in each transistor. Full square wave single-stage PWM pulses are generated by the gate for control. The power generated by the inverter was measured via a three-phase VI block for analysis, with voltages and currents displayed via a scope. To address potential noise during switching, an LC filter is employed to suppress noise output and stabilize power generation. Another power meter measures power from the grid, with waveforms displayed via scope blocks. Before synchronization between power from the DCMFC and the grid occurs, four requirements must be met: the grid voltage must match the inverter output voltage, the inverter frequency must match the grid frequency, the phase sequence must be the same, and the phase angle of the inverter must match that of the grid. Additionally, Institute of Electrical and Electronics Engineers (IEEE)-specific requirements were evaluated during the simulation for this study. In addition, this study critically examined whether the DCMFC inverter model conforms to conventional requirements, International Electrotechnical Commission (IEC) standards, and IEEE standards for the integration of inverters into the grid. Compared with previous studies on inverter modeling for grid connections, past studies have emphasized the efficacy of biomass power plants and different inverter topologies for photovoltaic (PV) systems. The current study focuses on optimizing a multilevel inverter-based model, achieving low total harmonic distortion (THD) below IEEE standards. In this study, the use of a multilevel inverter configuration proved highly effective in reducing the harmonic content, leading to synchronized phase sequences and enhanced coherence between the grid and inverter voltages. With a THD of 4.75% at 50 Hz, supported by a harmonic distortion value of 422.5, the chosen configuration significantly minimized the harmonic distortion. This success underpins the system’s reliability and efficiency, offering promising implications for practical applications. Wed, 01 Jan 2025 00:00:00 GMT http://hdl.handle.net/11189/10552 2025-01-01T00:00:00Z