Beam alignment for mmWave and THz: systematic review

Madhekwana, Sundire; Usman, Muhammad Arslan; Ayyub, Ahtisham; Politis, Christos Article — Published Version Beam alignment for mmWave and THz: systematic review Telecommunication Systems Provided in Cooperation with: Springer Nature Suggested Citation: Madhekwana, Sundire; Usman, Muhammad Arslan; Ayyub, Ahtisham; Politis, Christos (2025) : Beam alignment for mmWave and THz: systematic review, Telecommunication Systems, ISSN 1572-9451, Springer US, New York, NY, Vol. 88, Iss. 3, https://doi.org/10.1007/s11235-025-01318-7 This Version is available at: https://hdl.handle.net/10419/323473 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Telecommunication Systems (2025) 88:87 https://doi.org/10.1007/s11235-025-01318-7 Beam alignment for mmWave and THz: systematic review Sundire Madhekwana1,2 ·Muhammad Arslan Usman1·Ahtisham Ayyub1·Christos Politis1 Accepted: 28 May 2025 © The Author(s) 2025 Abstract This compact study investigates published research on beam management frameworks for large antenna arrays in mmWave and terahertz (THz) systems. The systematic literature review (SLR) focused on beam alignment and initial access frameworks for mmWave and THz. The investigation uncovered 596 relevant articles, including journals and conference papers from different sources using defined criteria, leading to 73 final studies post-filtering, following our SLR-defined 3-stage screening process. We explained the taxonomy of mmWave and THz beam alignment frameworks by classifying them into beam sweepingbased, context information-based, compressive sensing-based, and Machine Learning/Artificial Intelligence (ML/AI)-based frameworks, as well as Integrated Sensing and Communication (ISAC)-based frameworks, which combine ISAC with other frameworks. In addition, we presented research gaps by identifying and analyzing the limitations of the current frameworks. Lastly, challenges and possible opportunities for future research were highlighted. Keywords Beam alignment ·MmWave ·THz ·Compressive sensing ·Beam sweeping ·ML/AI ·Context information · Beam management 1 Introduction 1.1 Background The next-generation wireless network technology, 6G, has the potential to revolutionize the data user experience [1] as we know it from current technologies like 4G and 5G. The vast bandwidth availability in the mmWave and THz regions [2] paves the way for new applications that were previously unthinkable, such as holographic communication and applications in the metaverse domain. However, the main drawbacks and challenges are due to its small coverage range. The mmWave and THz signals suffer high path BSundire Madhekwana [email protected] BMuhammad Arslan Usman [email protected] Ahtisham Ayyub [email protected] Christos Politis [email protected] 1Faculty of Engineering, Computing and the Environment, Kingston University London, London KT1 2EE, UK 2Baseband Platform Software Department, Nokia, Ulm 89081, Germany losses. To address these challenges, beamforming technologies with large antenna arrays or multi-antenna systems are utilized to extend the coverage and enhance data throughput [3]. Multi-antenna systems, also known as Multiple-Input Multiple-Output (MIMO) systems [4], utilize multiple antennas at both the transmitter and receiver to significantly enhance communication performance. This technology is particularly crucial for beam alignment in Terahertz (THz) communications. The short wavelengths in the THz band enable a high number of antennas in a small space, resulting in narrow beams that pose challenges in maintaining a stable connection between the transmitter and receiver. By employing multiple antennas, MIMO systems can generate highly directional beams that can be precisely steered towards the intended receiver. This beamforming capability facilitates efficient signal transmission and reception, effectively overcoming the limitations of narrow beams and ensuring reliable communication within the THz spectrum. However, the use of multi-antenna systems at these frequencies makes the beams highly directional, so the beams need to be aligned before any link can be established. The beam alignment process establishes and maintains a communication link between a transmitter and a receiver by aligning the transmitter and receiver beams. This process becomes particularly crucial in the mmWave (30 GHz -300) GHz) 0123456789().: V,-vol 123 87 Page 2 of 53 S. Madhekwana et al. [5] and THz (0.1-10THz) [6] spectrum range because of the characteristics of these frequencies, including high path loss, vulnerability to blockages, and susceptibility to atmospheric conditions [7]. In lower-frequency wireless systems, omnidirectional antennas or simple sectorized antennas have been used, as the wider beam patterns can accommodate mobility and provide coverage over a larger area. However, using narrow beamforming antennas in mmWave and THz systems enables highly directional transmission, which offers several advantages, including increased spatial reuse, improved capacity, and reduced interference. Beam alignment plays a fundamental role in achieving the benefits of directional transmission in wireless communication systems. It involves two main processes: beam training and beam tracking. •Beam training - This is the first step in beam alignment, where the transmitter and receiver explore different beam directions to find the best beam pair with the highest received signal strength [8]. The transmitter typically sends a series of training signals with different beam directions, using a codebook or defined protocol, and the receiver measures the signal strength for each beam direction [9,10] and selects the best one. Through feedback or explicit signalling, the transmitter adjusts its beam direction to align with the receiver’s beam direction, which yields the strongest received signal. Codebook and protocol design in THz systems is essential to achieve robust beam training and overcome challenges such as beam squint, where beams optimized for the centre frequency LoS gain at other frequencies, causing misalignment and performance degradation. [11] introduced a specialized wideband codebook that ensures each beam maintains sufficient gain across the entire bandwidth, enabling low-complexity beam training that avoids full channel estimation and complex real-time optimization. By maximizing the worst-case wideband beam gain and tailoring spatial zones, this design delivers consistent signal quality even at coverage edges while simplifying beam alignment and reducing training overhead. Furthermore, THz massive MIMO systems require precise beam alignment to combat high-frequency challenges and severe path losses, yet traditional narrow beams are impractical under dominant LoS conditions and weak pilot signals. [12] introduces algorithms that generate wide-beam codewords with broad angular coverage, flat main lobes, and suppressed side lobes, enabling efficient beam training without full CSI and reducing real-time. •Beam tracking -[2,13]: This is performed to continuously update the beam directions as environmental conditions change beyond the initial alignment. This is particularly important in mobile scenarios or environments with moving obstacles. Many different methods can be used for beam tracking. [14] and [15] give such examples of methods that can be employed for Beam training and beam tracking. [14] introduces a unified 3D beam training and tracking procedure for THz communications that bypasses the need for real-time CSI. The authors used a hierarchical 3D codebook with wide beams for broad coverage and narrow beams for high gain, enabling a grid-based training protocol that efficiently identifies the optimal beam pair with far fewer tests than an exhaustive search. Once the beam pair is set, a grid-based hybrid tracking protocol employing two dynamic modes to refine and predict the beam. Whereas [15] introduced an exhaustive method to estimate angles for both direct LoS and Intelligent Reflecting Surface (IRS)-reflected links, due to the high complexity in THz massive MIMO systems, it proposes a low-complexity cooperative beam training scheme. This approach combines a partial search at the IRS–scanning key angle differences in the sine space–with a ternary-tree hierarchical search at the BS and user using two specialized codebooks to efficiently narrow down the best narrowbeam pair. Once training establishes an optimal beam pair, beam tracking continuously updates the pair based on received signal quality, ensuring sustained high beamforming gain in dynamic. Beam tracking helps maintain the alignment of the communication link, compensating for beam misalignment due to factors such as user mobility, blockages, or antenna misalignment. Beam alignment techniques can vary depending on system requirements, channel conditions, and available resources. Various frameworks and methods have been proposed to achieve efficient and accurate beam alignment [16]. Some of these frameworks include codebook-based techniques, compressed sensing-based approaches, machine learning-based methods, hybrid beamforming, channel estimation-based techniques, pilot-based beam alignment, and adaptive beam alignment. However, none of these frameworks have been standardized, implemented in real-world systems, or proven adequate for large-scale antenna arrays. The performance evaluation of beam alignment techniques involves several metrics, such as signal-to-noise ratio (SNR), bit error rate (BER), throughput, beam misalignment angle, beam training overhead, energy efficiency, and latency. Comparative analysis of different beam alignment techniques helps assess their strengths, limitations, and trade-offs across various scenarios and deployment contexts [10]. Challenges in beam alignment arise from the characteristics of mmWave and THz frequencies, including multi-path propagation, interference, hardware constraints, mobility, and environmental factors [2,10]. While mmWave and THz channels share several similar characteristics, the effects of path loss, atmospheric absorption, and scattering are significantly more pronounced in THz frequencies [17]. 123 Beam alignment for mmWave and THz: systematic... Page 3 of 53 87 Consequently, THz communication systems require more sensitive, rapid, and complex alignment techniques than mmWave due to the narrow beams and extreme signal attenuation. These challenges necessitate research and development efforts to overcome obstacles and improve the robustness and efficiency of beam alignment techniques. As wireless communication systems continue to evolve, integrating beam alignment frameworks with advanced networking technologies, such as 5G/6G and the Internet of Things (IoT), becomes essential. Cross-layer optimization, security considerations, machine learning advancements, and standardization efforts further contribute to the continuous improvement of beam alignment in wireless communication systems. Overall, beam alignment plays a critical role in enabling reliable and efficient communication in high-frequency wireless systems. Ongoing research and advancements in this field aim to address challenges, optimize performance, and facilitate deployment. Our motivation for this compact review of existing beam alignment frameworks is to explore unexploited opportunities and potential challenges. The systematic review aims to contribute to: •The knowledge of the design of alignment frameworks. •Improvement of beam alignment frameworks. The study intends to answer the following research questions (RQs), as shown in Table 1. 1.2 Principal contributions and organization We analyzed various existing beam management research works that cover beam alignment frameworks with a strong focus on large antenna arrays, mmWave, and THz systems. We investigated the studies in terms of their contribution to the areas depicted in the figure 1: The purpose of this study is to: •Develop a taxonomy of beam alignment frameworks and categorize existing beam alignment frameworks. •Compare them and identify their limitations. •Examine their prospects for standardization •Evaluate their suitability in meeting the challenges faced in beam alignment and the areas depicted in the figure 1. The conducted research examined the existing beam alignment frameworks in recognized journals and conferences. The developed taxonomy categorizes the beam alignment solutions and describes each framework to give an overview of existing difficulties and the latest existing techniques. All the selected frameworks have been evaluated in terms of beamforming architecture, beam alignment frameworks, Fig. 1 Focus of research area Contributions Environment, mobility, number of beams, performance metrics, key findings, etc. Additionally, the limitations of the frameworks were identified as well as the existing research gaps. Section 2 gives an overview of the previous works. Section 3 categorizes the results of existing solutions and techniques that help solve beam alignment challenges into five different classes. Section 4 discusses the latest research on beam alignment frameworks in detail. In this section, we review the research works within the defined frameworks, providing existing solutions and evaluating the identified research problems. Section 5 offers a qualitative comparison and outlines the limitations of the framework types. Section 6 presents future directions and opportunities. Finally, we conclude in Section 7. 1.3 Findings overview In this section, we provide an overview of the findings of the investigation. According to [19–21], beam alignment frameworks can be implemented in various ways: exhaustive searching, out-of-band information exchange via sensors mounted on the transmitter and receiver nodes, compressive sensing, machine learning, and Integrated Sensing and Communication (ISAC)-based beam alignment. These approaches can be used in numerous ways to develop beam alignment (BA) frameworks that address specific problematic scenarios. Table 2categorizes the BA frameworks identified based on their implementation features. To facilitate this classification, researchers decided to create the following categories: •Beam sweeping •Context Information •Compressive Sensing •ML/AI •ISAC beam alignment For this study, beam sweeping, context information, compressive sensing, ML/AI, and ISAC beam alignment were considered. We considered the beam sweeping technique as any method that uses either exhaustive beam searching, adaptive sweeping method, or hierarchical beam sweeping 123 87 Page 4 of 53 S. Madhekwana et al. Table 1 Research Questions Question Motivation RQ1: Which beam selection or alignment frameworks exist for mmWave and THz? The next-generation wireless network systems will use the vast bandwidth available in mmWave and THz regions with highly directional beams to compensate for the high path losses suffered by such signals. However, the Transmitter and receiver’s directional beams must be aligned [18]. The beam sweeping framework [16] was standardized for beam alignment in 5G NR with the assumption of a maximum of 64 beams. However, next-generation wireless system beyond 5G is anticipated to use very large antenna arrays with up to 1024 antenna elements and very directional beams. The beam-sweeping framework will not be adequate due to the high beam alignment latency, high energy consumption and high overhead that grows with an increasing number of beams. Although there is a substantial body of literature on this topic, further research is needed. This question is crucial for establishing existing frameworks and the challenges they have addressed, determining the status quo, identifying knowledge gaps, and exploring potential standardization solutions. RQ2: How do these Frameworks compare to each other? This question aims to compare the different frameworks to identify common issues that have been addressed and to highlight differences. Additionally, gaps are identified by showing the issues that have not been addressed by any of the established frameworks. RQ3: What are the challenges and limitations of these frameworks? Challenges in beam alignment arise from the characteristics of mmWave and THz frequencies, including multi-path propagation, interference, hardware constraints, mobility, and environmental factors. These challenges necessitate research and development efforts to overcome obstacles and improve the robustness and efficiency of beam alignment techniques. Although many of the challenges have been addressed by different approaches, some of these will spill over to 6G [10]. Specifically, the challenges and limitations that are not addressed by existing frameworks need to be identified through these questions. RQ4: What are the future directions and research opportunities? After establishing the existing solutions and their limitations, the future directions and research opportunities will be apparent, thereby answering this question. technique. Any technique that uses in-band or out-of-band signalling of position, direction, or other information to the transmitter or receiver was considered a context information technique. Compressive sensing techniques are those that exploit sparsity characteristics of the mmWave or THz channel. Approaches that use any of the machine learning techniques were classified as ML/AI. Approaches that combine any of the primary frameworks with Integrated Sensing and Communication (ISAC) were classified as ISAC-based beam alignment or hybrid. The summary of research distribution per category is depicted and summarized in figure 2. Table 2shows the studies that fall into each category and the distribution of the selected articles according to the framework categorization. Machine learning has the highest number of samples found. Figures 2and 3illustrate the distribution of publications by framework and year of publication, respectively. An initial classification was conducted to highlight key clusters within emerging beam alignment techniques. These Table 2 Beam Alignment Classification Classification References No.of Refs Compressive Sensing [22–31]10 Beam Sweeping [10,32–40]10 Context Information [41–55]16 Machine Learning [56–82]26 ISAC Beam Alignment [83–93]11 categories address various challenges by leveraging and capturing the unique characteristics of mmWave and THz signals. Notably, the data shows a significant increase in research activity from 2022 onward. It is also important to emphasize that frameworks involving Machine Learning (ML) and ISAC have become dominant in recent research, indicating their growing importance in this field. 123 Beam alignment for mmWave and THz: systematic... Page 5 of 53 87 Fig. 2 Distribution of Articles by Framework Fig. 3 Distribution of Articles by year 2 Related works There is a vast body of recent literature related to our work that has investigated beam management and alignment for mmWave and THz bands. These studies include [17,94– 100], and are described as follows: [17] and [94] highlighted major challenges in beam management for mmWave, discussing the trends and issues behind these challenges. Some of the identified challenges include inefficient beam sweeping, multi-panel optimization, high mobility, and DL/UL beam correspondence failures. The authors proposed potential solutions such as compressive sensing, machine learning (ML), reinforcement learning (RL), multi-base station cooperation, and robust codebook design. However, their work primarily focused on challenges related to standardized 5G NR beam management based on beam sweeping, and technical implementation was not thoroughly addressed. [95] Wang et al. provided a comprehensive overview of mmWave technologies in scenarios such as 5G networks, fixed wireless access (FWA), vehicle-to-everything (V2X), and indoor networks. The work highlighted challenges encountered in these scenarios, including high path loss, blockage, atmospheric absorption, and hardware limitations. Although the paper offered a broad overview of mmWave, it did not cover the THz bands or aspects like scalability, energy efficiency, and interoperability, and it lacked detailed technical depth in beam management. [96] Yi et al. presented an in-depth overview of beam training and tracking algorithms for mmWave in scenarios such as 5G networks, FWA, V2X, and indoor and urban environments. The study highlighted challenges such as high path loss, blockage, mobility, and real-time tracking and discussed how blind beam training, prior informationaided beam training, and ML-based mechanisms can reduce training overhead. The authors focused mainly on mmWave, with less emphasis on broader 5G/6G contexts, such as THz bands. However, aspects like low latency, beam management, scalability, and energy efficiency were not fully addressed. [97] Khan et al. provided extensive coverage of ML techniques specific to beam management for both mmWave and THz bands. The study explored how to address challenges like data availability, computational complexity and model generalization using supervised learning, reinforcement learning, and federated learning. Scenarios such as autonomous vehicles, smart cities, UAVs, and indoor networks were also considered. Despite the thorough exploration of ML applications in beam management, non-ML approaches were not covered. [98] Xue et al. focused on ML-based beam management for both mmWave and THz bands, examining current achievements with an emphasis on artificial intelligence (AI), integrated sensing and communication (ISAC), and reconfigurable intelligent surfaces (RIS). They also highlighted the use of multi-agent collaboration techniques like federated learning and transfer learning. Despite the extensive analysis of ML-based beam management, the paper did not provide a broader analysis of beam management, as it neglected advancements in non-ML-based approaches. [99] Häger et al. focused on design aspects that optimize beam training for mobile users in 6G mmWave networks. The authors examined the impact of mobility both during and after beam training, proposing a trade-off approach in the design of beam training systems. This approach balances the quality of the communication link, the delay introduced by the beam training process, and the coverage area. The design should incorporate a quality-of-service (QoS)-aware beam search algorithm that dynamically adjusts the beam training process according to the required link quality. Such algorithms enable a balance between speed and robustness, which is particularly crucial for mobile users who need quick yet reliable link establishment. However, the study primarily concentrated on design aspects, with less emphasis on 123 87 Page 6 of 53 S. Madhekwana et al. broad theoretical frameworks or scenarios involving nonmobile users. [100] presented a new approach to reducing multi-user interference in IRS-assisted massive MIMO systems. It uses a space-orthogonal scheme that decouples the precoder and decoder design. One part cancels interference using zero-forcing techniques. The other part maximizes the rate using singular value decomposition (SVD) and power allocation. To lower the complexity of joint beamforming, the paper introduces two IRS phase-shift design methods: water-filling segment matching (WSM) and phase iterative evolution (PIE). These methods balance performance with computational efficiency. The scheme achieves near-optimal sum-rate performance with much lower complexity than an exhaustive search. The paper’s strengths include breaking down a complex problem into simpler subproblems and using IRS-partial zero-forcing to improve spatial multiplexing gains. However, it relies on accurate channel models and dominant LoS assumptions, which may limit its use in scattering environments. Scalability remains a concern for very large systems. While existing research has made significant contributions in specific areas for instance, ML, we believe there is a gap in providing comprehensive coverage of beam alignment and management strategies. Our work aims to fill this gap by offering a broader analysis of advancements in beam management, exploring both ML/AI-driven techniques and traditional approaches. This wider scope allows us to address various methodologies and innovations, highlighting progress in all possible directions. 3 Beam alignment frameworks and taxonomy The selected articles were classified by constructing metadata forms to organize the details and include our defined remarks for those articles. We collected only the metadata that addressed our research questions, which included the title, research objectives, methodology, beamforming architecture, beam alignment frameworks environment, mobility, number of beams, performance metrics, etc. Figure 2 shows the distribution of the publications according to frameworks. An initial classification was made concerning our research topic to demonstrate outstanding clusters of the emerging beam alignment techniques categories solving different challenges by utilizing and capturing characteristics and aspects of the mmWave and THz signals. AI/ML dominates the ranks. 3.1 Beam sweeping We classified the published literature on beam sweeping into exhaustive beam sweeping and hierarchical beam sweeping. 3.1.1 Exhaustive beam alignment In an exhaustive beam-sweeping framework, both the transmitter and receiver scan through all the beams to identify the beam with the maximum power and most suitable for communication [10]. The exhaustive beam sweeping was standardized by the 3GPP for 5G NR in Release 15 [10,94]. The beam alignment overhead and delay increase with the number of beams when mmWave and THz are in operation. These challenges trigger the need for improvement in this framework. 3.1.2 Hierarchical and adaptive beam sweeping The research in this area investigates how the standardized beam-sweeping approach can be improved in terms of overhead, delay, and energy efficiency. The hierarchical framework’s beam sweeping is done adaptively. At first, wider beams are swept through, and a beam with higher power is then subsequently swept with narrower beams to determine the optimal beam accurately [34]. However, hierarchical beam sweeping is a common form of adaptive beam sweeping, allowing the system to progressively focus from wide to narrow beams. The implementation of beam sweeping can be codebook-based or hardware-based, e.g., where some antennas are switched off to produce large beams, and then all antennas are used for narrow beams. The main drawbacks of the framework are the inaccuracy caused by the wider beams that may not detect mobile stations at the cell edges [101] and the dependence on DL/UL beam correspondence, which introduces inflexibility and blockage [10]. 3.2 Context information As highlighted in beam sweeping-based BA, the overhead and delay grow with the number of beams used by the transmitter and receiver. Although the hierarchical approach reduces overhead and delay metrics, it is susceptible to errors and may not be suitable for applications like ultra-reliable low-latency communication (URLLC). The context information approach intends to reduce these metrics further by using out-of-band information that is communicated to both the receiver and the transmitter [41] to inform each other of their bearings and where to search for the beams from each other. The context information framework research can be further classified by the context information used to implement the framework, as shown in figure 4and the subsections below: 3.2.1 Location-aware beam alignment The location-aware frameworks exploit the position information of the base station and mobile station [41]. The base 123 Beam alignment for mmWave and THz: systematic... Page 7 of 53 87 Fig. 4 Beam Alignment Taxonomy station or mobile station derives the location information from GPS devices that are integrated into those entities. Once these entities obtain their positions, they beacon them towards the transmitter or receiver so that they know which area to search for suitable beams. 3.2.2 Camera-based alignment The camera-based framework uses the camera as a sensor for images of objects in the environment of the transmitter or receiver to determine the object sending the signal and its direction [19]. 3.2.3 Light detection and ranging Light detection and ranging (LiDAR) technology is used to acquire the contextual information of the transmitter and receiver [19,102]. The information is used to determine the direction of the beam. 3.2.4 Out-of-band measurements Out-of-band measurements is any type of signal that can be sent to the transmitter or receiver to predict the direction of the source of the signal [20,101]. Theoretically, a context information-based approach for beam alignment has a high potential to reduce latency, overhead, and energy consumption. However, the main shortcoming of such a framework is that the extra equipment needed for the in-band or out-of-band communication makes the equipment (base station and mobile station) more expensive. Hence, the industry may be reluctant to implement such an approach. 3.3 Compressive sensing (CS) CS is a technique for signal processing that makes it possible to reconstruct images or signals from a comparatively small number of samples or measurements [103,104]. CS relies on the sparsity principle and optimization algorithms for signal recovery. Sparse signals have a small number of non-zero components, while a significant number of the components are zeros or very close to zero. It is often quantified using the L1-norm of the signal, which is the sum of the absolute values of its components. Compressive sensing has applications in various fields, including image processing, medical imaging, and data compression. The equation below formalizes compressive sensing mathematically: y=x=a=a (see [105] for details) Where y is the low dimensional measurements, = random measurable matrix, = transform basis, a=sparse coefficients 123 87 Page 8 of 53 S. Madhekwana et al. The reconstruction of the signal problem will be solved as an optimization problem of the form: minal1,(1) such that: y=a(see [105] for details) Compressive sensing reduces the training overhead for beam alignment by leveraging the spatial clustering of multipaths in the channel [20]. The best beam is selected by probing and measuring a small set of randomly selected beams from the whole codebook. Although the compressive sensing technique may reduce overhead and energy consumption by probing fewer beams, it is complex due to the need for precise modelling and reconstruction algorithms and may face standardization and interoperability challenges. 3.4 Machine learning (ML) and artificial intelligence (AI) (ML/AI)) ML and AI are used to enhance initial beam alignment in wireless communication systems [106]. ML models are trained to estimate and predict wireless channel characteristics, such as signal strength, multipath propagation, and interference. These channel characteristics are critical for achieving optimal beam alignment. ML models learn from historical data, adapting to environmental factors like changes in surroundings, interference, or obstacles during wireless communication. ML/AI systems dynamically adjust beam parameters based on changing conditions, optimizing the initial alignment. Furthermore, weights and phase parameters can be adjusted for faster beam alignment. [107] presented the status quo, opportunities, and challenges for deep learning, including Convolutional Neural Networks (CNNs). Reinforcement learning trains systems to make decisions about initial beam alignment, learning optimal strategies through trial and error. ML and AI enable wireless communication systems to achieve faster, more accurate initial beam alignment, ensuring adaptability to diverse environmental conditions and enhancing overall reliability and performance. The machine learning and artificial intelligence (ML/AI) research work for beam alignment is further classified according to the machine learning model types, namely supervised, reinforcement, and deep learning, as depicted in figure 4. 3.4.1 Reinforcement Learning According to [108], Reinforcement learning is a method within AI systems where the system learns from its experiences to improve its decision-making. This approach emphasizes learning through trial and error, marking a significant shift in the capabilities of artificial intelligence systems. 3.4.2 Supervised learning Supervised learning involves using previously collected data, along with labeled information, to predict outcomes and through a training process, machine learning algorithms develop a function to anticipate output values. This system can then generate predictions for new input data after sufficient training, as defined by the authors of [109]. By comparing predicted outcomes with actual results, errors are identified and used to refine the model. In contrast, unsupervised learning is used when data lacks labels or classifications, aiming to uncover underlying patterns without predefined outputs by analyzing and deriving functions from unlabeled data, thus discovering hidden patterns through observation and analysis [109]. 3.4.3 Deep learning Deep learning uses computational models consisting of numerous layers to understand data by forming representations with various levels of complexity using neural networks [110]. The main drawback of the ML/AI approach is its high complexity due to the loads of data needed for training the models and the need for ongoing model improvements as the very expensive hardware necessary for running inference engines. 3.5 ISAC beam alignment frameworks Integrated Sensing and Communication (ISAC) integrates wireless communication and sensing in one system. Its goal is to enhance efficiency by sharing spectrum, hardware, and signals between the two tasks. ISAC is particularly valuable in advanced networks like 5G and 6G beam alignment frameworks, where high data rates and real-time environmental awareness are crucial. ISAC can be combined with different other beam alignment frameworks to form subcategories. The three key subcategories of ISAC are ISAC and Machine Learning (ML), ISAC and Beam Sweeping, and ISAC and Context Information. 3.5.1 ISAC and machine learning Machine Learning (ML) enhances ISAC by using data models to improve sensing and communication. For example, [31] integrates ISAC with computer vision and extended Kalman filters (EKF). Resulting in improved initial access (IA) and beam tracking in UAV networks The dual identity association (DIA) approach further enhances beam-tracking 123 Beam alignment for mmWave and THz: systematic... Page 15 of 53 87 shorter switch delays result in higher throughput. The study also suggests that the current RSRP-based beam selection might not always choose the best beam, indicating a need to reassess beam management criteria in 5G mmWave networks. Finally, [39] presents a beam alignment strategy for THz communication systems, using an iterative process to optimize beam directions and maximize data rates. It tackles challenges like severe path loss and the need for precise alignment in high-frequency environments. Despite its benefits in data rate and accuracy, the approach faces challenges with computational complexity and real-time feasibility. The study advances beam management in next-generation wireless networks, especially in the context of emerging THz technologies. Beam sweeping exhibits various challenges and considerations. Firstly, it introduces latency, which, while tolerable in some applications, becomes critical in low-latency scenarios. This latency issue is compounded by the overhead incurred in terms of time and frequency resources, impacting the network’s overall efficiency. Moreover, beam sweeping is sensitive to environmental changes, with factors like obstacles and interference affecting alignment accuracy and speed, potentially leading to suboptimal performance in dynamic settings. Scalability is a concern, particularly in scenarios with many users or devices, as managing multiple simultaneous beams and ensuring efficient alignment for each user poses challenges. Furthermore, the constant adjustment of beams through sweeping can contribute to increased energy consumption, raising concerns about the network’s sustainability and energy efficiency. Finally, the implementation and management of beam sweeping procedures, especially in millimetre-wave and THz frequencies, involve complexity, requiring sophisticated algorithms, signalling mechanisms, and hardware, which could escalate deployment and maintenance costs. 4.2 Context information frameworks In wireless communication, context information encompasses details about the environment, including device locations, obstacle presence, device movement, and overall network conditions, which is valuable for optimizing communication parameters. Beam alignment is essential in millimetre-wave and THz communication, utilizing directional beams through beamforming to enhance signal focus and reliability. A context-aware beam alignment approach integrates additional contextual information, such as precise device locations, obstacle awareness, and device dynamics (especially for mobile devices), facilitating proactive adjustments for optimal beam direction and interference reduction. The context information approach often employs adaptive algorithms to monitor and analyze contextual factors, dynamically adjusting beamforming parameters for better communication performance. Furthermore, contextual data plays a crucial role in Machine Learning (ML). It provides additional information that helps models make more accurate predictions. This enables informed decisions about beam alignment and adaptive responses to changing conditions. Tables 5and 6show the collected metadata of the selected Context Information frameworks research studies that addressed our research questions, which include year, title [Reference], Research objectives, Methodology, Beamforming Architecture, Beam Alignment frameworks Environment, Mobility, Number of beams, Performance metrics, etc. [42] explains the transmission model for beamforming using a Butler matrix and parameters characterizing radio spatial propagation. It introduces the beam alignment problem as a multi-hypothesis testing issue, discussing various detection frameworks, including fixed-length detection and variable-length detection (SCET), with SCET presented as an adaptive and robust improvement over fixed-length methods. The paper details an experimental setup at TU Dresden for mmWave experimentation with hybrid beamforming, evaluating beam selection methods in different scenarios. The results demonstrated the adaptability and efficiency of SCET and SCT in single-path and multi-path scenarios compared to fixed-length methods. The paper also highlighted the adaptability and efficiency of the SCET framework in realistic mmWave scenarios. The study acknowledged its robustness and superiority over fixed-length methods. [41] addresses the beam alignment challenges in millimetre-wave (mmWave) communication, where high data transmission and reduced interference are crucial. The proposed algorithm leverages the location information of mobile users and potential reflectors to efficiently align beams in mmWave-band communications. The key idea is to enable the base station and mobile user to work together to scan a limited number of beams within the error boundary of the noisy location information. This initial set of beams guides the search for future beams, reducing alignment overhead, power consumption, and time resources. The algorithm’s effectiveness is demonstrated through simulation results, demonstrating improved system performance even when dealing with noisy location data. The paper compares the proposed approach with existing methods, highlighting its adaptability, efficiency, and its reduced complexity. Overall, the research aims to enhance the beam alignment process in mmWave and THz communication, particularly in mobile environments, by exploiting location-aware coordination. [43] explores "Frequency-Dependent Beamforming”, leveraging fixed and variable length tests with a focus on mmWave communication systems. It introduces an exhaustive search (ES) based testing approach, including a Variable Length Testing Scheme (VLT) utilizing ES and a Generalized Likelihood Ratio Test (GLRT) considering cross-correlation values 123 87 Page 16 of 53 S. Madhekwana et al. and noise variance estimates, along with a termination threshold determined by the likelihood of a false alarm. The Frequency Dependent Beamforming (FDB) section employs a linear system model with Rician fading, adapting GLRT for hypothesis testing, and introducing a length-dependent threshold based on an F-distribution. Simulation results compare Fixed Length Testing (FLT) with true time delay for different scenarios, emphasizing the advantages of variable length testing, particularly in higher SNR regimes. The conclusion highlights the time efficiency of the proposed frequency-dependent beamforming, achieving power loss measures in medium to high SNR, and suggests further research for improving time efficiency and reducing power losses in challenging conditions. [44] focuses on beam alignment strategies for mmWave massive MIMO systems to enhance communication performance, exploring various codebooks, including orthogonal, hierarchical, and frequency-dependent types. It discusses the optimal beamformer choice, focusing on the orthogonal codebook after a specific number of observations (N) and emphasizing the minimum required samples for the Least Squares (LS) solution. Simulations considered a channel with multiple paths and a receiver with a 64-element uniform linear array. The paper compares fixed and variablelength beam selection techniques using the Butler codebook and introduces the Sequential Competition and Elimination Test (SCET) for adaptive test length and robustness. The hierarchical codebook (MLVL) and Frequency-Dependent Beamforming (FDB) codebook are evaluated, showing that variable-length detection, especially with SCET, enhances adaptivity, resulting in rate gains and delay reduction. Different codebook types impact detection performance differently, with orthogonal codebooks benefiting from elimination mechanisms and frequency-dependent codebooks eliminating the need for beam switching. [45] investigated the initial beam acquisition/alignment in mmWave communication systems. The authors proposed a training protocol that uses a frequency-selective beam probing network to scan all beamformers from a particular codebook. This approach exploits the sparse nature of mmWave channels, mapping each beamformer to different frequencies. The paper compares two hardware designs for parallel beam training and evaluates their feasibility. The results demonstrated that frequency-selective beam probing outperforms exhaustive search in terms of effective transmission rates in rapidly changing environments, crucial for mmWave systems with large beamforming codebooks and short coherence times. The study highlights the importance of reducing timing overhead in communication systems, especially in scenarios with high mobility and spatial resolution. The proposed approach demonstrates advantages in time efficiency and transmission rates. It offers potential benefits in sparse scattering environments common in mmWave communications. The paper suggests exploring continuous pilot subcarriers during data transmission for improved channel tracking for future work. [46] introduces an innovative rapid access method for mmWave systems in cellular networks, aiming to reduce the complexity of beam training. This is achieved through the utilization of retro-directive arrays (RDA) and a ZadoffChu (ZC) sequence-based preamble. The proposed method involves two stages: in the first part, the base station (BS) performs transmit (Tx) beam sweeping with activated RDA, and if a mobile station (MS) is detected in any of the beam directions, the RDA retransmits the signal to the BS. In the second part, the BS transmits the preamble through the chosen Tx beam, and the MS performs Rx beam sweeping to measure the SNR. The paper demonstrated a notable decrease in complexity when compared to the traditional exhaustive search approach. Simulation results confirmed the technique’s ability to accurately detect beam alignment parameters, highlighting its potential in addressing the complex challenges of beam training. The proposed approach, leveraging retro-directive arrays and Zadoff-Chu sequences, offers a promising solution for efficient initial access in mmWave networks. [47] tackles the beam alignment challenge in mmWave communication networks by proposing a groundbreaking framework aided by Ultra-wideband (UWB). The innovation lies in integrating UWB technology to estimate optimal angles for beam alignment, eliminating the necessity for exhaustive space searching and addressing latency and complexity issues. The Multi-Frequency MUSIC (MF-MUSIC) algorithm expands the traditional MUSIC algorithm into the frequency domain, enabling accurate angle estimation with a small number of UWB antennas. The paper recognizes practical challenges with limited antennas on CommercialOff-The-Shelf (COTS) devices and provides a solution, enhancing its real-world applicability. While comprehensive numerical evaluations and comparisons are strong, they lack real-world validation and rely on assumptions. The main contributions of [48] include a novel approach to coordinate computation using accelerometer rotation angles, addressing azimuth errors. The paper proposes an efficient association of mobile stations (MS) with base stations (BS) through a transmitted uplink control signal, considering power, response vectors, and the geometric channel. Two broadcasting approaches, single and multi-broadcasting, facilitate data dissemination to multiple MS users. Performance evaluation indicates advantages such as significantly reduced access times, high success rates, and energy efficiency in comparison to existing schemes. Limitations include potential challenges in real-world implementation and the developmental stage of mmWave technology. In their future work, they suggest exploring outages caused by blockage and mobility. 123 Beam alignment for mmWave and THz: systematic... Page 17 of 53 87 Table 5 Context Information based Framework research studies Ref Objectives Method Archite cture Frame work Environ ment Mobility Beam Nr. [42] Evaluates the performance of a sequential competition and Elimination Test (SCET) - a variable length beam selection framework Experim entally Analog BF Context information based LOS and NLOS N/A 64 [41] Determines the optimal beams by utilising the user’s position, the base station’s location, or certain obstructions on the channel Simul ation Analog BF Context information based NLOS N/A 64 [43] Achieves beam alignment by equipping the transmitter with a frequency-dependent beamformer that can simultaneously verify all spatial angles and, consequently, any beamformer that might come from a comparable codebook. Simula tion Hybrid BF Frequencydependent beamforming NLOS N/A 16,32, 64 [44] Compares the beam alignment accuracy and acquisition time of three types of orthogonal butler matrix strategy, hierarchical/multilevel search and frequency-dependent beamforming framework using both fixed and variable length Simul ation Hybrid BF Context information based NLOS N/A 8,16, 32,64 123 87 Page 18 of 53 S. Madhekwana et al. Table 5 continued Ref metrics Scenario Frequen cy Merits Demerits Beam Pattern [45] Used a strategy that simultaneously probes all beams from a codebook to determine AoA or AoD and exploits the sparsity of mmWave channels. The method maps every beamformer of the codebook to multiple frequencies using a frequency-selective beam probing network and spectral analysis at the receiver to determine favourable beamformers AoA or AoD Simul ation Analog BF Frequency-selective beam probing NLOS N/A 8 [46] Proposed a retrodirective array (RDA)-based initial access method to lower overhead associated with the beam training overhead. The RDA transmits the incoming signal back to the source with a directivity that matches the size of the antenna array. Simul ation Analog BF Retrodire ctive array (RDA) NLOS N/A BS:256, MS:8 [47] Proposed a mmWave communication architecture with jointly situated UWB antennas to find the optimal angles for mmwave beam alignment. Simul ation Analog BF co-located UWB antennas NLOS N/A - [48] When a new MS enters a cell, it scans the channel for a beacon indicating the bearing of the BS, broadcast by a nearby MS that has accomplished beam association. Then, using a digital compass, the location of the MS is reflected in the adjusted coordinates of the BS. Simul ation digital BF Probing beacon NLOS N/A - [42] SNR and Latency Cellular, UAV, FWA mmWave Reduces training overhead, adaptable to varying channel conditions Computational complexity and hardware impairments impact accuracy Orthogonal beams using a Butler Matrix setup 123 Beam alignment for mmWave and THz: systematic... Page 19 of 53 87 Table 5 continued Ref metrics Scenario Frequen cy Merits Demerits Beam Pattern [41] Through put Cellular, V2X, UAV mmWave Reduced alignment overhead, enhanced alignment accuracy Computational overhead requires precise location data Adaptive beams based on location data [43] SNR Cellular, UAV mmWave, sub-THz Reduces beam alignment time, higher SNR, adaptable to channel conditions Computational complexity in multi-path scenarios, high hardware cost Adaptive beam patterns basedontimedelay across different frequencies [44] SNR Cellular, V2X mmWave Reduces beam-training overhead, adaptable to SNR conditions Computational complexity, reliance on prior knowledge of SNR Orthogonal beams via Butler Matrix, Frequency-dependent beams [45] SNR and throughput Cellular, V2V, FWA mmWave Reduces beam alignment time, better performance in rapidly changing environments Increased hardware complexity due to delay elements Frequency-dependent beam patterns utilizing OFDM modulation [46] SNR,DP Cellular mmWave Faster beam alignment with reduced complexity, accurate detection Higher hardware cost, complexity in managing retrodirective arrays Retrodirect ive beam patterns [47] SNR, Throughput, AC Cellular mmWave Significant improvement in AoA estimation accuracy with fewer antennas, reduction of communication delays Performance drops in NLoS scenarios, requires a calibration step for Highly directional [48] SNR, EE, ST Cellular mmWave Fast access times, high success rates, reduced energy consumption Higher cost, complex beamforming structure, azimuth tilt errors Omni-directional and directional BF= Beam Forming, RI=Rank indicator, PMI=precoding matrix Index, CQI =channel quality indicator, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, BDA = beam detection accuracy, MDP = miss detection probability, DP = Detection Probability, C = complexity, LOS = line of sight, NLOS = Non line of Sight, SNR = Signal to noise ratio, UAV = Unmanned aerial vehicle, FWA = fixed wireless access, V2X = Vehicle to anything, MS = mobile station, BS = basestation 123 87 Page 20 of 53 S. Madhekwana et al. Table 6 Continued Context Information based Framework research studies Ref Objectives Method Archite cture Frame work Enviro nment Mobil ity Beam Nr. [49] The study used stochastic geometry and Student’s t-distribution statistics analytical model to examine the effect of uplink power control on the downlink beam alignment errors in an mmWave network as they depend on several link factors Anal ytical Analog BF Uplink power control NLOS N/A - [50] The Authors proposed an energy-controlled THz pulse-level beam switching called TRPLE. The impulse radio, which emits pulses lasting femtoseconds, allows beam direction control at the pulse level rather than the packet level. TRPL framework solves the pulse-to-pulse and symbol-to symbol separation parameters to maximize the data rate while meeting the interference requirements Simul ation Analog BF pulse-level beam switching NLOS N/A - [61] The study evaluated the performance of beamforming leveraging space division multiple access (SDMA) depending on the design, i.e. array pattern, number of array elements and angle of separation. Signal of interest and signal of no interest scenarios were defined for different base stations’ angles of separation and number of antenna elements. Simula tion Analog BF locationaware LOS N/A 128 [51] Efficient initial access and wireless power |transfer to energy-neutral devices Simula tion Analog BF Location LOS, NLOS N/A varia ble [52] Reduce beam alignment time for mmWa ve communications using third-party camera data Experi mental Analog BF camera LOS, NLOS Yes 31 [53] Reduce training overhead and enhance beamforming accuracy Simula tion, Experi mental Hybrid BF Location LOS, NLOS Yes Varia ble [54] Use CV and ML for beam management, reducing CSI reliance Simula tion Hybrid BF Camera LOS Yes varia ble [55] Create a Digital Twin (DT) to improve beam management Experi mental, Simulation Hybrid BF Camera LOS, NLOS Yes Not specified [49] SNR, CP Cellular mmWave Reduces UE power consumption, maintains downlink SNR in NLOS Degradation in LOS SNR with FPC, beam alignment errors in LOS Directional, sectorbased 123 Beam alignment for mmWave and THz: systematic... Page 21 of 53 87 Table 6 continued Ref metrics Scenario Frequen cy Merits Demerits Beam Pattern [50] DR, SNR Small cells, WLAN THz High data rate (167 Gbps), effective for long distances (up to 20m), efficient multiplexing High complexity in beam control, limited by pulse separation, molecular absorption Highly directional with narrow beams [61] SNR Cellul ar, D2D, 5G Ultra-Dense Networks mmWa ve Significant interference suppression, enhanced spatial multiplexing for ultra-dense networks Requires accurate positioning, computational complexity increases with more dense environments Directional with spatial separation [51] PG, Power Density Indoor WPT mmWa ve Efficient power delivery using beam diversity; scalable for dense environments Dependence on environment awareness; beam sweeping increases setup time and energy consumption Directional [52]AC,Time Consumption Indoor, Smart City mmWa ve Reduces beam alignment time by up to 1/50 compared to traditional methods Reliance on camera availability and placement, processing overhead Directional [53] Throug hput, Training overhead Cellul ar mmWa ve Reduces real-time channel training, improves communication rates even with moderate location errors Relies on location accuracy, environmental changes affect performance Directional [54]AC, Overhead Cellul ar, UAV mmWa ve Reduces overhead and improves real-time beam alignment Depends on camera data and environmental conditions Directional [55] AC Cellul ar mmWa ve Higher accuracy in beam alignment reduces overhead Camera dependency, high computational overhead Digital twin with adaptive mapping BF= Beam Forming, RI=Rank indicator, PMI=precoding matrix Index, CQI =channel quality indicator, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, BDA = beam detection accuracy, MDP = miss detection probability, DP = Detection Probability, C = complexity, LOS = line of sight, NLOS = Non line of Sight, SNR = Signal to noise ratio, UAV = Unmanned aerial vehicle, FWA = fixed wireless access, V2X = Vehicle to anything, MS = mobile station, BS = base station, AC = accuracy, PG = path gain, WPT = wireless power transfer 123 87 Page 22 of 53 S. Madhekwana et al. [49] investigates the impact of uplink power control on downlink beam alignment errors in mmWave cellular networks, employing stochastic geometry and Student’s t-distribution statistics. The authors created an analytical framework that considers link parameters, uplink power control, and SNR coverage. The results demonstrate a substantial reduction in user equipment (UE) power consumption through effective uplink power control without compromising downlink SNR coverage. The study highlights fractional power control for uplink pilot signal transmission, an underexplored area in mmWave systems. Key contributions include exploring the influence of uplink power control on downlink beam alignment errors, introducing an analytical model, and showcasing significant UE power consumption reduction with minimal effects on downlink SNR coverage. Acknowledging limitations related to assumptions, the paper suggests future research directions involving real-world validation, dynamic power control strategies, and extensions to additional scenarios for a comprehensive understanding of power control mechanisms in emerging wireless communication technologies. In summary, the research provides valuable insights into optimizing power consumption in mmWave cellular networks while ensuring satisfactory downlink SNR coverage. [50] focuses on advancing Terahertz (THz) communication by developing and optimizing a Medium Access Control (MAC) protocol framework. The authors aim to tailor the MAC framework for the downlink in THz networks, considering THz channel characteristics and emphasizing pulse-level beam-switching using large antenna arrays based on graphene for macro-scale communication. The proposed framework aims to establish a "pseudo wired" wireless link through focused transmission via line-of-sight or reflected paths. The paper explores key components, including the optimization of Inter-Pulse Separation (IPS) and Inter-Symbol Separation (ISS), revising transmission scheduling, discussing beam acquisition strategies, addressing beam switching delay, and presenting numerical results for both LOS and NLOS. [61] investigates the application of beamforming and spatial multiplexing in the context of 5G Ultra-Dense Networks (UDN), particularly emphasizing the transition to mmWave. The study highlights adaptive beamforming (AB) and spatial signal processing as crucial elements in overcoming mmWave limitations. The paper explores adaptive beamforming scenarios, addressing challenges like beam alignment and presenting analog, digital, and hybrid methods. The research evaluates spatial multiplexing performance through simulations employing the Least Mean Squares (LMS) algorithm. It focuses on the interference suppression rate (ISR) under various conditions. Key contributions include assessing location-aware beamforming and developing a simulation model to understand the impact of angular separation and antenna elements on interference suppression. In conclusion, the research significantly advances the understanding of beamforming and spatial multiplexing in 5G UDN, providing valuable insights for further exploration and practical implementation. Context-based beam alignment is an optimization technique in wireless communication systems that utilizes contextual information such as user locations, environmental conditions, interference levels, and network dynamics to dynamically adjust beam parameters. While this approach aims to enhance system performance, reliability, and efficiency, it faces several common limitations. These include the complexity introduced to system design, challenges in achieving real-time adaptation, reliance on accurate contextual information, concerns regarding interoperability across diverse systems, and potential security and privacy issues associated with sensitive contextual data. To address these challenges, common recommendations include the development of robust algorithms, integration with artificial intelligence and machine learning for improved adaptability, and collaborative standardization efforts to ensure consistency, real-world validation to assess effectiveness in various environments, and the implementation of strong security measures, such as encryption and authentication, to safeguard sensitive information. 4.3 Compressive sensing (CS) frameworks CS is employed to take advantage of the mmWave channels’ intrinsic sparsity. The sparsity characteristics allow for the accurate recovery of dominant paths with fewer training resources. It also reduces the training overhead by providing a means to recover essential channel information with a sparse set of measurements. Additionally, compressive sensing can be combined with other techniques (e.g., Kalman Filtering [25]) to adapt to the dynamic nature of time-varying channels, ensuring accurate beam alignment over changing conditions. Tables 7and 8show a list of studies included in this section. The selected studies in Tables 7and 8collectively provide a contemporary understanding of the status quo in Compressive Sensing (CS) beam alignment. These studies showcase the potential of CS-based frameworks in addressing challenges associated with mmWave communication. Notably, the integration of CS with Kalman Filtering, as demonstrated by Kun-Hsien Lin, presents a promising approach for accurate channel recovery with reduced overhead. Erfan Khordad’s exploration of beam alignment in dynamic environments emphasizes the importance of adaptability and a switching mechanism between beam training and tracking to balance accuracy and overhead. [25]’s work further underscores the potential of CSKalman Filtering integration in achieving accurate beam alignment with significantly lower Mean Squared Error 123 Beam alignment for mmWave and THz: systematic... Page 23 of 53 87 (MSE) compared to traditional methods. However, the status quo also reveals critical aspects that warrant attention. The impact of CS-based frameworks in highly mobile scenarios and their robustness under extreme channel dynamics require further investigation. Real-world deployment challenges, such as hardware constraints and synchronization issues, are recognized but not extensively discussed. Adaptability to diverse environments and comprehensive analysis of CS performance under varying signal-to-noise ratios (SNR) remain areas for more in-depth investigation. The trade-off between accuracy and overhead reduction, as well as comparative assessments with alternative beam alignment techniques, need more detailed consideration. In conclusion, the status quo of CS-based beam alignment is promising, with notable advancements in integration strategies and performance improvements. Yet, there exists a need for continued research to address challenges in mobility, real-world implementation, environmental adaptability, SNR sensitivity, and a comprehensive understanding of trade-offs in practical mmWave communication scenarios. 4.4 ML/AI frameworks The application of machine learning (ML) and artificial intelligence (AI) principles in beam alignment involves a data-driven approach, utilizing historical data to train models on relevant features, such as signal strength and environmental factors. Supervised learning is employed for tasks with labelled datasets, while unsupervised learning aids in understanding data patterns. Reinforcement learning trains systems to make beam alignment decisions through trial and error. ML models adapt in real time to dynamic environmental changes, and optimization algorithms adjust beamforming parameters for optimal performance. Deep learning, particularly Convolutional Neural Networks (CNNs), may be used for image processing and pattern recognition. The principles also include the creation of adaptive systems that dynamically adjust beam parameters based on changing conditions and predictive analytics for forecasting potential misalignments. Together, these principles enhance the automation, optimization, and adaptability of beam alignment processes, improving efficiency and reliability in various technologies. Tables 9,10,11 and 12 show the collected metadata of the selected ML/AI framework-based research studies that addressed our research questions, which include year, [Reference], Research objectives, Methodology, Beamforming Architecture, Beam Alignment frameworks Environment, Mobility, Number of beams, Performance metrics, etc. The work of [56] explores beam alignment in mmWave massive MIMO systems using ML. The proposed framework using ML, introduces a neural network (NN) trained offline in simulated environments that uses partial beams to forecast the beam distribution vector. Based on the NN predictions, all users’ beams are then simultaneously aligned. The framework demonstrated superior overall performance, considering total training time slots and spectral efficiency. The approach eliminates the need for historical user location data during NN training, contributing to reduced system overhead. The research indicates that ML techniques can significantly enhance beam alignment by improving efficiency and reliability. [57] proposes a new approach to overcome the challenges of beam alignment in mmWave systems, which are crucial for addressing propagation issues at high frequencies. The paper introduces a distributed beam-alignment strategy utilizing an adversarial multi-armed bandits (MAB) approach, leveraging a single bit of feedback indicating whether the SINR surpasses a predetermined cutoff. The main breakthrough is a redesigned reward function that draws inspiration from the mmWave channels’ sparse structure, efficiently reinforcing good beam directions and penalizing poor ones. The proposed algorithm, MEXP3, builds upon the exponential weights (EXP3) algorithm, demonstrating theoretical guarantees for regret. The results of the mmWave numerical simulations with user mobility showcase the superior performance of MEXP3 when compared to EXP3 and other policies, emphasizing its adaptability to dynamic wireless environments. [58] suggests BsNet, a deep learning-based beam selection technique. The motivation stems from the challenges posed by the narrow and highly directional beam characteristics of mmWave bands, making beam alignment costly and challenging. BsNet addresses this by treating beam alignment as an image reconstruction problem, leveraging deep neural networks (DNNs) for beam domain image reconstruction. The approach involves offline training and online prediction stages, significantly decreasing the online beam selection overhead. The paper introduces the concept of eigen-beam extraction and uses a learning-based method for BsNet training, achieving remarkable results in terms of scalability, robustness, and performance. Simulation results demonstrate BsNet’s superiority over existing methods, such as Modified Rosenbrock’s Direct Search (MRDS) and local learning-based clustering algorithm with feature selection (LLCfs), offering high spectral efficiency while minimizing the search overhead. The proposed approach exhibits promising potential for efficient beam selection in mmWave communication scenarios, demonstrating adaptability to various environments and scenarios. [59] proposes a machine learning solution for beam management in 5G NR using geolocation side information. The method models the mapping between the geolocations of user equipment (UE) and the beams or cells that serve them in a multiuser, multicell environment by using support vector machines (SVMs). Furthermore, a multiuser scheduling approach is shown that reduces real-time channel state information (CSI) feedback 123 87 Page 24 of 53 S. Madhekwana et al. Table 7 Compressive Sensing Framework research studies Ref Objectives Method Archite cture Frame work Environ ment Mobility Beam Nr. [22] Demonstrated that limiting randomness in compressed sampling to local sets achieves robustness to structured errors due to carrier frequency offset Simula tion Analog BF Compress ive sensing LOS N/A 64 [23] Aims to optimize beam alignment in multiuser mmWave MIMO systems under practical low SNR conditions by investigating compressive sensing techniques, proposing a trial-based protocol and a novel deterministic construction for the CS sensing matrix, and demonstrating superior performance compared to existing methods. Simula tion Analog BF Compress ive sensing LOS, NLOS Yes Variable [24] Develop a deterministic sensing matrix for millimeter-wave beam alignment using a Kronecker-based structure, exploiting the sparse nature of mmWave channels. Simulation Analog BF Compress ive sensing NLOS, LOS N/A variable [25] Aimed to address the challenges of tracking and training beams in a time-varying millimetre-wave channel while minimizing overhead by combining Compressed Sensing (CS) and Kalman Filtering. Simula tion Analog BF Compress ive sensing NLOS, LOS N/A Variable [26] To reduce overhead in beam alignment for mmWave/THz systems using compressive sensing. Simula tion Hybrid BF Compress ive Sensing LOS N/A 32 [27] To reduce computational complexity in mmWave beam alignment using a deterministic, sparse sensing matrix Simula tion Hybrid BF Compres sive Sensing LOS N/A variable 123 Beam alignment for mmWave and THz: systematic... Page 31 of 53 87 more data. However, the approach is heavily reliant on highquality training data, and its computational complexity may limit real-time applications in resource-constrained environments. Additionally, scaling to larger networks with more APs and users presents challenges in managing multiple beams simultaneously. [81] presents a deep learning method called deep regularised waveform learning (DRWL) to improve beam prediction in non-cooperative mmWave systems, addressing the challenge of limited training data. Using data augmentation techniques like cyclic time shift (CTS) and signal splicing, DRWL predicts optimal beams without coordination between transmitter and receiver, reducing the need for exhaustive beam sweeping. The method improves beam prediction accuracy even with fewer training samples, maintaining high performance by leveraging waveform characteristics. However, the model introduces computational complexity, making real-time deployment difficult in resource-constrained environments, and its performance depends on input signal quality. Further testing is needed to confirm its effectiveness in dynamic, real-world scenarios. [82] explores using computer vision (CV) and machine learning (ML) to improve mmWave beam management. Traditional systems rely on channel state information (CSI), leading to high overhead and delays. This new approach uses camera data to select beams without needing CSI. The framework focuses on improving efficiency, robustness, and scalability with ML models. Simulations show better beam alignment and reduced computational costs. However, the system depends on reliable camera data, which may be impacted by blocked views or poor conditions. Labeling data for ML training is also challenging. The system faces further issues with multiuser environments, integration complexity, and environmental sensitivity. The research papers spanning [56–60,62–69], [70–82] collectively provide a thorough examination of ML/AI-based beam alignment techniques. These methodologies, exemplified by innovations like HBA (Hierarchical Beam Alignment), are designed to optimize beamforming for improved communication performance, demonstrating promising outcomes in terms of regret performance, scalability, and beam alignment latency reduction. However, the papers identify several limitations, including reliance on specific channel models that may not fully represent real-world mmWave complexities, posing challenges for generalization to diverse scenarios. The impact of mobility, dynamic obstacles, and environmental changes on beam alignment is also insufficiently addressed. Practical implementation aspects, such as computational complexity and energy efficiency, are often overlooked, with deployment challenges needing more attention. The trade-off between exploration and exploitation in learning algorithms, especially in dynamic mmWave channels, requires further exploration. In essence, while ML/AIbased beam alignment techniques exhibit significant progress in optimizing mmWave communication, challenges persist in generalization, practical implementation, and adaptability to dynamic scenarios, necessitating future work to address these limitations and explore real-world deployment challenges, mobility, and energy efficiency for a more holistic solution in mmWave communication systems. 4.5 ISAC beam alignment frameworks Integrated Sensing and Communication (ISAC) is an emerging technology. It combines wireless communication and sensing within one system. ISAC offers new ways to optimize beam alignment. Which is essential for efficient data transmission and environmental awareness. Machine Learning (ML), beam sweeping, and context information frameworks can enhance ISAC. These methods boost performance in dynamic environments. The following research highlights advancements and innovations in ISAC, providing an overview of key efforts driving this field forward, see Tables 13 and 14. In [83], Zhiqiang Xiao et al. presented a beamforming technique for mmWave integrated sensing and communication (ISAC) systems, aiming to create a cost-effective analog beamforming approach that enables a MIMO ISAC system to simultaneously sense targets and communicate with user equipment (UE). Their research introduces a flexible double-beam codebook to create separate beams for sensing and communication. They also propose simultaneous beamsweeping schemes. These include single and double-beam methods to optimize the process and reduce overhead. [84] presents an ISAC system using Reconfigurable Intelligent Surfaces (RIS). It improves communication and target sensing efficiency. The authors propose a Simultaneous Beam Training and Target Sensing (SBTTS) scheme. This scheme distinguishes RIS from targets by accumulating energy in different domains. It reduces training overhead. A Positioning and Array Orientation Estimation (PAOE) scheme further improves efficiency with a fast search algorithm. The system performs well in both line-of-sight and non-line-of-sight scenarios. However, it faces complexity, noise sensitivity, and pilot overhead challenges. Deployment is also difficult due to large networks’ hardware needs, real-time control, and scalability. [85] introduces the Integrated Sensing and Communication (ISAC) beam alignment method for Terahertz (THz) networks. The method, called Joint Synchronization Signal Block (SSB) and Reference Signal (RS)-based Sensing (JSRS), reduces line-of-sight (LoS) blockages and corrects beam misalignment caused by user mobility. With high-resolution THz sensing, JSRS detects and prevents misalignment early, improving network coverage. The paper shows JSRS enhances beam alignment and coverage, especially in urban vehicle-to-everything 123 87 Page 32 of 53 S. Madhekwana et al. Table 9 ML/AI based Framework research studies Ref Objectives Method Archite cture Frame work Enviro nment Mobil ity Beam Nr. [56] The article suggested a method for beam alignment with partial beams that uses machine learning and doesn’t require any prior knowledge, like user location data. Simula tion Hybrid BF ML-NN NLOS N/A 5 [57] To be considered aligned, a signal must reach a certain SINR threshold before it can be considered aligned by a distributed beam alignment scheme. Simula tion Hybrid BF ML-RL NLOS Yes 32 [58] Proposed a deep neural network (DNN) solution that required no knowledge of the channel but treated the beam selection as an image reconstruction problem and the DNN is used to construct a beam domain image. The solution consists of offline training and online prediction. Eigen-beam theorem entry treated as starting value for beam domain image reconstruction (BDIR). The off-line Eigen-beam extraction reduces the overhead of online-beam alignment. Simula tion Analog and Hybrid BF ML-DNN NLOS N/A 32,64, 128,256 [59] Proposed a mmWave beam alignment solution that uses geolocation side information and leveraging Support Vector Machine (SVM) model for mapping UE location and covering beam in a multi-user multi-cell scenario. Subsequently, the beam assignment information from user-adjacent cells helps to reduce the latency for channel state information (CSI) feedback. Simula tion Analog and Hybrid BF MLSupervised LOS N/A 8 [60] Proposed ML-based beam alignment framework that is trained to forecast the best access point (AP) and the most suitable beam choice given an UE’s GPS coordinates. Whereby the GPS coordinates are reported by the UE or by some other technique. Simula tion Hybrid and Analog BF ML-DCNN NLOS N/A 64 [62] The authors derived an AP beam-training framework by using the spatial correlation between various beams in addition to the ability of the Deep Convolutional Neural Network (DCNN) to retrieve features The AP framework only probes a fixed set of the entire beam space and determines the best Simula tion Analog BF DCNN framework NLOS N/A 128 [63] The authors proposed a CNN strategy for performing beam selection between the transmitter and receiver. The employed CNN skip connections and hyperparameter optimization to balance between accuracy and computational complexity. Simula tion Analog BF CNN framework NLOS Yes 61 123 Beam alignment for mmWave and THz: systematic... Page 33 of 53 87 Table 9 continued Ref metrics Scenario Frequen cy Merits Demerits Beam Pattern [56] SNR, SE Cellular mmWa ve Reduces training time, does not require user location data, efficient beam alignment High computational complexity, requires simulated training environments Directional [57] SNR Cellular mmWa ve Low-latency beam alignment, optimal performance in sparse mmWave channels High computational complexity, requires tuning of reward parameters for varying conditions Directional Beam Pattern [58] SNR Cellular mmWa ve Reduces beam search overhead by up to 90%, maintains 99% of spectral efficiency compared to exhaustive search Requires extensive training, potential computational complexity in real-time applications Directional Beam Pattern [59] SNR Cellular, V2X mmWa ve Reduces latency by up to 50%, decreases signaling overhead by 34%, and improves beam pairing accuracy. High reliance on geolocation data, complexity in multi-cell scenarios, computational overhead in training Directional Beam Pattern [60] BPA Cellular mmWa ve Reduces search space for optimal AP and beam alignment, robust against dynamic scatterers and location uncertainty High computational complexity, only considers MISO scenario, requires more development for UE-side beam training. Directional Beam Pattern [62] SNR, BPA, Laten cy Cellular, V2X mmWa ve Reduces computational complexity by 15%, high accuracy (70.4%) in beam selection Requires extensive training dataset, computational complexity in real-time deployment Directional, Adaptive [63] SNR Cellular mmWa ve Reduces signaling overhead by 90%, high prediction accuracy with 10% beam soundings Computational complexity, requires careful model tuning to prevent over fitting Direction BF= Beam Forming, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, BPA = beam Prediction accuracy, SE = Spectral Efficiency, ML = Machine Learning, RL = Reinforced Learning , SNR = Signal to Noise Ration, NN = neural Networks, DNN= Deep NN, DCNN Deep Convolutional NN, EE =Energy Efficiency, UDN =Ultra-Dense Networks (UDN), D2D = Device-to-Device (D2D), SBP =Spatial Beam Prediction,TBP=Temporal Beam Prediction, DL = Deep Learning, LSTM = Long and Short Memory, SL=supervised Learning 123 87 Page 34 of 53 S. Madhekwana et al. Table 10 Continued ML/AI based Framework research studies - continued Ref Objectives Method Archite cture Frame work Enviro nment Mobil ity Beam Nr. [64] The authors proposed to train Matrix factorization ML and Nonnegative Matrix factorization models by probing a tiny subset of a massive beam’s codebook of TX and RX to predict the SNRs of the beams of the codebook. Subsequently, they derived equations for optimizing the matrix factorization method Simula tion Analog BF ML-NN NLOS N/A 1024 [65] The authors studied a site-specific-sounding codebook and designed a NN frame that uses measurements from the sounding codebook to predict the best beam. The NN framework model captured the site-specific beams that are associated with the particular characteristics of the propagation environment. Simula tion Analog BF ML - NN NLOS N/A 128 [66] The authors proposed a gridless beam alignment framework that predicts near-optimal beam weights from a continuous beam set using unsupervised training that uses few beams for channel sounding. Experim ental Analog BF ML-NN NLOS N/A variable [67] Proposed a deep reinforcement learning-based beam alignment framework that can switch between different beam alignment methods depending on the radio channel thereby achieving good power and spectral efficiency by using channel circumstances to regulate active RF chains Simula tion Hybrid BF ML-DRL NLOS Yes 64 123 Beam alignment for mmWave and THz: systematic... Page 35 of 53 87 Table 10 continued Ref Objectives Method Archite cture Frame work Enviro nment Mobil ity Beam Nr. [68] Improved the Kolmogorov model (KM) by introducing discrete monotonic optimization to reduce complexity and improve scalability. Furthermore, testing advanced hypotheses can also be conducted using the Kolmogorov-Smirnov (KS) criterion, which requires no subjective threshold setting compared to frequency estimation used in KM Simula tion Analog BF ML-KM NLOS N/A 16 [69] The authors proposed a hierarchical beam alignment algorithm by using knowledge about the inter-beam correlation structure and channel variation. The authors conceptualised the beam alignment problem as a "multi-armed bandit" problem and solved by characterizing the beam correlation structure in the multi-channel as a multimodal function. Additionally, use the previously gained knowledge of the channel fluctuations to properly consider reward uncertainty Simula tion Analog BF MLMAB NLOS N/A 512 [70] Optimize beam management and CSI acquisition using ML in X-MIMO systems for 6G networks Simula tion Hybrid BF ML-NN LOS, NLOS Yes 8, 16, 32 123 87 Page 36 of 53 S. Madhekwana et al. Table 10 continued Ref metrics Scenar io Frequen cy Merits Demerits Beam Pattern [64] SNR Cellular mmW ave, THz Reduces beam sweeping overhead by 3x, achieves over 90% alignment accuracy High training complexity, dependent on site-specific training data Directional Beam Pattern [65] SNR Cellular mmWa ve Reduces sweeping overhead by 21x, achieves 0.32 dB of the theoretical upper bound High computational complexity, requires site-specific training Directional [66] SNR Cellular mmWa ve Reduces training overhead, adapts to environmental changes, maximizes both EE and SE High computational complexity, sensitive to changes in real-world environments, relies on deep learning Directional Adaptive [67] EE, SE Cellular mmWa ve Reduced computational complexity, improved beam alignment performance High complexity in previous KM models, computational limitations hierarchical beam pattern [68] SNR Cellul ar, WLAN, FWA, UAV mmWa ve Significantly reduced computational complexity, improved beam alignment performance High computational complexity with larger antenna arrays Overlapping, hierarchical [69] SNR UDN, D2D mmWa ve Improved interference suppression, location-aware communication, optimized beamforming for spatial multiplexing in dense environments Requires accurate positioning and synchronization, computational complexity in high-density environments Directional [70] RSRP, SNR, SE Cellul ar mmWa ve Significant gains in spectral efficiency, adaptability, reduced feedback requirements Increased computational complexity, challenges in real-time operations and generalization. Dynamic and Adaptive BF= Beam Forming, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, BPA = beam Prediction accuracy, SE = Spectral Efficiency, ML = Machine Learning, RL = Reinforced Learning , SNR = Signal to Noise Ration, NN = neural Networks, DNN= Deep NN, DCNN Deep Convolutional NN, EE =Energy Efficiency, UDN =Ultra-Dense Networks (UDN), D2D = Device-to-Device (D2D), SBP =Spatial Beam Prediction,TBP=Temporal Beam Prediction, DL = Deep Learning, LSTM = Long and Short Memory, SL=supervised Learning 123 Beam alignment for mmWave and THz: systematic... Page 37 of 53 87 Table 11 Continued ML/AI based Framework research studies Ref Objectives Method Archite cture Frame work Environ ment Mobil ity Beam Nr. [71] Optimize beam tracking and rate adaptation in mmWave systems using reinforcement learning Simula tion Hybrid BF ML-RL LOS, NLOS Yes Varia ble [72] Optimize initial access and beam alignment for security and efficiency using deep learning Simula tion Real-world data Hybrid BF ML-DL LOS, NLOS Yes Limite d set [73] Optimize beam training at THz frequencies using MAB algorithms for fast and adaptive beam selection Simula tion Hybrid BF ML-RL LOS, NLOS Yes Variab le [74] Enhance beam management with AI/ML to improve prediction accuracy, reduce latency, and optimize performance in 5G-Advanced Simula tion Survey Hybrid BF MLSBP ML-TBP) LOS, NLOS Yes Variab le [75] Optimize beam alignment in mmWave systems using a deep learning-based hierarchical framework Simula tion Hybrid BF ML-DL LOS, NLOS Yes Variab le [76] Improve energy efficiency in mmWave networks using the A2C learning framework for beam selection and power optimization Simula tion Hybrid BF ML-RL LOS, NLOS Yes Variab le 123 87 Page 38 of 53 S. Madhekwana et al. Table 11 continued Ref metrics Scenar io Frequen cy Merits Demerits Beam Pattern [71] Throug hput, Outage duration Cellul ar mmWa ve 182% throughput gain, reduced outage duration, low overhead, adaptable to mobility Computational complexity in real-time operations, tuning of ATS parameters for performance. Dynamic and Adaptive [72] BPA, RSRP Cellul ar mmW ave 75% reduction in beam search space, 99.66% accuracy, improved security, energy efficiency Reduced RSRP by 6.5 dB for security, trade-off between signal strength and security Dynamic and Adaptive [73] Adaptation Speed, SE Cellul ar THz Improved spectral efficiency, faster adaptation, and superior performance with contextual data. Higher complexity for contextual algorithms, latency trade-offs with Probing-LinUCB Dynamic and Adaptive [74] BPA RSRP Laten cy Cellul ar mmWa ve 63.5% increase in prediction accuracy, reduced overhead, and better high-mobility performance High computational complexity, difficulties with model generalization, privacy concerns Dynamic and Adaptive [75] BPA, SE, Overhead Cellul ar mmWa ve Higher accuracy with fewer measurements, reduced overhead, enhanced spectral efficiency High computational complexity, need for extensive training data, challenges in real-time deployment Dynamic and Adaptive [76] EE, Overhead, SNR Cellul ar mmWa ve More than doubles energy efficiency, adapts to various conditions, and integrates into Open RAN High computational complexity, extensive training data, scalability issues in multi-user environments Dynamic and Adaptive BF= Beam Forming, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, SE = Spectral Efficiency, ML = Machine Learning, RL = Reinforced Learning , SNR = Signal to Noise Ration, NN = neural Networks, DNN= Deep NN, DCNN Deep Convolutional NN, EE =Energy Efficiency, UDN =Ultra-Dense Networks (UDN), D2D = Device-to-Device (D2D), SBP =Spatial Beam Prediction,TBP=Temporal Beam Prediction, DL = Deep Learning, LSTM = Long and Short Memory, SL=supervised Learning, BPA=Beam Prediction Accuracy 123 Beam alignment for mmWave and THz: systematic... Page 39 of 53 87 Table 12 Continued ML/AI based Framework research studies Ref Objectives Method Archite cture Frame work Enviro nment Mobility Beam Nr. [77] Optimize beam selection in mmWave MIMO systems using a deep contextual bandit learning framework Simulation Hybrid BF ML-DL LOS, NLOS Yes Variab le [78] Improve beam tracking efficiency and reduce overhead using LSTM in multi-cell mmWave environments Simula tion Analog and Digital BF ML-DL and -LSTM LOS, NLOS Yes Variab le [79] Improve beam selection efficiency and reduce overhead in dynamic, blockage-prone environments Simula tion Hybrid BF ML-NN LOS, NLOS Yes Variab le [80] Use ML algorithms to optimize beam management in D-MIMO systems and reduce measurement complexity Simula tion Hybrid BF ML-SL LOS, NLOS Yes Variab le [81] Improve beam prediction with limited data using deep learning in non-cooperative mmWave environments Simula tion Hybrid BF ML-DL LOS, NLOS Yes variab le [82] Use CV and ML for beam management, reducing CSI reliance Simula tion Hybrid BF ML-SL LOS Yes variab le 123 87 Page 40 of 53 S. Madhekwana et al. Table 12 continued Ref metrics Scenario Frequen cy Merits Demerits Beam Pattern [77] Throug hput, Overhead Cellul ar mmWa ve Improved beam selection efficiency, reduces exhaustive search, adaptable to dynamic environments High computational complexity, requires significant training data, limited real-world validation Dynamic and Adaptive [78] Power Consumption, Overhead, BPA Cellul ar mmWa ve Reduced power consumption and overhead, high accuracy in beam tracking, adaptable to environments Computational complexity, reliance on training data, fixed vs adaptive constraint complexities Dynamic and Adaptive [79] Over head, SNR, BPA Cellul ar mmW ave 90% accuracy, reduces beam search time by 86.4%, performs well in dynamic environments like urban V2X Computational complexity, data dependency, parameter optimization for various environments. Dynamic and Adaptive [80] Over head, BPA Cellul ar mmW ave Reduces measurement overhead, improves beam selection accuracy, efficient in dynamic environments Data dependency, computational complexity, challenges in scaling to large networks Dynamic and Adaptive [81] BPA, Overhead Cellul ar mmWa ve High prediction accuracy with limited training samples, reduced beam sweeping overhead High computational complexity, performance dependent on signal quality, limited real-world testing Dynamic and Adaptive [82] BPA, Overhead Cellul ar UAV mmWa ve Reduces overhead and improves real-time beam alignment Depends on camera data and environmental conditions Predefined beam codebook (YOLO, etc.) BF= Beam Forming, RSRP = Reference Signal Received Power, RSS = Received Signal Strength, SE = Spectral Efficiency, ML = Machine Learning, RL = Reinforced Learning , SNR = Signal to Noise Ration, NN = neural Networks, DNN= Deep NN, DCNN Deep Convolutional NN, EE =Energy Efficiency, UDN =Ultra-Dense Networks (UDN), D2D = Device-to-Device (D2D), SBP =Spatial Beam Prediction,TBP=Temporal Beam Prediction, DL = Deep Learning, LSTM = Long and Short Memory, SL=supervised Learning, BPA=Beam Prediction Accuracy 123 Beam alignment for mmWave and THz: systematic... Page 47 of 53 87 Table 15 continued Feature/ Framework Beam Sweeping Context Information Compressive Sensing ML/AI ISAC Frameworks Robustness Limited robustness in dynamic environments Can enhance robustness by considering environmental factors Robust in sparse channel conditions but may struggle in highly dynamic scenarios Can adapt and improve robustness over time with continuous learning High, sensitive to environment but improved beamforming accuracy and highly adaptable Standardization Potential Generally simpler and widely used, making it easier to standardize across different systems Standardization may be feasible for certain aspects, but context information can be diverse and application-specific May face challenges in standardization due to variations in algorithms and reconstruction methods. Standardization is challenging due to the diversity of machine learning models and the need for tailored solutions Moderate to High, reliant on 5G like standards andnlikely to be adopted in RIS aided 6G Standards Interoperability Generally, more interoperable as it relies on basic beam adjustment mechanisms Interoperability can be a challenge if different systems use diverse context information sources and processing methods Interoperability challenges may arise due to variations in sensing hardware and algorithms. Interoperability is a concern due to variations in machine learning models, architectures, and training data High, scals to large networks with multiple reflecting surfaces and UEs Scalability Highly scalable as it involves basic beam adjustments, suitable for various deployment scales Scalability challenges may arise if context information sources become too complex or diverse Scalability depends on the efficiency of sensing hardware and algorithms, potentially facing limitations in large-scale deployments Scalability can be an issue due to the computational demands, especially during the training phase High asit integrates well with existing mmWave setups in large scale systems Network Scenarios All All All All All 123 87 Page 48 of 53 S. Madhekwana et al. 5 Comparison and limitations of existing frameworks In this section, RQ2 is addressed. Table 15 shows a qualitative analysis of performance metrics used to compare all the frameworks identified in this review. In comparing the frameworks for beam alignment, each framework offers distinct advantages and challenges. Beam Sweeping, characterized by its systematic adjustment of beam direction, is simple to implement and widely used, making it easily scalable and interoperable. However, its limited robustness in dynamic environments and higher latency in finding optimal alignment may be drawbacks. Context Information, which incorporates additional environmental data, enhances robustness but may suffer from complexity and interoperability challenges, particularly with diverse sources of context information. Compressive sensing, relying on sparse channel exploitation, offers robustness in certain conditions but requires complex algorithms and may face challenges in standardization and interoperability due to variations in hardware and methods. ML/AI-based approaches leverage machine learning for adaptive strategies, promising improved robustness over time. Still, they demand high computational resources and may struggle with standardization and interoperability due to diverse models and training data. ISAC systems present a promising solution for future networks, combining communication and sensing. They offer benefits like better efficiency and performance. However, challenges such as computational complexity, infrastructure requirements, and real-time processing must be addressed. As research progresses, ISAC will likely play a major role in shaping 6G and other advanced networks. Machine learningbased and ISAC-based frameworks are both scalable and interoperable with other frameworks, however, they are heavy in terms of computation. Despite heavy computational resource requirements and complexity, they have a potential for standardization. In essence, while each framework presents unique strengths, they also pose distinct complexities and limitations that are shown in Table 16. 6 Future directions and research opportunities This section will address RQ4. The 3GPP Release 15 introduced the beam sweeping-based protocol for 5G to address the beam alignment [10,94]. However, as a higher frequency spectrum and larger antenna arrays are deployed in 5G and 6G, this approach won’t be adequate. The beam-sweeping approach won’t address all the challenges that are associated with beam alignment. Instead, the overhead, complexity and latency associated with beam sweeping scale up with the size of the beam space [10]. Some of the challenges and gaps that need to be solved in addressing beam alignment were identified in [10,111–113] and can be summarized as follows: 6.1 Dynamic and mobile environments Many existing beam alignment frameworks primarily focus on static scenarios and may not adequately address the challenges posed by dynamic environments and mobile users. Research is needed to develop beam alignment techniques that can efficiently track and adapt to Fast-moving users or changing environmental conditions, ensuring continuous and reliable connectivity [10,112,113]. 6.2 Robustness against environmental factors Environmental factors, such as blockages, reflections, diffraction, and atmospheric conditions, can significantly impact the performance of beam alignment frameworks. Developing robust beam alignment techniques that can handle varying environmental conditions, including adverse weather conditions, dense urban environments, and indoor scenarios, is an important research area [113]. 6.3 Interference mitigation Interference from neighbouring systems or coexisting networks can hinder beam alignment performance. Research is needed to develop interference-aware beam alignment frameworks that can effectively mitigate interference effects and maintain high-quality communication links in crowded spectrum environments [94], [111]. 6.4 Energy efficiency considerations Beam alignment techniques often involve energy-intensive operations, such as beam training and tracking. Considering the energy constraints of battery-powered devices and the growing demand for energy-efficient communication systems, research is needed to explore energy-efficient beam alignment methods and protocols without compromising the performance and reliability of the communication links [112]. 6.5 Standardization and interoperability While several beam alignment frameworks have been proposed, there is a lack of standardization and interoperability across different systems and vendors. Efforts are required to develop common standards and protocols for beam align123 Beam alignment for mmWave and THz: systematic... Page 49 of 53 87 ment, facilitating seamless integration and interoperability among different devices, networks, and technologies [113]. 6.6 Real-world deployment and validation Many existing beam alignment frameworks have been evaluated through simulations or limited experimental setups. More research is needed to validate these frameworks in real-world deployments, considering practical constraints, real-life channel conditions, and diverse deployment scenarios. This would provide valuable insights into the actual performance, limitations, and scalability of beam alignment techniques [10]. 6.7 Security and privacy considerations Beam alignment frameworks may be vulnerable to security threats, including eavesdropping, spoofing, or unauthorized beam manipulation. Further research is needed to investigate security and privacy aspects related to beam alignment and develop robust mechanisms to ensure secure and private communication in beamforming-based systems [111], [113]. Addressing these research gaps will contribute to the development of more efficient, reliable, and adaptive beam alignment frameworks, paving the way for the successful deployment of mmWave and THz communication systems in various scenarios and applications. 7 Conclusion The SLR was conducted on published research focusing on beam alignment and initial access frameworks for mmWave and THz. The SLR identified the most relevant 73 research papers according to the author’s knowledge during the review period. 596 papers were initially selected using the defined search strings from digital libraries such as IEEE, Springer, Elsevier, and Francis & Taylor. Subsequently, a three-stage filtering mechanism was applied to the title, abstract, and full text, including the assessment of paper quality. The chosen studies were then explored, sorted and evaluated according to the defined RQs. In this study, we presented state-of-the-art frameworks for beam alignment in mmWave and THZ, and we explained the taxonomy of the beam alignment frameworks by classifying them into beam sweeping-based, context information-based, compressive Sensing-based machine Learning/Artificial Intelligence (ML/AI) -based and ISAC-based frameworks. To improve beam alignment in terms of latency reduction, energy consumption reduction and overhead, the existing literature body was explored to identify the most common beam alignment frameworks. While each framework presents unique strengths, they also pose distinct complexities and limitations. Beam Sweeping and Context Information offer simplicity and robustness enhancements but may struggle with dynamic environments and interoperability issues. Compressive sensing provides robustness in sparse channels but demands complex algorithms and faces standardization challenges. ML/AI approaches promise continuous improvement in robustness but come with high computational demands and potential interoperability hurdles. ISAC systems present a promising solution for future network beam management, combining communication and sensing. They offer benefits like better efficiency and performance. However, challenges such as computational complexity, infrastructure requirements, and real-time processing must be addressed. Thus, the choice among these frameworks hinges on factors such as deployment environment, computational resources, and the need for adaptability and standardization. In addition, we presented research gaps by identifying and analyzing the limitations of the current frameworks. Lastly, challenges and possible opportunities for future research were highlighted. Acknowledgements I thank Nokia for supporting this work. Authorcontributions Sundire Madhekwana wasresponsible forresearch, preparation, presentation and critical evaluation of the works presented in this SLR. Muhammad Arslan Usman contributed towards the research planning, supervision and technical editing. Ahtisham Ayyub contributed towards data visualisation and presentation of the articles this SLR covers. Christos Politis provided his oversight, supervision and mentorship in critically reviewing this article for its scientific rigour. Data Availability No datasets were generated or analysed during the current study. Declarations Competing interests The authors declare no competing interests. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecomm ons.org/licenses/by/4.0/. References 1. Uwaechia, A. N., & Mahyuddin, N. M. (2020). A comprehensive survey on millimeter wave communications for fifth-generation 123 87 Page 50 of 53 S. Madhekwana et al. wireless networks: Feasibility and challenges. IEEE Access, 8, 62367–62414. 2. Xue, Q., Ji, C., Ma, S., Guo, J., Xu, Y., Chen, Q., & Zhang, W. (2024). A survey of beam management for mmwave and thz communications towards 6g. IEEE Communications Surveys & Tutorials., 26(3), 1520–1559. 3. Zhang, J., Björnson, E., Matthaiou, M., Ng, D. W. K., Yang, H., & Love, D. J. (2020). Prospective multiple antenna technologies for beyond 5g. IEEE Journal on Selected Areas in Communications, 38(8), 1637–1660. 4. Ning, B., Tian, Z., Mei, W., Chen, Z., Han, C., Li, S., Yuan, J., & Zhang, R. (2023). Beamforming technologies for ultra-massive mimo in terahertz communications. IEEE Open Journal of the Communications Society, 4, 614–658. 5. Al-Falahy, N., & Alani, O. Y. (2019). Millimetre wave frequency band as a candidate spectrum for 5g network architecture: A survey. Physical Communication, 32, 120–144. 6. Petrov, V., Pyattaev, A., Moltchanov, D., & Koucheryavy, Y. (2016). Terahertz band communications: Applications, research challenges, and standardization activities. In: 2016 8th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT), pp. 183–190. IEEE. 7. Wang, X., Kong, L., Kong, F., Qiu, F., Xia, M., Arnon, S., & Chen, G. (2018). Millimeter wave communication: A comprehensive survey. IEEE Communications Surveys I& Tutorials., 20(3), 1616–1653. 8. Narengerile, N. (2022). Channel estimation and beam training with machine learning applications for millimetre-wave communication systems. 9. Attaoui, W., Bouraqia, K., & Sabir, E. (2022). Initial access i& beam alignment for mmwave and terahertz communications. IEEE Access, 10, 35363–35397. 10. Heng, Y., Andrews, J.G., J., M., V., V., A., A., L., N.B., & Zhang, J.C. (2021). Six key challenges for beam management in 5.5 g and 6g systems. IEEE Communications Magazine 59(7). 11. Ning, B., Mei, W., Zhu, L., Chen, Z., & Zhang, R. (2024). Codebook design and performance analysis for wideband beamforming in terahertz communications. IEEE Transactions on Wireless Communications. 12. Ning, B., Wang, T., Huang, C., Zhang, Y., & Chen, Z. (2023). Wide-beam designs for terahertz massive mimo: Sca-atp and ssarv. IEEE Internet of Things Journal, 10(12), 10857–10869. 13. Duan, Q., Kim, T., & Ghauch, H. (2021). Km learning for millimeter-wave beam alignment and tracking: Predictability and interpretability. IEEE Access, 9, 117204–117216. 14. Ning, B., Chen, Z., Tian, Z., Han, C., & Li, S. (2021). A unified 3d beam training and tracking procedure for terahertz communication. IEEE Transactions on Wireless Communications, 21(4), 2445–2461. 15. Ning, B., Chen, Z., Chen, W., Du, Y., & Fang, J. (2021). Terahertz multi-user massive mimo with intelligent reflecting surface: Beam training and hybrid beamforming. IEEE Transactions on Vehicular Technology, 70(2), 1376–1393. 16. Giordani, M., Mezzavilla, M., & Zorzi, M. (2016). Initial access in 5g mmwave cellular networks. IEEE communications Magazine, 54(11), 40–47. 17. Tataria, H., Shafi, M., Molisch, A. F., Dohler, M., Sjöland, H., & Tufvesson, F. (2021). 6g wireless systems: Vision, requirements, challenges, insights, and opportunities. Proceedings of the IEEE, 109(7), 1166–1199. 18. Yi, W., Zhiqing, W., & Zhiyong, F. (2024). Beam training and tracking in mmwave communication: A survey. China Communications. 19. Zecchin, M., Mashhadi, M. B., Jankowski, M., Gündüz, D., Kountouris, M., & Gesbert, D. (2022). Lidar and position-aided mmwave beam selection with non-local cnns and curriculum training. IEEE Transactions on Vehicular Technology, 71(3), 2979–2990. 20. Ali, A., González-Prelcic, N., & Heath, R. W. (2017). Millimeter wave beam-selection using out-of-band spatial information. EEE Transactions on Wireless Communications, 17(2), 1038–1052. 21. Li, X., Fang, J., Duan, H., & Li, H. (2019). A sparse encoding and phaseless decoding approach for fast mmwave beam alignment. ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 22. Myers, N.J., & Heath, R.W. (2019). Localized random sampling for robust compressive beam alignment. ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 23. Khordad, E., Collings, I. B., Hanly, S. V., & Caire, G. (2022). Compressive sensing-based beam alignment schemes for time-varying millimeter-wave channels. IEEE Transactions on Wireless Communications, 22(3), 1604–1617. 24. Khordad, E., Collings, I.B., & Hanly, S.V. (2019). A kroneckerbased sparse compressive sensing matrix for millimetre wave beam alignment. 2019 13th International Conference on Signal Processing and Communication Systems (ICSPCS). IEEE. 25. Lin, K. H., & Liu, K. H. (2022). A novel beam alignment scheme for mobile millimeter-wave communications based on compressed sensing aided-kalman filter. IEEE Open Journal of the Communications Society, 3, 1515–1527. 26. Masoumi, H., Verhaegen, M., & Myers, N.J. (2023). In-sector compressive beam alignment for mmwave and thz radios. arXiv preprint arXiv:2308.13268 27. Khordad, E., Collings, I.B., & Hanly, S.V. (2019). A kroneckerbased sparse compressive sensing matrix for millimeter wave beam alignment. In: 2019 13th International Conference on Signal Processing and Communication Systems (ICSPCS), pp. 1–5. IEEE. 28. Myers, N. J., Mezghani, A., & Heath, R. W. (2018). Swift-link: A compressive beam alignment algorithm for practical mmwave radios. IEEE Transactions on Signal Processing, 67(4), 1104– 1119. 29. Khordad, E., Collings, I. B., Hanly, S. V., & Caire, G. (2022). Compressive sensing-based beam alignment schemes for time-varying millimeter-wave channels. IEEE Transactions on Wireless Communications, 22(3), 1604–1617. 30. Yan, H., & Cabric, D. (2019). Compressive initial access and beamforming training for millimeter-wave cellular systems. IEEE journal of selected topics in signal processing, 13(5), 1151–1166. 31. Song, Y., Ge, C., Qiu, L., & Zhang, Y. (2023). 2ace: Spectral profile-driven multi-resolutional compressive sensing for mmwave channel estimation. In: Proceedings of the Twentyfourth International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing, pp. 41–50. 32. Fernandes, F., Rom, C., Harrebek, J., & Manchón, C.N. (2022). Improving beam management signalling for 5g nr systems using hybridbeamforming. IEEE Wireless Communications and Networking Conference (WCNC). IEEE. 33. Li, Y. N. R., Gao, B., Zhang, X., & Huang, K. (2020). Beam management in millimeter-wave communications for 5g and beyond. IEEE Access, 8, 13282–13293. 34. Qian, J., Masouros, C., Tanabe, K., Sasaki, E., Zein, N., & Marumoto, T. (2022). Beam-pattern assisted low-complexity beam alignment for fixed wireless mmwave xhaul. ICC 2022-IEEE International Conference on Communications. IEEE. 35. Giordani, M., Polese, M., Roy, A., Castor, D., & Zorzi, M. (2018). Initial access frameworks for 3gpp nr at mm-wave frequencies. 17th Annual Mediterranean Ad Hoc Networking Workshop (MedHoc-Net). IEEE. 123 Beam alignment for mmWave and THz: systematic... Page 51 of 53 87 36. Khirwadkar, P., Shahsavari, S., Hücümeno˘glu, M.C., & Pal, P. (2023). Max-min hierarchical codebooks for millimeter-wave channel sensing with bandit algorithms. In: 2023 57th Asilomar Conference on Signals, Systems, and Computers, pp. 387–392. IEEE. 37. Ram, A., López, M., Rodr, I., Sørensen, T. B., Barrio, S. C., Padilla, P., Valenzuela-Valdés, J. F., Mogensen, P., et al. (2024). Fr2 5g networks for industrial scenarios: Experimental characterization and beam management procedures in operational conditions. IEEE Transactions on Vehicular Technology., 73(9), 13513–13525. 38. Baron-Hyppolite, A., Abreu, J.V., Santos, J.F., DaSilva, L.A., & Kibilda, J. (2023). Adaptive beam management for secure mmwave communications using software-defined radios. In: MILCOM 2023-2023 IEEE Military Communications Conference (MILCOM), pp. 243–244. IEEE. 39. Huang, W., Zhang, W., & Wang, C. (2023). Iterative achievable rate adaptive beam alignment for thz communication systems. In: 2023 8th International Conference on Communication, Image and Signal Processing (CCISP), pp. 450–456. IEEE. 40. Feng, Y., Wei, J., Dinh, P., Ghoshal, M., & Koutsonikolas, D. (2023). Beam management in operational 5g mmwave networks. In: Proceedings of the 7th ACM Workshop on Millimeter-Wave and Terahertz Networks and Sensing Systems, pp. 13–18. 41. Orikumhi, I., Kang, J., Park, C., Yang, J., & Kim, S. (2018). Location-aware coordinated beam alignment in mmwave communication. 2018 56th Annual Allerton Conference on Communication, Control, and Computing (Allerton). IEEE. 42. Marandi, M.K., Wei, S., Khodapanah, B., Rave, W., & Fettweis, G. (2022). Experimental evaluation of a variable length beam selection framework in a usrp based testbed with mmwave frontends and butler matrices. IEEE International Symposium on Phased Array Systems I& Technology (PAST). IEEE. 43. Jans, C., Rave, W., & Fettweis, G. (2022). Fast beam alignment via true time delay frequency dependent beamforming using fixed and variable length tests. IEEE International Symposium on Phased Array Systems I& Technology (PAST). IEEE. 44. Marandi, M.K., Jans, C., Rave, W., & Fettweis, G. (2021). Evaluation of detection accuracy and efficiency of considered beam alignment strategies for mm-wave massive mimo systems. 55th Asilomar Conference on Signals, Systems, and Computers. IEEE. 45. Jans, C., Song, X., Rave, W., & Fettweis, G. (2020). Frequencyselective analog beam probing for millimeter wave communication systems. IEEE Wireless Communications and Networking Conference (WCNC). IEEE. 46. Kim, Y., Lee, H., & Cho, Y.S. (2020). Fast initial access technique for millimeter-wave systems using retrodirective arrays. International Conference on Information and Communication Technology Convergence (ICTC). 47. Wang, J., Chen, X., Liu, X., & Dudek, G. (2021). Uwb-assisted fast mmwave beam alignment. ICC 2021-IEEE International Conference on Communications. IEEE. 48. Aldalbahi, A., Siasi, N., Mazin, M.A. A. & Jasim. (2022). Digital compass for multi-user beam access in mmwave cellular networks. Digital Communications and Networks. 49. Zia, M.S., Blough, D.M., & Weitnauer, M.A. (2022). Uplink power control and snr-dependent beam alignment errors in mmwave cellular networks. IEEE 33rd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC). IEEE. 50. Lin, J., & Weitnauer, M. A. (2019). Pulse-level beam-switching for terahertz networks. Wireless Networks, 25(6), 3047–3062. 51. Deutschmann, B.J., Wilding, T., Larsson, E.G., & Witrisal, K. (2022). Location-based initial access for wireless power transfer with physically large arrays. In: 2022 IEEE International Conference on Communications Workshops (ICC Workshops), pp. 127–132. IEEE. 52. Zhang, T., Wang, Y., Ouyang, M., Xing, L., & Gao, F. (2024). Third-party camera aided beam alignment real-world prototype for mmwave communications. In: 2024 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1–6. IEEE. 53. Wu, D., Zeng, Y., Jin, S., & Zhang, R. (2023). Environment-aware hybrid beamforming by leveraging channel knowledge map. IEEE Transactions on Wireless Communications., 23(5), 4990–5005. 54. Zheng, K., Yang, H., Ying, Z., Wang, P., & Hanzo, L. (2023). Vision-assisted millimeter-wave beam management for nextgeneration wireless systems: Concepts, solutions, and open challenges. IEEE Vehicular Technology Magazine, 18(3), 58–68. 55. Arnold, M., Major, B., Massoli, F.V., Soriaga, J.B., & Behboodi, A. (2024). Vision-assisted digital twin creation for mmwave beam management. arXiv preprint arXiv:2401.17781. 56. Ma, W., Qi, C., & Li, G. Y. (2020). Machine learning for beam alignment in millimetre wave massive mimo. IEEE Wireless Communications Letters, 9(6), 875–878. 57. Chafaa, I., Belmega, E.V., & Debbah, M. (2020). Exploiting channel sparsity for beam alignment in mmwave systems via exponential learning. IEEE International Conference on Communications Workshops (ICC Workshops). IEEE. 58. Lin, C.H., Kao, W.C., Zhan, S.Q., & Lee, T.S. (2019). Bsnet: A deep learning-based beam selection method for mmwave communications. IEEE 90th Vehicular Technology Conference (VTC2019-Fall). IEEE. 59. Arvinte, M., Tavares, M., & Samardzija, D. (2019). Beam management in 5g nr using geolocation side information. 53rd Annual Conference on Information Sciences and Systems (CISS). IEEE. 60. Heng, Y., & Andrews, J. G. (2021). Machine learning-assisted beam alignment for mmwave systems. IEEE Transactions on Cognitive Communications and Networking, 7(4), 1142–1155. 61. Lazarev, V., & Fokin, G. (2020). Beamforming and spatial multiplexing performance evaluation in 5g ultra-dense networks. In: International Youth Conference on Electronics, Telecommunications and Information Technologies: Proceedings of the YETI 2020, St. Petersburg, Russia, pp. 709–717. Springer. 62. Wang, Z., Shan, W., Zhang, L., Ma, S., Xiao, M., Wei, N., & Li, S. (2022). Intelligent beam training with deep convolutional neural network in mmwave communications. GLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE. 63. Vankayala, S.K., Kumar, S., Thirumulanathan, D., Mathur, A., Yoon, I. S., & Kommineni. (2022). Deep-learning based beam selection technique for 6g millimeter wave communication. IEEE 33rd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC). IEEE. 64. Ktari, A., Ghauch, H., & Rekaya, G. (2022). Matrix factorization for blind beam alignment in massive mmwave mimo. IEEE Wireless Communications and Networking Conference (WCNC). IEEE. 65. Heng, Y., Mo, J., & Andrews, J. G. (2022). Learning site-specific probing beams for fast mmwave beam alignment. IEEE Transactions on Wireless Communications, 21(8), 5785–5800. 66. Heng, Y., & Andrews, J.G. (2022). Grid-less mmwave beam alignment through deep learning. GLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE. 67. TNarengerile, T.J., Patras, P., & Ratnarajah, T. (2022). Deep reinforcement learning-based beam training with energy and spectral efficiency maximisation for millimetre-wave channels. EURASIP Journal on Wireless Communications and Networking 2022(1), 110 68. Duan, Q., Kim, T., Ghauch, H., & Wong, E.W. (2020). Enhanced beam alignment for millimeter wave mimo systems: A kolmogorov model. GLOBECOM 2020-2020 IEEE Global Communications Conference. IEEE. 123 87 Page 52 of 53 S. Madhekwana et al. 69. Yang, P., Wu, W., Zhang, N., Shen, X., Yang, P., Wu, W...., & Shen, X. (2021). Machine learning-based beam alignment in mmwave networks. Millimeter-Wave Networks: Beamforming Design and Performance Analysis, 37–71. 70. Dreifuerst, R.M., & Heath, R.W. (2024). Hierarchical ml codebook design for extreme mimo beam management. IEEE Transactions on Machine Learning in Communications and Networking. 71. Krunz, M., Aykin, I., Sarkar, S., & Akgun, B. (2023). Online reinforcement learning for beam tracking and rate adaptation in millimeter-wave systems. IEEE Transactions on Mobile Computing, 23(2), 1830–1845. 72. Ozmat, U., Yazici, M. A., & Demirkol, M. F. (2023). Secure initial access and beam alignment using deep learning in 5g and beyond systems. IEEE Access., 12, 46–59. 73. Dinh-Van, S., Mo, Y.K., Zidan, J., Fowler, D.S., Evans, A., Higgins, M.D., et al. (2024). Rapid beam training at terahertz frequency with contextual multi-armed bandit learning. In: 2024 IEEE International Conference on Industrial Technology (ICIT), pp. 1–7. IEEE. 74. Xue, Q., Guo, J., Zhou, B., Xu, Y., Li, Z., & Ma, S. (2023). Ai/ml for beam management in 5g-advanced. arXiv preprint arXiv:2309.10575 75. Yang, J., Zhu, W., Tao, M., & Sun, S. (2023). Hierarchical beam alignment for millimeter-wave communication systems: A deep learning approach. IEEE Transactions on Wireless Communications., 23(4), 3541–3556. 76. Dantas, Y., Iturria-Rivera, P.E., Zhou, H., Bavand, M., Elsayed, M., Gaigalas, R., & Erol-Kantarci, M. (2023). Beam selection for energy-efficient mmwave network using advantage actor critic learning. In: ICC 2023-IEEE International Conference on Communications, pp. 5285–5290. IEEE. 77. Mohsenivatani, M., Ali, S., Rajatheva, N., & Latva-Aho, M. (2023). Deep contextual bandits learning-based beam selection for mmwave mimo systems. In: 2023 IEEE International Mediterranean Conference on Communications and Networking (MeditCom), pp. 211–216. IEEE. 78. Shah, S. H. A., & Rangan, S. (2023). Lstm-aided selective beam tracking in multi-cell scenario for mmwave wireless systems. IEEE Transactions on Wireless Communications, 23(2), 890–907. 79. Sheng, Y., Xu, J., & Tao, X. (2023). A two-stage prediction-based beam selection algorithm in mmwave massive mimo systems. In: 2023 IEEE 34th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), pp. 1–6. IEEE. 80. Hegde, D.N., Sarajlic, M., Sarkar, A., et al. (2023). Machine learning (ml)-assisted beam management in millimeter (mm) wave distributed multiple input multiple output (d-mimo) systems. arXiv preprint arXiv:2401.05422 81. Huang, H., Gui, G., Gacanin, H., Yuen, C., Sari, H., & Adachi, F. (2023). Deep regularized waveform learning for beam prediction with limited samples in non-cooperative mmwave systems. IEEE Transactions on Vehicular Technology, 72(7), 9614–9619. 82. Zheng, K., Yang, H., Ying, Z., Wang, P., & Hanzo, L. (2023). Vision-assisted millimeter-wave beam management for nextgeneration wireless systems: Concepts, solutions, and open challenges. IEEE Vehicular Technology Magazine, 18(3), 58–68. 83. Xiao, Z., Chen, S., & Zeng, Y. (2024). Simultaneous multi-beam sweeping for mmwave massive mimo integrated sensing and communication. IEEE Transactions on Vehicular Technology., 73(6), 8141–8152. 84. Chen, K., Qi, C., Dobre, O. A., & Li, G. Y. (2023). Simultaneous beam training and target sensing in isac systems with ris. IEEE Transactions on Wireless Communications., 23(4), 2696–2710. 85. Chen, W., Li, L., Chen, Z., Liu, Y., Ning, B., & Quek, T. (2024). Isac-enabled beam alignment for terahertz networks: Scheme design and coverage analysis. IEEE Transactions on Vehicular Technology. 86. Cui, Y., Nie, J., Cao, X., Yu, T., Zou, J., Mu, J., & Jing, X. (2024). Sensing-assisted high reliable communication: A transformerbased beamforming approach. IEEE Journal of Selected Topics in Signal Processing. 87. Jiang, H., Wang, Z., & Liu, Y. (2024). Sense-then-train: A novel beam training design for near-field mimo systems. arXiv preprint arXiv:2402.15634 88. Li, L., Chen, W., Chen, Z., Hu, T., Mei, W., & Ning, B. (2024). Enhancing terahertz communications coverage with isac-assisted beam management. IEEE Wireless Communications, 31(1), 34– 40. 89. Tariq, S., Arfeto, B.E., Khalid, U., Kim, S., Duong, T.Q., & Shin, H. (2024). Deep quantum-transformer networks for multi-modal beam prediction in isac systems. IEEE Internet of Things Journal. 90. Yanpeng, C., Qixun, Z., Zhiyong, F., Wen, Q., Ying, Z., Zhiqing, W., & Ping, Z. (2024). Sensing-assisted accurate and fast beam management for cellular-connected mmwave uav network. China Communications. 91. Chen, H., Liu, L., Xue, S., Sun, Y., & Pang, J. (2023). Active sensing for beam management: A deep-learning approach. In: 2023 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1–6. IEEE. 92. Jiang, T., Sohrabi, F., & Yu, W. (2023). Active sensing for twosided beam alignment and reflection design using ping-pong pilots. IEEE Journal on Selected Areas in Information Theory, 4, 24–39. 93. Yang, J., Wen, C.-K., Yang, X., Xu, J., Du, T., & Jin, S. (2023). Multi-domain cooperative slam: The enabler for integrated sensing and communications. IEEE Wireless Communications, 30(1), 40–49. 94. Li, Y. N. R., Gao, B., Zhang, X., & Huang, K. (2020). Beam management in millimeter-wave communications for 5g and beyond. IEEE Access., 8, 13282–13293. 95. Wang, X., Kong, L., Kong, F., Qiu, F., Xia, M., Arnon, S., & Chen, G. (2018). Millimeter wave communication: A comprehensive survey. IEEE Communications Surveys & Tutorials, 20(3), 1616– 1653. 96. Yi, W., Zhiqing, W., & Zhiyong, F. (2024). Beam training and tracking in mmwave communication: A survey. China Communications. 97. Khan, M. Q., Gaber, A., Schulz, P., & Fettweis, G. (2023). Machine learning for millimeter wave and terahertz beam management: A survey and open challenges. IEEE Access, 11, 11880–11902. 98. Khan, M. Q., Gaber, A., Schulz, P., & Fettweis, G. (2023). Machine learning for millimeter wave and terahertz beam management: A survey and open challenges. IEEE Access, 11, 11880–11902. 99. Häger, S., Sabbah, B., & Wietfeld, C. (2024). Design aspects for efficient beam training in 6g millimeter-wave networks with mobile users. cni.etit.tu-dortmund.de. 100. Ning, B., Wang, P., Li, L., Chen, Z., & Fang, J. (2022). Multiirs-aided multi-user mimo in mmwave/thz communications: A space-orthogonal scheme. IEEE Transactions on Communications, 70(12), 8138–8152. 101. Heng, Y., & Andrews, J.G. (2022). Grid-less mmwave beam alignment through deep learning. GLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE. 102. Klautau, A., González-Prelcic, N., & Heath, R. W. (2019). Lidar data for deep learning-based mmwave beam-selection. IEEE Wireless Communications Letters, 8(3), 909–912. 103. Jain, J.K., & Chauhan, D. (2023). Innovative model of internet of things for industrial applications. In: Opportunities and Chal123 Beam alignment for mmWave and THz: systematic... Page 53 of 53 87 lenges of Industrial IoT in 5G and 6G Networks, pp. 95–118. IGI Global, ???. 104. Myers, N. J., Mezghani, A., & Heath, R. W. (2018). Swift-link: A compressive beam alignment algorithm for practical mmwave radios. IEEE Transactions on Signal Processing, 67(4), 1104– 1119. 105. Candès, E. J. (2006). Compressive sampling. Proceedings of the International Congress of Mathematicians., 3, 1433–1452. 106. Cousik, T.S., Shah, V.K., Erpek, T., Sagduyu, Y.E., & Reed, J.H. (2022). Deep learning for fast and reliable initial access in aidriven 6g mm wave networks. IEEE Transactions on Network Science and Engineering. 107. Ma, K., Wang, Z., Tian, W., Chen, S., & Hanzo, L. (2021). Deep learning for beam-management: State-of-the-art, opportunities and challenges. arXiv preprint arXiv: 108. Sutton, R.S., & Barto, A.G. (2018). Reinforcement Learning: An Introduction. MIT press, ???. 109. Saravanan, R., & Sujatha, P. (2018). A state of art techniques on machine learning algorithms: a perspective of supervised learning approaches in data classification. In: 2018 Second International Conference on Intelligent Computing and Control Systems (ICICCS), pp. 945–949. IEEE. 110. LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436–444. 111. Chowdhury, M.Z., Shahjalal, S., Ahmed, M., & Jang, Y.M. (2020). 6g wireless communication systems: Applications, requirements, technologies, challenges, and research directions. IEEE Open Journal of the Communications Society. 1, 957–975 112. Huang, C., Hu, S., Alexandropoulos, G. C., Zappone, A., Yuen, C., Zhang, R., & Debbah, M. (2020). Holographic mimo surfaces for 6g wireless networks: Opportunities, challenges, and trends. IEEE Wireless Communications, 27(5), 118–125. 113. Khan, M. Q., Gaber, A., Schulz, P., & Fettweis, G. (2023). Machine learning for millimeter wave and terahertz beam management: A survey and open challenges. IEEE Access., 11, 11880–11902. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 123