AI-enhanced channel estimation and signal processing for MIMO systems in 5g/6g radio frequency networks

*Corresponding author: Aditi Roy Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0. AI-enhanced channel estimation and signal processing for MIMO systems in 5G/6G radio frequency networks Aditi Roy 1, * and Tasriqul Islam 2 1 University of the Cumberlands, Kentucky. 2 American National University, Salem VA. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 Publication history: Received on 02 December 2024; revised on 08 January 2025; accepted on 10 January 2025 Article DOI: https://doi.org/10.30574/gjeta.2025.22.1.0253 Abstract This study explores AI-Enhanced Channel Estimation and Signal Processing for MIMO Systems in 5G/6G Radio Frequency Networks, addressing key challenges in optimizing network performance. It examines critical research questions and objectives, structuring the analysis around advanced methodologies and frameworks tailored to the field. By leveraging AI-driven approaches, the study systematically enhances the credibility and reliability of the results, highlighting significant outcomes such as improved channel estimation accuracy, reduced latency, and enhanced spectral efficiency. These findings contribute to advancing domain knowledge and practice by introducing innovative strategies for addressing the complexities of next-generation wireless networks. The research also underscores the need to generalize its observations while offering pragmatic recommendations for practitioners, policymakers, and scholars. It identifies actionable insights and proposes future research directions to extend the applicability of AI in wireless communication systems. This study bridges theoretical advancements and practical implementations, emphasizing the transformative potential of AI-driven signal processing in MIMO systems. Ultimately, the work advocates for more interdisciplinary research to maximize the benefits of AI technologies in radio frequency networks, laying the foundation for future exploration in the 5G/6G landscape. By addressing critical gaps and presenting new perspectives, this research strengthens the case for adopting AI-enabled solutions in the telecommunications industry. Keywords: AI-Enhanced Signal Processing; MIMO Systems; 5G/6G Radio Frequency Networks; Channel Estimation 1. Introduction The use of multiple-input multiple-output (MIMO) systems has evolved wireless communication through significant improvements in spectral efficiency, data rates, and network performance. MIMO technology, which employs numerous transmitting and receiving antennas, is widely used in various telecommunication standards and, especially, in the current 5G networks. As we move toward the realization of 6G, MIMO systems are expected to play an even more critical role given the persistent need for high-speed, reliable connectivity [1], [2]. Enhanced features like beamforming, spatial multiplexing, and diversity gain are facilitated by these systems, making them crucial in scenarios that demand improved reliability across diverse terrains [5]. 1.1. Problem Statement However, MIMO system deployment brings several challenges, particularly in the areas of channel estimation and signal processing. The analysis and design of MIMO systems becomes complex when the number of antennas and users is Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 22 large, or when multiple transmission setups are involved. These factors increase the overheads and resources required. In the following studies, accurate channel state information (CSI) is essential for improving system performance. However, obtaining CSI is often constrained by hardware limitations, noise, and the inherently dynamic nature of wireless systems. Existing channel estimation methods for single antenna systems are insufficient to address the demands of large-scale MIMO in 5G and 6G networks. This challenge is compounded by the integration of innovative technologies, such as massive MIMO, mmWave communication, and reconfigurable intelligent surfaces (RIS) [3], [6]. 1.2. Objective In this context, the application of artificial intelligence (AI) has emerged as a promising solution to overcome these challenges in wireless communication. AI-enabled techniques offer effective options for addressing various aspects of MIMO systems, providing unique solutions for channel estimation, signal processing, and resource management. This article explores the potential of using AI to optimize MIMO system performance and accuracy. By leveraging machine learning (ML) and deep learning (DL) approaches, researchers can design efficient, self-learning frameworks for updating CSI acquisition and mitigating interference. The aim of this work is to provide an overview of the subject and illustrate how AI aids in the transition from 5G to 6G networks, while also highlighting the limitations of previous methods [7], [8], [11]. Figure 1 A Conceptual diagram showing the role of MIMO systems in 5G/6G networks 1.3. Structure This article is divided into different sections to give the reader a systematic approach to the discussion of the topic. After the introduction, Section 2 is all about ‘Understanding MIMO Systems,’ where its working methodology and inclusion in wireless networks are explained elaborately. Section 3 provides an overview of the issues arising from large-scale systems, specifically in relation to large-scale MIMO systems, on aspects of channel estimation and resulting signal processing and computational requirements [1], [2]. In Section 4, MIMO structures combined with AI are discussed, and the understanding of AI techniques applied is provided. The final section of the paper is Section 5, which covers case studies and recent developments in AI-based MIMO systems, including their real-world applications and results. Section 6 on future trends and research directions in the use of AI in wireless communications assesses the general utility of AI in relation to wireless communication. Last but not least, Section 7 is devoted to the conclusion and highlights the prospects of AI applications to develop the next generation of wireless networks [3], [5]. To sum up, this article aims at presenting information that is relevant to researchers, engineers, and policymakers actively working on the development of WLANs and other wireless communication systems, which urgently need fresh scientific input in today’s world of rapid technological innovation [4]. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 23 2. Related Work The development of wireless communication systems has been in constant synergy with channel estimation, signal processing methods, and, more recently, the implementation of artificial intelligence. The works that form the background to the present study are discussed in this section, particularly conventional channel estimation approaches, innovations in signal processing, and the early integration of AI into wireless communication systems. 2.1. Conventional Channel Identification Schemes From the very beginning of wireless communication systems development, channel estimation has been one of the key required features for accurate decoding of transmitted signals. Early research in wireless communication commonly used conventional approaches, which mainly depended on sound mathematical theories and deterministic paradigms. Among the first techniques that were introduced was the pilot-based approach, where some symbols within the transmitted data were known, allowing receivers to estimate the channel conditions by comparing them with the received signal. These methods showed high reliability in systematic platforms but suffered from segmental and extensive environments, including moving objects or serious multipath effects [6], [7]. Another widely used technique was blind channel estimation, which relies on the statistical characteristics of the received signal without requiring reference pilot symbols. Although these methods reduce the overhead of pilot symbols, most of them imposed significant computational complexity and had large errors in conditions of low SNR and large channel fluctuations. Other filtering techniques, including the LMS and RLS algorithms, improved channel estimation by adapting to the changing environments due to the variation of its parameters. Nevertheless, these methods were considered too rigid and encountered issues with determining the compromise between the speed of convergence and the desired accuracy of estimation, especially in real-time cases [8], [9]. 2.2. Currently Employed Algorithms in Signal Processing Signal processing has played a critical role in enhancing wireless communication networks by increasing the reliability of data transmission and improving frequency efficiency. Traditional techniques that significantly contribute to combating noise, interference, and fading include equalization, diversity combining, and error correction. Techniques like ZF and MMSE equalizers were used across systems to mitigate ISI experienced due to multipath propagation. Although these techniques provided satisfactory performance, they typically required accurate CSI, which could pose a challenge in rapidly fading scenarios [10], [11]. Table 1 Comparative Summary of Prior Research on Traditional vs. AI-Driven Network Security Approaches Aspect Traditional Approaches AI-Driven Approaches Gaps Identified Key Methods - Rule-based systems (e.g., firewalls, IDS). - Manual monitoring and response. - Machine learning algorithms for threat detection. - Automated response systems. - Limited scalability of traditional methods. - Need for more real-time data in AI systems. Results - Effective against known threats. - Higher reliance on human intervention. - High accuracy in detecting emerging threats. - Reduced response time to incidents. - Traditional methods struggle with advanced persistent threats. - AI systems require large datasets. Scalability - Limited ability to scale with growing network size. - Highly scalable with cloud integration and adaptive algorithms. - Resource constraints in traditional systems. - High cost of AI implementation. Adaptability - Static rules and configurations; slow to adapt to new threats. - Dynamic learning capabilities to identify novel patterns and anomalies. - Traditional approaches lack flexibility. - AI depends heavily on quality of training data. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 24 Cost - Lower initial investment but higher operational costs due to manual interventions. - Higher initial investment but lower long-term costs due to automation. - Balancing cost-effectiveness with security needs. Human Dependency - Heavy reliance on IT staff for monitoring, updates, and threat mitigation. - Minimal dependency on human operators postdeployment. - Skill gaps in AI-based implementations among staff. Threat Detection Accuracy - Effective for known threats but fails against zero-day and sophisticated attacks. - High precision in identifying zero-day and polymorphic threats. - Inconsistent performance of AI in low-data environments. Used together with spatial, frequency, and time diversity, signal reliability was improved through transmission over multiple paths. When those schemes, for instance, the maximal ratio combining (MRC) and selection combining (SC), were implemented, there were significant gains in fading reduction. Furthermore, new levels of coding, including the turbo codes and the low-density parity check (LDPC), brought a new level of error correction Dong [2000]. Nevertheless, the existing signal processing techniques left some issues unsolved, such as the wireless network developing into the complexity of modern mass MIMO systems and millimeter-wave communication. 2.3. Initial Applications of AI in Wireless Communications Wireless communication is a good example of how introducing AI marked a shift from pre-defined models to learningbased environments. These narrow AI application areas included channel estimation, resource allocation, and interference management. Supervised learning, for example, was first used to improve channel estimation precision since the models were trained on vast volumes of labeled data on the channel array. It also revealed that these approaches offered significant enhancements to conventional methods, particularly in circumstances where channel characteristics lacked simplicity or linearity [1]. AI algorithms, particularly optimization and reinforcement learning algorithms, have been applied to efficiently allocate carrier spectrum, power, and other resources. These methods proved more effective than traditional optimization algorithms due to their ability to adapt to the ever-varying demands of the network and its users. Similarly, interference management techniques aim to leverage time-tested AI tools that help predict interference and manage networks with high traffic density [2][5]. Deep learning also pushed the boundaries of AI in wireless communications. In MIMO systems, applications of neural networks, including CNNs and RNNs, have been used to address signal detection, modulation classification, and beamforming. These techniques demonstrated higher efficiency than traditional extraction methods due to the richness of high-dimensional data and the ability to learn more subtle features for accurate and efficient communication processes [6][7]. However, the adoption of AI in wireless systems was not without challenges. A major obstacle was the lack of standard datasets, while another challenge arose from the unrealistic constraints of computing and processing power on otherwise promising AI solutions. Moreover, the interpretability of AI models remains a concern, as the black-box nature of most models makes it impossible to certify their safety for critical applications [8][9]. 3. Nonetheless, MIMO systems are not yet fully realized in 5G/6G networks 3.1. Architecture Overview Large-scale MIMOs are integral to the foundational design of 5G networks and are expected to further amplify their importance for 6G networks. The most basic form of MIMO involves signal transmission through multiple antennas at both the transmitter and receiver sides to increase spectral efficiency, reliability, and data rates. In 5G networks, MIMO configurations are significantly different from previous generations, incorporating ‘massive MIMO’ technology that provides hundreds of antennas to the base stations. This technology also enables spatial multiplexing, where several users receive data simultaneously, thereby increasing the network’s capacity [10][11]. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 25 Figure 2 A block diagram illustrating the MIMO architecture in 5G/6G It has been anticipated that these enhancements are going to build on this framework in 6G. Holographic MIMO and IRS are seen as extensions to the current methods with antenna arrays, but the former is still in its developmental stage and captures the aim of owning an adaptive surface in what direction electromagnetic waves should travel. Holographic MIMO is based on an ultra-dense antenna array setup that will be able to form tailored wavefronts from the intended transmitters for efficient transmission while with the advent of IRS including and improving reflection and refraction in signal propagation. Collectively, these advancements should help 6G networks effectively support and add terahertz frequencies and ultra-high speed communication as needed to augment the insatiable consumer demand for lowlatency high-capacity services for applications in augmented reality, autonomous systems, and real-time machine learning. 3.2. Channel Estimation Problems MIMO systems have a number of distinctive advantages but at the same time they bring a number of problems, especially for channel estimation. MIMO systems require CSI, of which achieving accurate results in the settings of today’s 5G and the tomorrow’s anticipative 6G is a challenge. Figure 3 A flowchart showing the challenges in channel estimation Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 26 One of the most critical challenges is the pilot contamination that occurs in multi-cell systems where non-orthogonal pilot signals inter_cross in adjacent cells. In this interference, CSI decreases and affects the MIMO systems’ overall efficiency and consumes substantial pilot resources compared to the number of antennas in Massive MIMO structures. Another difficulty is losses of signal due to hardware imperfections like nonlinearities present in amplifiers; phase noise, and quantization errors originating from using ADCs with finite resolution to estimate channel effects. These impairments are more prominent, especially in those bands that were typically challenging at lower frequencies such as Millimeter wave (mm-Wave) and Terahertz bands. Another major issue is high dimensionality which becomes a paramount problem in huge MIMO and beyond. As the number of available antennas and frequency bands is further increased the size of the channel matrix increases exponentially and poses computational and storage issues. This high-dimensionality is not only challenging in channel estimation but also in other processes such as signal processing and resource allocation carried out afterward. Solving such problems calls for creative solutions that include machine learning for CSI prediction, novel pilots for interference control, and analog-digital beamforming to reduce dimensionality without compromising on performance. 3.3. Signal Processing Needs The adoption of MIMO systems in 5G as well as in future 6G networks require signal processing techniques that could satisfy the requirements for low latency rates, low energy consumption and system reliability. Signal detection and demodulation in massive MIMO systems observe the ability to detect various data streams at the same time, which may be impaired by different noises, interfirs, as well as hardware constraints. This task becomes even more challenging in the extremely dense environment as well as in the high-frequency scenarios of 6G. Latency becomes a fundamental value, especially when working with real-time systems such as self-driving cars, smart factories or virtual reality. It is thus seen that the power of traditional signal processing methodologies can fail to deliver the desired ultra-low latency imperative of 6G. To this end, new methods such as compressed sensing and the use of deep learning, which substantially reduce the amount of time required to recover signals, are under development. Figure 4 A bar chart comparing latency, energy efficiency, and robustness Other important factor is energy efficiency due to the rising number of antennas, and higher frequency operation in the case of MIMO systems. The initial massive MIMO signal processing methods were conventional methods that required a large amount of power, which might not be ideal for further large scale usage. Much research is going into designing low energy-consuming algorithms like, low-resolution ADCs, hybrid beamforming and model-based machine learning algorithms to overcome this problem without a drastic fall in performance. Furthermore, hardware improvement including energy-harvesting antennas and efficient processing units, are assumed to support these algorithm improvements. It is almost as critical as performance, especially where organizations operate in conditions that are volatile and uncertain. Sustainable growth of MIMO systems is contingent on their capability to counteract channel fluctuations, jamming, and hardware distortion. This entails signals that can be adapted to dynamically change their signal processing Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 27 in response to changing conditions on the network. For example, the reinforcement learning-based methods can control the flow of system parameters depending on the needs of various scenarios, whereas robust optimization can consider uncertainties in CSI and hardware limitations. 4. AI Driven Channel Estimation Artificial Intelligence (AI) has become the new frontier in addressing the challenges of wireless communication systems. Among all the fundamental applications of AI, one of the most promising is channel estimation for multiple-input multiple-output (MIMO) systems. Specifically, AI-based methods are designed to improve the channel estimation processes by utilizing more advanced learning algorithms [1], [2], [5]. In this context, the role of AI is highlighted, with a focus on machine learning, feature engineering, performance measures, and real-world examples that demonstrate the superiority of AI-driven approaches. 4.1. Machine Learning Models The pattern employed by AI-driven channel estimation is largely expressed through complex machine learning algorithms, which are particularly useful given the complexity of MIMO systems. For instance, neural networks are frequently utilized because they allow programmers to model systems with non-linear associations and learn complex relationships in datasets [7], [9]. Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are particularly noteworthy; CNNs excel at capturing spatial aspects, while RNNs are more suited for addressing temporal dynamics. Another critical learning paradigm is reinforcement learning, which enables systems to adapt channel estimation and optimize performance in evolving environments without the need for feedback [5], [12]. Generalization—the ability of a model to perform well in diverse environments—and robustness, which is further enhanced by creating multiple models, make ensemble methods particularly well-suited for various channel conditions [6], [8]. The combination of these models results in a flexible and highly effective approach to enhancing channel estimation in real-world scenarios. Figure 5 A diagram summarizing the machine learning pipeline for channel estimation 4.2. Feature Engineering Feature engineering is identified as a significant factor in determining the efficacy of AI models in channel estimation. In MIMO systems, the channel is complex and vast, requiring the extraction and incorporation of discriminative important features in the feature space to capture the essence of channel characteristics. These features include signalto-noise ratio (SNR), delay spread, and channel state information (CSI) [1], [5]. Preprocessing of these features is crucial for achieving good performance, as it enhances the model's capability to learn these features more effectively. Further enhancements can be made through domain-specific features, such as antenna correlation and spatial geometry [2], [5]. High-dimensional data is typically transformed before analysis, using feature selection methods such as principal component analysis (PCA) or mutual information-based methods. This process not only speeds up AI computations but also reduces the risk of overfitting, allowing the model to better learn conditions it has not encountered before [5]. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 28 4.3. Performance Metrics The comparison of results obtained from AI-based channel estimation models presents challenges associated with assessing multiple criteria. Accuracy is the fundamental performance parameter, as it reflects the difference between the estimated channel conditions and the actual values, serving as one of the most straightforward yardsticks for evaluating the overall fit of the chosen model to the channel characteristics. Another important measure is computational complexity, as real-time systems, in which these models are applied, require efficient data processing to avoid delays. We explore some of the complexity metrics involving resource usage and speed, assessing the applicability of such models in systems with limited hardware capabilities, such as mobile devices or IoT [1], [6]. Additionally, robustness metrics consider the degree of system degradation caused by factors like interference, fading, or mobility of the channel. By keeping these metrics in check, AI-driven solutions can be developed and implemented with better performance, while remaining realistically applicable [5]. 4.4. Case Studies Several case studies and simulations have shown the positive implications of using AI in channel estimation for MIMO systems. One specific example is the use of deep learning models for predicting channel states in a system based on massive MIMO technology. By training a CNN on synthesized datasets that closely resemble real-world channel conditions, researchers demonstrated that the improvement in estimation accuracy was significantly higher than with traditional methods. The CNN was particularly robust in addressing the dimensionality problem, which is a common characteristic of many massive MIMO systems [6], [7]. Table 2 Performance Metrics Comparison of AI Models for Channel Estimation Model Accuracy Computational Complexity Training Time Generalization Capability Real-Time Suitability Deep Neural Networks (DNN) High Moderate Long High Moderate Convolutional Neural Networks (CNN) Very High High Long Very High Moderate Reinforcement Learning (RL) Moderate Very High Very Long Moderate High Support Vector Machines (SVM) Moderate Low Short Low High Ensemble Models High Very High Long Very High Moderate Another case study involved using reinforcement learning for channel estimation adaptation in situations where the channel conditions are changing. This approach framed the estimation task as a decision-making problem, where the model learned the best strategy based on feedback received from the environment. This method was particularly successful in dynamic channel conditions that do not permit the use of traditional estimation approaches [5], [6]. Other works demonstrating the use of ensembles of classifiers have focused on robustness and generality. For example, decision trees and neural networks were combined to estimate channels in high interference and multipath environments commonly found in urban terrains. The decentralized approach proved to be more effective at filtering noise and maintaining accuracy across varying conditions compared to using individual models, showcasing the power of ensemble strategies in such contexts [7], [8]. This success is also reflected in simulations, which demonstrate the computational superiority of AI-driven methods. For instance, the feature selection techniques used in the AI models provided estimation times almost identical to traditional models while maintaining similar accuracy. This balance between speed and accuracy is critical for real-time operations, such as in self-driving cars or automated systems like conveyor lines [14], [15]. 5. Artificial Intelligence Aided Signal Processing Artificial intelligence has revolutionized signal processing by improving system efficiency, enhancing signal recovery, and simplifying the implementation of new signal processing frameworks. This section explores AI's multifaceted role Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 29 in signal processing, focusing on three areas: the application of deep learning models for signal recovery, their implementation into real-world systems, and performance evaluation of AI-based methods. 5.1. AI Models for Signal Recovery Signal recovery, traditionally achieved through algorithmic methods, often imposes strict mathematical assumptions such as linearity or stationarity on the data. AI, particularly deep learning, has disrupted this paradigm by offering more flexible and efficient methods that can capture the complex functional dependencies inherent in real-world signals [9], [10]. Deep learning frameworks play a crucial role in signal recovery tasks such as equalization, detection, and filtering. Equalization, which is essential for reducing Inter-Symbol Interference (ISI) in communication systems, benefits from AI's ability to learn dynamic channel properties. Compared to conventional equalizers, neural network-based equalizers are better at adapting to real-time channel changes. This feature is particularly important in mobile environments, such as vehicular or satellite communications, where channel parameters change rapidly [11], [12]. Detection is another key component of signal processing, as it involves identifying transmitted symbols from the received signal. To address the issue of detecting signals that have been distorted by noise or interference, AI-based detectors incorporate advanced structures like Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). These networks are especially effective in handling nonlinear distortions or non-Gaussian noise, a challenge for traditional detection algorithms [13], [14]. AI has also made significant contributions to filtering, which aims to isolate the desired signal while rejecting noise and interference. This study demonstrates that deep learning models can be effectively used in adaptive filtering, as they are data-driven and capable of benefiting from large datasets. Unlike traditional filters, which are designed with fixed coefficients, AI-based filters adapt by updating their parameters in response to the input signal, allowing for more effective noise reduction and signal amplification [15], [16]. 5.2. Integration with Existing Systems As mentioned in the prior section, the theoretical benefits of signal processing with AI tools are numerous, the starting of these techniques are not without their issues. There are key questions concerning computational complexity, compatibility with traditional signal-processing systems, and real-time performance when incorporation of AI models into signal processing architecture is considered. At present, one of the biggest challenges is the compute requirements of AI models and especially deep learning architectures. Most of them may be too complex to implement on existing systems especially because current models tend to need large processor space and memory space for storage. To counter this problem, methods known as model compression, quantization, and pruning are used. These methods help to prune the size and complexity without huge setbacks in the efficiency of these neural networks and make it possible to implement them on the small edge nodes or embedded systems. One of the challenges is the compatibility of the developed AI-driven modules with the existing signal processing pipelines. In general, legacy systems are based on deterministic algorithms while AI is probabilistic thus necessitating change of system architecture. To bridge such a gap, it is necessary to develop new forms of machinery – the designing of which implies the integration of AI components into typical algorithms. For example, AI models can be used in functions such as channel estimate or noise filtering before their results are processed by current equalization or detection units. This provided me the synergistic effect of keeping traditional paper and pen benefits while being free from many of AI jurisdictions. Online operation is an important constraint in many signal processing applications especially in communication systems. The AI models take more time while considering the computations to be done which makes its usability a drawback to a degree. To solve this problem, there must be improvements in the inference times via hardware like GPUs or TPUs and better software frames. Parallel processing and pipelining are the other methods that augment the realtime properties of the AI systems to exhibit sufficient timing characteristics of the present day applications. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 36 [15] Liu, D., Zhang, Y., & Zhang, H. (2005). A self-learning call admission control scheme for CDMA cellular networks. IEEE Transactions on Neural Networks, 16(5), 1219–1228. https://doi.org/10.1109/TNN.2005.852253 [16] Chen, Y.-S., Chang, C.-J., & Ren, F.-C. (2004). Q-learning-based multirate transmission control scheme for RRM in multimedia WCDMA systems. IEEE Transactions on Vehicular Technology, 53(1), 38–48. https://doi.org/10.1109/TVT.2003.822662 [17] Chen, Y.-S., Chang, C.-J., & Ren, F.-C. (2006). Situation-aware data access manager using fuzzy Q-learning technique for multi-cell WCDMA systems. IEEE Transactions on Wireless Communications, 5(9), 2539–2547. https://doi.org/10.1109/TWC.2006.879261 [18] Chen, Y.-H., Chang, C.-J., & Huang, C. Y. (2009). Fuzzy Q-learning admission control for WCDMA/WLAN heterogeneous networks with multimedia traffic. IEEE Transactions on Mobile Computing, 8(11), 1469–1479. https://doi.org/10.1109/TMC.2009.80 [19] Glorennec, P. Y. (1994). Fuzzy Q-learning and dynamical fuzzy Q-learning. In Proceedings of the IEEE International Fuzzy Systems Conference (pp. 474–479). Orlando, Florida. https://doi.org/10.1109/FUZZY.1994.343660 [20] Cui, Y., & Lau, V. K. (2010). Distributive stochastic learning for delay-optimal OFDMA power and subband allocation. IEEE Transactions on Signal Processing, 58(9), 4848–4858. https://doi.org/10.1109/TSP.2010.2053690 [21] Bernardo, F., Agustí, R., Pérez-Romero, J., & Sallent, O. (2011). Inter-cell interference management in OFDMA networks: A decentralized approach based on reinforcement learning. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), 41(6), 968–976. https://doi.org/10.1109/TSMCC.2011.2104947 [22] Vucevic, N., Pérez-Romero, J., Sallent, O., & Agustí, R. (2011). Reinforcement learning for joint radio resource management in LTE-UMTS scenarios. Computer Networks, 55(7), 1487–1497. https://doi.org/10.1016/j.comnet.2011.01.015 [23] Fathi, M., Maihami, V., & Moradi, P. (2013). Reinforcement learning for multiple access control in wireless sensor networks: Review, model, and open issues. Wireless Personal Communications, 72, 535–547. https://doi.org/10.1007/s11277-013-1045-y [24] Dai, H., Huang, Y., & Yang, L. (2015). Game theoretic max-logit learning approaches for joint base station selection and resource allocation in heterogeneous networks. IEEE Journal on Selected Areas in Communications, 33(6), 1068–1081. https://doi.org/10.1109/JSAC.2015.2390731 [25] Mertikopoulos, P., Belmega, E. V., Moustakas, A. L., & Lasaulce, S. (2011). Distributed learning policies for power allocation in multiple access channels. IEEE Journal on Selected Areas in Communications, 30(1), 96–106. https://doi.org/10.1109/JSAC.2012.120104 [26] Wang, L., Wu, K., Hamdi, M., & Ni, L. M. (2013). Attachment-learning for multi-channel allocation in distributed OFDMA-based networks. IEEE Transactions on Wireless Communications, 12(4), 1712–1721. https://doi.org/10.1109/TWC.2013.022113.120734 [27] Li, X., Gao, J., Liu, Y., Xie, G., Mao, J., & Deng, P. (2012). Resource allocation for MIMO-OFDMA downlink based cognitive radio systems with imperfect channel learning. Science China Information Sciences, 56, 1–14. https://doi.org/10.1007/s11432-012-4567-0 [28] Xu, C., Sheng, M., Wang, X., Wang, C.-X., & Li, J. (2014). Distributed subchannel allocation for interference mitigation in OFDMA femtocells: A utility-based learning approach. IEEE Transactions on Vehicular Technology, 64(6), 2463–2475. https://doi.org/10.1109/TVT.2014.2354372 [29] Nguyen, T. T., Nguyen, H. H., Sartipi, M., & Fisichella, M. (2023). Multi-vehicle multi-camera tracking with graphbased tracklet features. IEEE Transactions on Multimedia, 26, 972-983. [30] Nguyen, T. T., Nguyen, H. H., Sartipi, M., & Fisichella, M. (2024). LaMMOn: language model combined graph neural network for multi-target multi-camera tracking in online scenarios. Machine Learning, 113(9), 6811-6837. [31] Nguyen, T. T., Nguyen, H. H., Sartipi, M., & Fisichella, M. (2024). Real-time multi-vehicle multi-camera tracking with graph-based tracklet features. Transportation research record, 2678(1), 296-308. [32] Areo, G. (2024). Optimized Neural Network for Cybersecurity and Smart Camera Parking System Detection in IoT. Global Journal of Engineering and Technology Advances, 2025, 22(01), 021-037 37 [33] Chandrashekar, K., & Jangampet, V. D. (2019). HONEYPOTS AS A PROACTIVE DEFENSE: A COMPARATIVE ANALYSIS WITH TRADITIONAL ANOMALY DETECTION IN MODERN CYBERSECURITY. INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING AND TECHNOLOGY (IJCET), 10(5), 211-221. [34] Chandrashekar, K., & Jangampet, V. D. (2020). RISK-BASED ALERTING IN SIEM ENTERPRISE SECURITY: ENHANCING ATTACK SCENARIO MONITORING THROUGH ADAPTIVE RISK SCORING. INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING AND TECHNOLOGY (IJCET), 11(2), 75-85.