A survey of privacy-preserving mechanisms on quality of experience in next-generation networks

Journal Pre-proof A Survey of Privacy-Preserving Mechanisms on Quality of Experience in Next-Generation Networks Rodrigo Dutra Garcia, Sankar Gowri Ramachandran, Christian Esteve Rothenberg, Bhaskar Krishnamachari, J´ o Ueyama PII: S1389-1286(25)00865-5 DOI: https://doi.org/10.1016/j.comnet.2025.111899 Reference: COMPNW 111899 To appear in: Computer Networks Received date: 20 May 2025 Revised date: 17 October 2025 Accepted date: 26 November 2025 Please cite this article as: Rodrigo Dutra Garcia, Sankar Gowri Ramachandran, Christian Esteve Rothenberg, Bhaskar Krishnamachari, J´ o Ueyama, A Survey of Privacy-Preserving Mechanisms on Quality of Experience in Next-Generation Networks, Computer Networks (2025), doi: https://doi.org/10.1016/j.comnet.2025.111899 This is a PDF of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability. This version will undergo additional copyediting, typesetting and review before it is published in its final form. As such, this version is no longer the Accepted Manuscript, but it is not yet the definitive Version of Record; we are providing this early version to give early visibility of the article. Please note that Elsevier’s sharing policy for the Published Journal Article applies to this version, see: https://www.elsevier.com/about/ policies-and-standards/sharing#4-published-journal-article. Please also note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ©2025 Published by Elsevier B.V. A Survey of Privacy-Preserving Mechanisms on Quality of Experience in Next-Generation Networks Rodrigo Dutra Garciaa,∗,1,Gowri Sankar Ramachandranc,Christian Esteve Rothenbergb, Bhaskar Krishnamacharidand Jó Ueyamaa aInstitute of Mathematics and Computer Science, University of São Paulo, São Carlos, 13566-590, SP, Brazil bSchool of Electrical and Computer Engineering, Universidade Estadual de Campinas, Campinas, 13083-970, SP, Brazil cQueensland University of Technology, Faculty of Science, Brisbane, QLD 4000, Queensland, Australia dUSC Viterbi School of Engineering, University of Southern California, Los Angeles, CA 90089, California, USA ARTICLE INFO Keywords: Quality of Experience Privacy B5G 6G Network Technologies ABSTRACT Quality of Experience (QoE) and privacy protection are central concerns in modern network technologies, especially with the rise of user-centric applications in Beyond 5G (B5G) and emerging 6G networks. However, as the number of connected devices and the volume of exchanged data increase, protecting user privacy becomes more challenging, especially as stronger confidentiality measures restrict access to the data needed to improve QoE. This creates a persistent trade-off, as telecommunications providers (Telecoms) must balance the need to analyze user behavior for service quality improvements while protecting sensitive data. This survey analyzes the trade-off between privacy and QoE, focusing on how privacy-preserving mechanisms are adapted to the constraints of next-generation communication networks. It highlights network-level impacts such as latency, overhead, scalability in mobile environments, and operational challenges for Mobile Network Operators (MNOs) and Service Providers (SPs). We review prominent techniques such as Federated Learning (FL), Encrypted Traffic Inference, Blockchain, and Differential Privacy (DP), clarifying their distinct roles, strengths, and limitations in managing the privacy-QoE trade-off. It highlights challenges related to network-level impacts such as latency, overhead, and scalability in mobile environments, as well as operational challenges for MNOs and SPs. Furthermore, a synthesis of research issues and potential future research directions is presented. 1. Introduction 1.1. Background In the era of advanced network technologies, QoE evaluates the overall acceptability of a service or application from the user’s perspective (Kougioumtzidis et al.,2022). Unlike Quality of Service (QoS), which focuses on technical network-centric metrics (e.g., latency, throughput), QoE considers the user’s satisfaction by incorporating both objective performance indicators and subjective user-centric perceptions influenced by human factors, including perceptual, cognitive, and behavioral aspects of service interaction (Laiche et al.,2020;Bouraqia et al.,2020). Subjective methods rely on direct user input, such as surveys, questionnaires, or scoring frameworks like the Mean Opinion Score (MOS), where participants numerically rate their perceived quality or satisfaction with a service or application (Chen et al.,2024). These methods capture personal and psychological aspects of the user experience (e.g., facial expressions, heart rate) for understanding perceptual quality (Bingöl et al.,2022;Amour et al.,2018). Examples of QoE assessment include using MOS ratings in Virtual Reality (VR) under variable network conditions (Rossi et al.,2024;Tondwalkar et al.,2024). Further examples involve evaluating QoE for multimedia streaming and video services (Chen et al.,2024;Barakabitze et al.,2020). The increasing prevalence of digital services like VR and immersive holographic experiences in 6G makes the user experience a key consideration (Tondwalkar et al.,2024; Dang et al.,2020), which means future 6G networks must prioritize QoE over traditional QoS metrics. Researchers focus on developing solutions to improve QoE by employing the end-user’s perception (Tondwalkar et al.,2024;Chen ∗Corresponding author [email protected] (R.D. Garcia); [email protected] (G.S. Ramachandran); [email protected] (C.E. Rothenberg); [email protected] (B. Krishnamachari); [email protected] (J. Ueyama) ORCID(s): 1Rodrigo Dutra Garcia is supported by FAPESP Grant No. 2023/15919-1. Garcia et al.: Preprint submitted to Elsevier Page 1 of 35 Privacy-Preserving QoE in Next-Gen Networks Quality of Service (network related) User Perspective Perceived Quality User Experience/Satisfaction Psychological Factors Latency Bandwidth/Throughput Packet Loss Frame Rate Quality of Experience User Application Influence Factors Figure 1: QoE Overview in VR. This illustration conceptually differentiates between QoS, focusing on network-centric metrics such as latency, bandwidth, packet loss, and frame rate, and QoE, which contains the user perspective. QoE metrics include perceived quality, user experience/satisfaction, and psychological factors, representing the user’s subjective evaluation of the service. et al.,2024;Laghari et al.,2023). Machine learning models are being developed to predict user-centric QoE metrics and tailor services to individual needs (Ahmad et al.,2021;Kougioumtzidis et al.,2023,2024). Figure 1illustrates a conceptual overview to clarify the user-centric nature of QoE using VR as an example. 1.2. Motivation Since QoE contains multiple dimensions, telecommunications providers collect a broad range of network and user data to evaluate and improve service quality (Yeznabad et al.,2024;Dobreff et al.,2023). A substantial portion of this data can be linked to individual users and devices. It includes GPS-based location data (Feng et al.,2024a;Esper et al., 2023), user preferences, device context, and behavioral indicators (Zhang et al.,2022;Setayesh and Wong,2025). Video streaming services, for instance, regularly collect usage patterns and network metadata (Setayesh and Wong, 2025;Artioli,2024). Although this information helps enhance service performance, it also enables detailed analysis of user behavior without clear user understanding or control, raising concerns about privacy and data governance. These concerns are amplified in environments with increasing numbers of connected devices and high volumes of data exchange, where protecting user information becomes more complex (Mao et al.,2023;Dang et al.,2020). In such contexts, privacy protection in QoE often involves trade-offs between personalization and data safeguards (Feng et al., 2024a). Implementing privacy protection mechanisms creates a challenge: stronger privacy measures often reduce service quality, as QoE improvements typically require access to user data (Jin et al.,2024). Additionally, users’ perception of privacy significantly influences QoE, as trust and privacy concerns directly affect QoE (Zhou et al.,2019). This creates a conflict between privacy protection and understanding user requirements, as service providers (SPs) must analyze user behavior and preferences to improve service quality while preserving data confidentiality. This privacy-QoE trade-off particularly impacts areas where finding the balance between privacy protection and QoE data collection remains an active area of research. For instance, video streaming services require real-time quality monitoring while protecting user viewing patterns (Wassermann et al.,2020c). Mobile networks face challenges in maintaining service quality under encryption (Wassermann et al.,2020a), while Internet of Things (IoT) applications balance device constraints with privacy requirements (Le et al.,2023) and healthcare applications demand strict privacy compliance while delivering consistent QoE (Lin,2022). Decentralized approaches such as Federated Learning (FL) are expected to be a leading approach for sixthgeneration networks (6G) (Zuo et al.,2023;Kazmi et al.,2024;Javeed et al.,2024). However, some other studies explored the security and privacy concerns of FL, particularly focusing on data poisoning and inference attacks (Rao et al.,2025;Ben Saad et al.,2023;Sirohi et al.,2023). These attacks can compromise the integrity of models and expose sensitive user data. Moreover, the large volume of data generated by 6G services, combined with the demand for real-time processing and analysis, creates significant challenges for maintaining user privacy and ensuring data security (Mao et al.,2023;Rachakonda et al.,2024;Sandeepa et al.,2024). As networks become more complex and user expectations increase, telecoms and SPs need to understand and develop methods that optimize QoE without Garcia et al.: Preprint submitted to Elsevier Page 2 of 35 Privacy-Preserving QoE in Next-Gen Networks compromising the privacy and security of user data. This introduces new challenges in managing the user experience and protecting personal information. Existing surveys have explored QoE across a wide range of application domains, including cloud-based interactive services (Arellano-Uson et al.,2024), IoT systems (Sanyour et al.,2022), and various forms of video streaming (Choon and Lim,2024;Ghafil and Ali,2021;Chiariotti,2021;Chen et al.,2024), such as 360-degree content and adaptive formats. Additional work has addressed telemeetings and videoconferencing (Skowronek et al.,2022), immersive technologies like VR and XR (Vlahovic et al.,2022), and application-specific contexts such as mobile health (Nasralla et al.,2023). These efforts consistently highlight the complexity of QoE and the necessity of multi-layered models that combine subjective evaluations with objective system performance indicators. However, while these studies have advanced the understanding of QoE measurement and optimization, they have mainly focused on performance metrics and user experience factors without examining the privacy aspect when improving user experience. To address this gap, the present survey examines QoE from a privacy perspective. It provides a structured analysis of current methodologies, identifies key challenges, and outlines prospective directions for research at the intersection of user experience and data privacy. This survey aims to analyze privacy-preserving techniques in QoE management and their associated challenges across different application domains, identify applications that involve user-sensitive data, and examine the predominant data types, evaluation methods, and open-source tools used in this context. This article has two main purposes. First, we present a survey that establishes how privacy-preserving techniques have been applied in QoE measurement and optimization. Second, we present a detailed analysis examining the specific advantages and limitations of each method, identify technical challenges, and propose future research directions. We examine how the architectural design of next-generation communication networks determines which mechanisms are chosen, how they are deployed, and what trade-offs they involve. The analysis distinguishes the roles, capabilities, and data access limitations of SPs, mobile network operators (MNOs), and edge providers. 1.3. Contributions Our contributions are the following: 1. A survey of privacy-preserving techniques for QoE in network contexts, including a structured classification of current approaches and their applications; 2. A comparative analysis of privacy-preserving approaches, clarifying their distinct roles and trade-offs in QoE contexts. This analysis includes: •A review of key methods such as FL, Encrypted Traffic Inference, Blockchain, and DP regarding their ability to balance privacy protection with QoE. •Identification of key technical and practical challenges, particularly concerning scalability, real-time processing, and the inherent privacy-QoE trade-off in emerging network technologies. •A structured threat analysis that maps key privacy threats in QoE monitoring to specific privacy-preserving techniques and clarifies the associated privacy-utility trade-offs. 3. A synthesis of key research issues and outlines future directions, including privacy-preserving computation, enhancing FL for real-world deployment, and the development of user-centric evaluation methods. 1.4. Organization This article is organized as follows. Section 2provides an overview of QoE and privacy in network technologies. Section 3reviews the existing literature and positions this survey within the broader context of privacy research. Section 4describes the survey methodology. Section 5presents the results, categorizing the selected studies by the privacy-preserving techniques used, the application domains addressed, and the types of research contributions. It then synthesizes the findings through a structured threat analysis and a review of the key technical and operational challenges identified in the literature. Section 6presents the discussion of the findings. Section 7outlines open issues and future research directions, and Section 8concludes the survey. Table 1shows the acronyms used in this survey. Garcia et al.: Preprint submitted to Elsevier Page 3 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 1 List of Acronyms. Acronym Definition Acronym Definition AES Advanced Encryption Standard QoE Quality of Experience AR Augmented Reality QoS Quality of Service B5G Beyond 5G QUIC Quick UDP Internet Connections CDN Content Delivery Network RAN Radio Access Network DP Differential Privacy SGX Software Guard Extensions DPI Deep Packet Inspection SNR Signal-to-Noise Ratio DWT Discrete Wavelet Transform SP Service Provider EEG Electroencephalography STFT Short-Time Fourier Transform EDA Electrodermal Activity TEE Trusted Execution Environment ECG Electrocardiogram TLS Transport Layer Security FL Federated Learning UAV Unmanned Aerial Vehicle FoV Field of View VR Virtual Reality GDPR General Data Protection Regulation XR eXtended Reality IoV Internet of Vehicles ZKP Zero-Knowledge Proof ISP Internet Service Provider KPI Key Performance Indicator LSH Locality-Sensitive Hashing MEC Mobile Edge Computing MNO Mobile Network Operator MOS Mean Opinion Score 2. Quality of Experience and Its Relationship to Privacy 2.1. Quality of Experience QoE is the user’s perceived satisfaction with an application or service in their specific context of use. Unlike QoS, which focuses on technical metrics, QoE includes a user-centric approach emphasizing the end user’s perception of service quality. Machine learning models to predict QoE have gained popularity, enabling adaptive, user-centric services that meet the evolving needs of modern telecommunications and multimedia systems (Kougioumtzidis et al., 2022). QoE metrics contain not only network QoS parameters like packet loss, latency, jitter, signal quality, round-trip time, retransmission rate, and throughput, but also user-specific metrics, including physiological indicators, such as heart rate variability and respiration rate, to capture user emotional and perceptual factors. QoE can be measured using subjective or objective methods. Objective methods typically consider technical performance indicators that can be measured quantitatively. In contrast, subjective methods rely on user surveys, focus groups, and direct user feedback to gauge satisfaction (Chiariotti,2021). Non-QoS-based metrics, focusing on user-centric factors like emotions, contextual influences, and physiological responses, are increasingly recognized for QoE assessment. Measures such as heart rate, EEG, and EDA correlate strongly with user emotions, cognitive load, and immersion levels, particularly in AR/VR environments (Keighrey et al.,2021;Salgado et al.,2018). Subjective scores of vibrotactile experience in VR have been used as ground truth for QoE evaluation in tactile VR (Zhang et al.,2024). Frontal lobe power asymmetry (FLPA), derived from EEG, has been investigated as a predictive physiological indicator of QoE in tactile VR (Zhang et al.,2024). Subjective QoE assessments in multimedia communications have explored perceived overall quality and audio quality (Vijayakumar et al.,2024). Physiological signals such as ECG and respiration have been used to predict these subjective quality perceptions in multimedia (Vijayakumar et al.,2024). Factors related to user experience quality in VR gaming, compared to non-VR, have been analyzed, including physiological responses (Parvez et al.,2023). Physiological parameters like EDA, heart rate, blood volume pulse, temperature, and blood pressure capture physiological responses to VR environments (Parvez et al.,2023). Qualitative surveys evaluating calmness and energy level also provide an understanding of user perception in VR experiences (Parvez et al.,2023). Video streaming and immersive environments are evaluated using a range of metrics capturing different aspects of the user’s perception and system performance. Visual fidelity is a primary concern, often assessed through resolution and perceptual quality scores. Studies frequently investigate the impact of varying video resolutions on QoE (Tran et al.,2024;Amirpour et al.,2024) and explore methods to enhance perceived visual quality (Llorente et al.,2024). Garcia et al.: Preprint submitted to Elsevier Page 4 of 35 Privacy-Preserving QoE in Next-Gen Networks Signal-to-noise ratio (SNR), often perceptual SNR, is also used as an objective metric reflecting the quality of the video signal itself (Wang et al.,2024b). Spatial aspects are essential in immersive media, with Field of View (FoV) being a key metric, particularly in 360°video streaming, where accurate FoV prediction is essential for QoE optimization (Wang et al.,2024c;Hou et al.,2024). Temporal aspects like frame rate and smoothness are also crucial, with research examining how frame rate variations affect gaming QoE (Liu et al.,2023) and developing systems to maintain high frame rates for enhanced experiences (Kokiadis et al.,2024). In immersive scenarios like 360°video and VR, immersion level and presence, alongside comfort and the user’s sense of presence, become important QoE indicators (Elwardy et al.,2024;Althoff et al.,2023). Furthermore, system-centric metrics like latency, including end-to-end latency, startup delay, and rebuffering time, directly impact user frustration and are frequently used to quantify QoE degradation (Farahani et al.,2024;Zeng et al.,2024). Considering the interactive nature of many applications, metrics like interaction responsiveness and usability are also relevant in evaluating the overall QoE (Ouardi et al.,2024). Emerging research also considers cognitive load as a QoE factor, recognizing that excessive cognitive demands can negatively impact the user experience, especially in demanding applications like VR and AR (Elwardy et al.,2024). To ensure conceptual clarity, we define the terminology for three aspects of QoE that appear frequently in the works analyzed in this survey: modeling, forecasting, and optimization. These definitions are the following: •QoE modeling: refers to the process of creating a formal representation, such as a mathematical function or machine learning model, that links objective technical metrics (like QoS parameters) and contextual factors to a subjective measure of user-perceived quality (such as the MOS). The aim is to identify and model the factors that influence user satisfaction (Skaka - Čekić and Baraković Husić,2023;Corcuera Bárcena et al.,2023b). •QoE forecasting (or QoE prediction): utilizes an existing QoE model to estimate future QoE values or quality degradation events based on current and historical data. This serves as a proactive measure to detect potential issues before they affect the user (Al-Quraan et al.,2024;Ghafil and Ali,2021;Corcuera Bárcena et al.,2023a; Nguyen et al.,2023). •QoE optimization: is the active process of adjusting system or network parameters (such as resource allocation, bitrate selection, or routing) in real-time to improve or maintain a target level of QoE. This is the practical application of results derived from modeling and forecasting (Li et al.,2022;Darwich et al.,2024;Liu et al., 2020a). 2.2. Privacy and its intersection with QoE Protecting user data and personal information from unauthorized access, use, or disclosure is a crucial aspect of ensuring both user trust and regulatory compliance in modern digital service delivery (Mao et al.,2023). In addition to network metrics (i.e., QoS), QoE influence factors can contain sensitive user data, including service preferences, localizations, user emotions, contextual influences, physiological responses, device information, and usage patterns (Taoetal.,2019;Bingöl etal.,2024;KatsigiannisandRamzan,2018;Boffettietal.,2025).Users’ growingawareness of network monitoring has made them more cautious about their online behavior, resulting in less reliable and insufficient data for accurate QoE assessment (Zhou et al.,2019). User privacy intersects with QoE in complex ways, often involving trade-offs between personalization and data protection. The relationship between QoE and privacy in network technologies involves several considerations: 1. Data collection for QoE improvement may compromise user privacy. Detailed user data can enhance service personalization and performance optimization, potentially improving QoE. However, as demonstrated by Zhou et al. (2019), this data collection raises privacy concerns, with users becoming increasingly cautious about their online behavior when aware of network monitoring. Studies in emerging network technologies further confirm how collecting user data, while beneficial for service quality, introduces substantial privacy risks (Corcuera Bárcena et al.,2023a). 2. Strong privacy measures might limit the ability to personalize services, potentially impacting QoE. While robust privacy protections protect user data, they may restrict the granularity of data available for QoE enhancements. This trade-off is evident in edge computing environments, where Esper et al. (2023) shows how privacy measures can affect service performance. Similarly, in immersive video streaming, Jin et al. (2024) demonstrates how privacy preservation can lead to decreased accuracy in viewpoint prediction. Garcia et al.: Preprint submitted to Elsevier Page 5 of 35 Privacy-Preserving QoE in Next-Gen Networks 3. Users’ perception of privacy can itself be a factor in their overall QoE (Zhou et al.,2019). Research by Gao et al. (2020) shows that user privacy concerns necessitate privacy-preserving data exchange methods to enable personalized QoE improvement in video streaming services. Additionally, Yan et al. (2019) demonstrates that privacy issues arise because platforms are reluctant to share user data, which complicates the implementation of distributed recommendations. 4. Privacy-preserving techniques may introduce performance overheads that could affect QoE. For example, encryption and anonymization processes might introduce latency or computational costs. This is particularly evident in blockchain-based privacy solutions, where He et al. (2022) shows how privacy mechanisms can introduce latency in network slice orchestration. Similarly, Le et al. (2023) demonstrate how privacy-preserving mechanisms in edge computing can substantially impact computational efficiency and service response times. Managing QoE presents multiple challenges (Panahi et al.,2024), one of which is the difficulty of collecting largescale network and user data under varied conditions while maintaining high data quality and accuracy. Privacy concerns add another layer of complexity to QoE research and management, as protecting user data and ensuring its security is essential (Zhou et al.,2019). Collecting personally identifiable user data related to QoE is particularly sensitive, necessitating privacy and security measures. SPs and researchers face a complex balancing act: they must collect sufficient data to improve QoE while simultaneously adhering to stringent data protection laws and regulations. These legal requirements often restrict data collection, storage, and usage practices. 2.3. The Dual Aspects of Privacy in QoE In this survey, privacy and QoE interact through two interconnected dimensions. The first dimension addresses how privacy methods affect system performance. The second dimension reflects users’ perceptions of privacy and their level of trust. These dimensions differ as follows: •The system-level impact of privacy mechanisms: This aspect concerns the objective, technical effects that privacy-preserving mechanisms have on network and application performance, which in turn indirectly affect QoE. Implementing technologies like FL, DP, or strong encryption is not without cost. As discussed in Section 5, these methods can introduce computational overhead, increase latency, and, in some cases, reduce the utility or accuracy of data-driven models (e.g., personalized recommendations or QoE prediction). This creates a direct privacy-utility trade-off, where stronger privacy guarantees may lead to a degradation in the very service metrics that contribute to a high QoE (Jin et al.,2024). As presented in Section 5, much of the research in this topic focuses on mitigating this trade-off by designing more efficient privacy-preserving algorithms. •The perceptual impact of privacy: This aspect addresses the subjective and psychological dimension of privacy. In this case, QoE is directly influenced by the user’s perception, trust, and feeling of control over their personal data, independent of specific system performance metrics (Baraković and Husić,2022). A user who feels their privacy is being invaded or who distrusts a SP may report a lower QoE, even if the technical performance (e.g., video resolution, latency) is optimal (Zhou et al.,2019). 3. Previous Survey Analysis and Motivation This section reviews existing surveys to position the context for this study. It first examines research on QoE across different application domains, then turns to privacy-related studies in next-generation networks. The review highlights a gap where QoE and privacy intersect, which this survey aims to explore. 3.1. QoE Across Different Application Domains QoE studies have addressed different application areas, including cloud services, interactive multimedia, IoT applications, mobile health systems, and distributed computing environments. For instance, Arellano-Uson et al. (2024) reviewed 28 studies on QoE in cloud-based interactive applications, including remote desktop services, cloud gaming, and web apps. They proposed a five-category taxonomy of measurement methods and mapped them to a threestage evaluation model: input collection, processing, and output generation. The study identified 21 input metrics and examined processing approaches from heuristics to machine learning. Outputs included video, audio, temporal metrics, and QoE indices. Key challenges included weak QoS–QoE correlations, lack of protocol-independent APIs, and limited validation. The authors recommended real-time, general-purpose monitoring tools for growing post-pandemic demand. Garcia et al.: Preprint submitted to Elsevier Page 6 of 35 Privacy-Preserving QoE in Next-Gen Networks Similarly, Laghari et al. (2023) provided a broader review of QoE research across cloud computing, examining work on general frameworks, cloud gaming, and social cloud environments up to mid-2023. They distinguished between objective and subjective assessment methods, including MOS-based ITU standards, and evaluated frameworks such as CLAMS, Follow-Me Cloud, Cloud2Bubble, and mobile cloud gaming platforms regarding monitoring coverage, analytic capabilities, and adaptability to policy changes. The review identified persistent limitations, including insufficient user-level data in SLA monitoring, scalability challenges in edge and fog deployments, and limited use of QoE-aware resource allocation strategies. The authors concluded by outlining research directions that included the integration of IoT and big data streams, the application of machine learning to service provisioning, and the design of mechanisms that connect SLA violations with measurable QoE degradation. They argued that QoE should be treated as a central consideration in service design rather than a derivative of QoS. Sanyour et al. (2022) conducted a four-stage systematic review that distilled 43 relevant studies. They showed that traditional user-centric QoE indicators, such as the MOS, were unsuitable for IoT environments characterized by machine-to-machine communication. By introducing the concepts of QoIoT and QoME and categorizing metrics into QoS-based, QoD-based, and hybrid classes, the authors argued that credible QoE evaluation in IoT must rely on objective measurements of network performance, data quality, and contextual parameters across all layers of the IoT architecture. Skowronek et al. (2022) provided a holistic survey of QoE in telemeetings and videoconferencing, integrating a taxonomy of over 120 human, system, and context factors with analysis of conversational-cognitive processes and current assessment methods; their study shows that QoE depends on both technical variables (latency, audiovisual fidelity, network topology) and non-technical dimensions such as cognitive load, social presence, and meeting purpose, and introduces the Telemeeting Profile Template to guide future evaluations. This holistic perspective on QoE assessment is particularly relevant in the context of immersive environments such as XR, which shares similar QoE challenges to those outlined by Arellano-Uson et al. (2024). Kougioumtzidis et al. (2022) found that ML-driven QoE prediction must evolve beyond video-streaming heuristics to cope with XR media, where factors such as presence, immersion and cybersickness interact with bitrate, latency and rendering load; by synthesising state-of-the-art models for VR/AR/MR and cloud games, the authors demonstrate how deep-learning architectures can fuse perceptual and physiological signals to enable real-time, adaptive QoE control in immersive services. Chen et al. (2024) conducted a broad review of QoE research on emerging immersive video formats, including stereoscopic 3D, 360°, holographic, and multisensory streams. The authors argue that future evaluations should go beyond visual quality metrics to include multidimensional measures that reflect immersion, interaction, and connectivity. Likewise, Chiariotti (2021) presented an end-to-end analysis of 360°video delivery, connecting projectionaware encoding, viewport-adaptive streaming, and perceptual modeling. The study also highlights open challenges, including the development of metrics sensitive to cybersickness and the application of reinforcement learning to balancerate, distortion, and user experience.Ghafil and Ali(2021) reviewedQoEresearch for adaptivevideo-streaming services, showing that network impairments such as increased join time and rebuffering events degrade user perception non-linearly; by combining subjective evidence (MOS, crowd tests) with objective signaland traffic-based metrics, the authors demonstrate that holistic QoE evaluation and forecasting require integrated models capable of capturing both user-level feedback and measurable network-layer dynamics. Both the video-streaming QoE survey and the IoT QoE review converge on the need for multi-dimensional assessment frameworks that integrate subjective user feedback with objective technical indicators, such as throughput, latency, and packet loss, to obtain a truly comprehensive view of QoE. Bouraqia et al. (2020) provided a comprehensive survey of QoE in streaming services over 4G, 5G, and B5G networks. It reviews subjective evaluation methods based on lab tests and crowdsourcing, objective metrics from signal and traffic data, and emerging hybrid models using machine learning. The authors argue that only these combined approaches can address the complexity of mobile wireless environments. Furthermore, Skaka - Čekić and Baraković Husić (2023) offers a comprehensive review of feature selection methods applied to video QoE modeling. This study investigates how these methods address the challenges of high dimensionality in complex QoE spaces to enhance prediction accuracy and categorizes feature selection techniques based on data collection, modeling steps, and QoE characteristics. Nasralla et al. (2023) conducted a detailed survey connecting future 6G technologies, such as reconfigurable intelligent surfaces, terahertz small cells, satellite aerial radio, mURLLC, and blockchain-based security, to QoE requirements in mobile health multimedia.. The study shows that meeting the strict latency and bandwidth demands of applications like telesurgery, holographic consultations, and AR/VR training depends on these technologies functioning together. It also outlines research challenges that must be addressed to deliver clinically acceptable user Garcia et al.: Preprint submitted to Elsevier Page 7 of 35 Privacy-Preserving QoE in Next-Gen Networks experiences using 6G capabilities. Wu et al. (2022) reviewed resource allocation objectives for 6G and proposed an exponential demand side QoE criterion that accounts for the high heterogeneity of future networks, including diverse radio layers, devices, and application classes. Using a social vehicular case study, the authors showed that this criterion distributes resources more fairly while maintaining satisfactory QoE for users. Moreover, da Silveira et al. (2024) systematically reviewed the transformative role of physiological data in assessing QoE and User Experience (UX), driven by cost-effective sensing devices, advanced analytical methods, and emerging immersive domains like VR/AR and mulsemedia (multi-sensory media), highlighting its growing utility as a complementary metric to traditional subjective-objective evaluations in interactive applications. 3.2. Privacy-Oriented Studies for the Next Generation Networks The survey by Khan et al. (2025) examined security and privacy threats in UAV-assisted networks within the B5G/6G context, focusing on physical-layer security (PLS), blockchain, FL, and post-quantum cryptography (PQC). It emphasized the need for layered defenses across UAV hardware, software, communication, and sensing components to address threats such as swarm attacks, signal interception, and regulatory noncompliance. The analysis highlighted PLS methods like secure beamforming and reconfigurable surfaces to counter eavesdropping, while noting scalability issues in blockchain-based authentication and data integrity. FL was presented as a privacy-preserving alternative for distributed learning, though the survey identified a lack of real-world validation and standardized frameworks. Although privacy was not explicitly framed through QoE, the study acknowledged trade-offs between security and operational demands in latency-sensitive applications. It positioned UAV security as a multidisciplinary challenge spanning cryptography, AI, and regulatory policy. Similarly, Benarous et al. (2025) analyzed location privacy in vehicular networks, focusing on pseudonym change strategies (PCSs) as defenses against tracking through beacon messages. PCSs were classified as noncooperative (vehicle-based), cooperative (involving peers or infrastructure), or hybrid (combining obfuscation, silence, and encryption). These methods aimed to disrupt pseudonym linkability and prevent trajectory reconstruction based on identifiers like MAC/IP addresses or contextual data. The survey addressed adversary models such as global passive attackers and cross-layer tracking, emphasizing the limitations of idealized simulations. It used metrics like entropy and anonymity set size (ASS) but acknowledged their inability to capture real-world tracking risks. The study also discussed conditional privacy, balancing anonymity with traceability, and critiqued static mix-zones and excessive pseudonym use for their potential impact on performance and safety. Mao et al. (2023) explored security and privacy challenges in 6G edge networks, emphasizing edge computing, caching,and intelligence.Ithighlightsdecentralized solutions likeFLand blockchain to address threats while balancing resource constraints, latency, and energy efficiency. Key issues include securing dynamic, distributed architectures and mitigating risks in IoT and edge-AI applications. However, while the survey addresses QoS trade-offs (e.g., latency, reliability), it does not deeply investigate user-centric QoE metrics, such as perceived service quality or user satisfaction, resulting in a lack of comparative understanding of how different privacy mechanisms are employed as integral components of systems and frameworks designed to improve user QoE. 3.3. Scope of this Survey Previous privacy surveys lacked analysis of mechanisms to improve QoE while preserving privacy. Additionally, they did not consider the trade-off between privacy protection and data utility, with some schemes degrading user QoE. This highlights the need for further research into integrated approaches that can balance or simultaneously improve both QoE and privacy. Table 2compares the focus of prior surveys with this work. Despite these surveys covering QoE in various services and new technologies, none specifically examines QoE from a privacy perspective, an important topic given growing privacy concerns in modern networks and the influence of stringent data protection regulations, which make privacy a necessary dimension of QoE analysis. Our work focuses on privacy-preserving QoE techniques. We examine key application domains and analyze technologies such as FL, Blockchain, DP, and Encrypted Traffic Inference, a technique that estimates QoE from network traffic patterns without decrypting the payload. The analysis highlights the distinct roles, capabilities, and trade-offs associated with each approach. We also provide an overview of the open-source tools and datasets used in existing studies and discuss the main challenges and future research directions. Garcia et al.: Preprint submitted to Elsevier Page 8 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 7 Privacy-preserving techniques across application domains (Part II). Application Domain Privacy Technique Works VR Camouflaged Tile Requests Wei and Yang (2022) End-to-End Encryption Islam et al. (2023) Not Specified Wei et al. (2023) Data Market Blockchain Bordel Sánchez et al. (2024), Jiang and Wu (2022) Telemedicine Systems ID-Based Cryptography Lin (2022) Access Control De Deus et al. (2021) Video Caching DP Zhang et al. (2022) Blockchain Li and Wan (2021) IoV LSH Xu et al. (2021) Anomaly Detection FL Guo et al. (2023) Cloud Computing FL Feng et al. (2024b) Encrypted Traffic Classification Encrypted Traffic Inference Dillbary et al. (2024) HTTP Adaptive Streaming Encrypted Traffic Inference Wassermann et al. (2020c) Healthcare Hybrid Encryption Scheme Jayaram and Prabakaran (2021) Location-Based Services DP Feng et al. (2024a) Network Slice Blockchain He et al. (2022) Proactive Caching FL Liu et al. (2020a) Traffic Estimation FL Valente et al. (2023) Ultra-Dense IoT Networks Elliptic Curve Cryptography Nyangaresi (2022) Vehicle-to-Everything FL Renda et al. (2022) Vehicular Crowdsensing Blockchain Zhao et al. (2022) Wireless Body Area Networks Blockchain Li and Zhang (2024) Wireless Networks FL Liang et al. (2022) XR Streaming Noisy Entropy Function Wei et al. (2024) User User devices devices ISP/Network Monitoring Point Encrypted Traffic Flow Encrypted Traffic Analysis for QoE Inference Packet-level features (size, inter-arrival times) Flow-level features (duration, volume, burst patterns) Advanced features (e.g., Spectral: STFT, DWT) 1. Feature Extraction (from encrypted patterns) 2. Analysis Engine Machine Learning Models Statistical Models Network Probe Web Browsing Video Streaming Service 3. Inferred QoE Metrics e.g., stalling, resolution, loading time, speed index Figure 5: Encrypted Traffic Inference. Encrypted user traffic is analyzed by extracting features from its patterns (not content) and using machine learning to estimate QoE metrics. •Model accuracy in QoE inference is consistently analyzed across studies (Gutterman et al.,2020; Wassermann et al.,2020c). •Real-time processing capabilities allow for timely QoE monitoring and decision-making (Wassermann et al.,2020c;Wehner et al.,2021). 3. Implementation Contexts: Garcia et al.: Preprint submitted to Elsevier Page 15 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 8 Machine-learning models and feature categories used for QoE inference from encrypted traffic. Works Models for Encrypted Traffic Inference Data / Feature Types Employed Corcuera Bárcena et al. (2023a); Orsolic and Skorin-Kapov (2020); Wassermann et al. (2020c) Decision Trees, Random Forests, Regression Trees, Hoeffding Trees (with fuzzy extensions) Packet-level statistics (size, inter-arrival time); flow and session metrics (throughput, counts, means, deviations); window-based aggregates that capture shortand long-term behaviour Corcuera Bárcena et al. (2023a) Bayesian Networks Packet and flow statistics; QoS indicators (round-trip time, packet loss); contextual variables such as UE position and cell load Corcuera Bárcena et al. (2023a) Support Vector Machines Packet and flow statistics; spectral transforms of traffic time-series (STFT, DWT) Dillbary et al. (2024); Islam et al. (2023)Deep, Convolutional, and Graph Neural Networks; AutoML pipelines combining shallow and deep learning Heterogeneous features: packet, flow, window, and chunk parameters; QoS header fields; spectral coefficients; contextual information Corcuera Bárcena et al. (2023a); Corcuera Bárcena et al. (2023b)Takagi–Sugeno–Kang (TSK) fuzzy models Packet and flow statistics; chunk-related parameters; QoS metrics •Streaming services primarily video platforms such as YouTube, but also other content types serve as the main testbed for encrypted QoE analysis (Gutterman et al.,2020;Wassermann et al.,2020c;Orsolic and Skorin-Kapov,2020). •Web browsing across device types (desktops, smartphones, tablets) and mobile applications are also studied to assess QoE inference in dynamic environments (Wassermann et al.,2020a;Wehner et al.,2021;Casas et al.,2021). 5.1.3. Blockchain Blockchain does not directly improve QoE objective metrics (e.g., latency, throughput); instead, by enforcing transparent and verifiable operations across administrative domains, it strengthens system integrity and indirectly supports QoE. This is especially relevant in settings like federated network slicing or vehicular crowdsensing, where different administrative entities must coordinate and share data. By ensuring trust among these parties, blockchain can indirectly support improvements in QoE. For example, blockchain and smart contracts can be employed to manage a decentralized data market by optimizing query costs through cooperative search while respecting delay tolerance (Jiang and Wu,2022). Blockchain can also be employed to provide mechanisms such as TEEs for multi-domain orchestration, zk-SNARK–based proofs, and reputation systems that align incentives between providers and users, reduce privacy risk, and sustain fair compensation in decentralized settings (e.g., vehicular crowdsensing) (He et al.,2022;Zhao et al., 2022). 5.1.4. Differential Privacy DP provides a framework for protecting user information while still allowing data analysis across different network settings. It works by introducing random noise to the data or to the results of computations, such as during model training in FL, thereby obscuring the influence of any single individual (Ouadrhiri and Abdelhadi,2022). This approach reduces the risk of disclosing sensitive information about specific users. Research in this area demonstrates that stronger privacy guarantees, as shown in DP, reduce the accuracy of QoE analysis, highlighting the trade-off between privacy protection and service quality. For instance, Esper et al. (2023) examined privacy-preserving mechanisms in MEC, revealing how enhanced privacy measures, such as location approximation, necessarily increased system latency and reduced offloading efficiency, directly impacting QoE. DP applications range from anonymizing IoT data sources to enable secure and low-latency edge processing, to more domain-specific implementations (Le et al.,2023). For example, DP is now used to formally measure user satisfaction at different privacy levels in location-based services (Feng et al.,2024a). In immersive video, Gaussian noise (spatial) and Fourier perturbation (temporal) protect viewport/gaze data while preserving utility for predictive algorithms and adaptive bitrate allocation (Jin et al.,2024). 5.1.5. Other Approaches to Privacy Preservation Some studies have explored alternative privacy-preserving approaches, each with distinct advantages and limitations. User behavior trajectories in XR streaming, such as viewpoint and position, are consistently tracked and Garcia et al.: Preprint submitted to Elsevier Page 16 of 35 Privacy-Preserving QoE in Next-Gen Networks uploaded, which inherently creates a privacy vulnerability. This data is not anonymous, as studies show that users can be identified within minutes, which enables the inference of highly sensitive personal attributes (Wei et al.,2024). One strategy proposed to address the balance between QoE and privacy in XR streaming is to gradually reduce the amount of noise applied to trajectory data over time. This approach, known as Decreasing Noise Entropy (DNE), reduces noise entropy within a defined observation window, applying less distortion to samples that are most relevant for predicting the user’s viewpoint. This strategy allows DNE to reduce QoE loss by 55% to 100% compared to uniform noise at equivalent privacy levels, while keeping QoE degradation negligible (0–3%) even when maximum privacy is enforced (Wei et al.,2024). In IoT environments, this challenge can be addressed through a multi-layer QoE model that incorporates a “privacy preference” component. The model applies the Artificial Fish Swarm Algorithm to assign a score based on the user’s sensitivity to different data types and the provider’s reliability, expressed as an exponential function. Service selection is then guided by this score, prioritizing services that better reflect user privacy expectations while reducing exposure risks and maintaining QoE (Jia et al.,2020). Cryptographic systems and authentication mechanisms are prominent in preserving user privacy, but their contribution to QoE depends on how effectively they balance security guarantees with computational and network demands (Jia et al.,2022;Foko Sindjoung et al.,2023). For instance, hybrid cryptographic models that combine symmetric and asymmetric encryption provide a degree of resilience by limiting unauthorized access and reducing the risk of data corruption. Their effectiveness, however, lies not only in securing transmissions but also in supporting privacy-preserving authentication through dynamic key management. The generation of new session keys at each initiation prevents key reuse, which disrupts long-term user activity monitoring but also introduces the challenge of sustaining system performance under continuous renewal (Foko Sindjoung et al.,2023). Healthcare settings illustrate a related tradeoff: while hybrid schemes such as combining AES with attribute-based encryption can lower latency and computational overhead, thereby accelerating processes like disease prediction, they must still manage the tension between fine-grained access control and the resource constraints of real-time clinical systems (Jayaram and Prabakaran, 2021). In video services, adversarial perturbations can be applied to volumetric content to obscure 3D face models while preserving perceptual quality for viewers (Tang et al.,2020). The strength of this approach lies in the imperceptibility of the perturbations, which preserves QoE, yet its deployment highlights an underlying vulnerability: without such protections, volumetric data may be exploited to bypass authentication systems. Data isolation protects privacy at the network level by limiting the amount of information shared between SPs. In this model, the client acts as a mediator between the CDN and ISP, transmitting only a list of CDN addresses to the ISP’s PPNet server for network optimization,whilewithholding the specific content requested.Atthe same time,the CDN retains networkperformance data, ensuring that the ISP’s proprietary information is not disclosed (Akpinar and Hua,2020). Both adversarial perturbations and data isolation reflect the principle of minimizing information exposure, though they intervene at different levels: one modifies the content itself, whereas the other restricts metadata exchange. 5.2. Challenges in Privacy-Preserving Implementation The reviewed studies highlight recurring challenges, which we grouped into five categories: Understanding User Needs and Network Monitoring, Privacy Beyond Encrypted Traffic, Communication and Scalability, Security and Authentication, and Service Integration Between SPs and MNOs. Table 10 summarizes these categories and their impact on QoE. 5.2.1. Understanding User Needs and Monitoring Network Performance As outlined in Section 5.1.2, end-to-end encryption prevents DPI and constrains QoE monitoring. This directly impacts their ability to: (i) guarantee QoS and QoE; (ii) manage network resources and optimize services; (iii) reliably predict service performance, such as edge offloading, even when privacy protections reduce the precision of user location data; and (iv) understand complex user interactions and service behavior in mobile applications. A related challenge is that encryption not only constrains providers’ ability to monitor QoE but also limits users’ ability to track how their data is being used. IoT devices can exfiltrate private data, such as voice recordings, to manufacturer servers without user consent. This exposes the limitations of TLS in IoT contexts, where encrypted traffic prevents user oversight and creates privacy and security trade-offs, affecting QoE (Zhou et al.,2020). A key challenge for operators is not the existence of methods to infer QoE from encrypted traffic (as discussed in Section 5.1.2), but their practical deployment. Two specific issues dominate: Garcia et al.: Preprint submitted to Elsevier Page 17 of 35 Privacy-Preserving QoE in Next-Gen Networks •ML for QoE/KPI inference from encrypted traffic: The effectiveness of ML presented in Section 5.1.2 depends on the availability of suitable data and the ability to generalize across diverse environments. Acquiring large, labeled datasets that capture application-level KPIs alongside network traffic is resource-intensive and difficult to scale, especially for mobile services under variable conditions (Orsolic and Skorin-Kapov,2020). Even when models are trained successfully, they often perform poorly outside controlled settings, as device heterogeneity, fluctuating network states, and user behavior introduce unseen variations (Wassermann et al., 2020c). These challenges highlight the need for automated data collection pipelines and model designs that remain reliable across different applications and deployment contexts. •Generalization and device heterogeneity: Even when sufficient training data is available, ML models often fail to generalize across diverse usage contexts. Models built for a single device type or service, such as desktops or a specific streaming platform, lose accuracy when applied to smartphones, tablets, or other applications (Wassermann et al.,2020b). This problem is intensified in mobile environments, where fluctuating network conditions and variable user behavior introduce patterns unseen during training (Casas et al.,2021). To improve performance across different contexts, some approaches, such as multi-device and multi-content training, produce models that are more reliable in real-world deployments (Wassermann et al.,2020b). Other strategies within FL also address heterogeneity. One approach is to train separate models for different user groups, preserving group-specific data patterns (Porcu et al.,2022). Another is to expand local datasets with synthetic samples, which helps reduce bias, speeds up training, and lowers communication overhead (Ickin et al.,2020). A further option is vertical FL (vFL), where models are trained collaboratively across nodes that each hold different subsets of features. This approach has been shown to outperform models trained independently on local data (Ickin et al.,2021). 5.2.2. Integrating Privacy Considerations (beyond just Encrypted Traffic Inference) While encryption inherently protects privacy from external observers, some applications (e.g., Location-Based Services, Edge Offloading, Immersive Video) require sharing user data, which then reintroduces privacy concerns. DP discussed in Section 5.1.4 has been proposed as a privacy mechanism. However, its main challenge is structural: stronger privacy guarantees generally reduce the usefulness of the data. Prior studies (Esper et al.,2023;Feng et al., 2024a) show this trade-off explicitly, with Feng et al. (2024a) demonstrating that higher DP parameter values reduce the level of location privacy available to users. For sensitive data like location, viewport, or gaze motion, traditional DP methods may fall short due to strong temporal or spatial correlations. Novel adaptations are proposed (e.g., in immersive video streaming) to introduce noise while accounting for these correlations, aiming to maintain data usefulness for service optimization (Jin et al., 2024). Parameter configuration presents another significant challenge, as explored by Jin et al. (2024) in their work on privacy-preserving video streaming. In the context of temporal privacy using Fourier perturbation, they emphasize that choosing the DP parameter involves a trade-off: smaller values increase reconstruction error, while larger values raise perturbation error. In centralized settings, the parameter must be chosen in a differentially private way to protect data privacy. A central privacy challenge in proactive tile-based VR video streaming is the exposure of sensitive information through traces of users’ FoV and eye movements. This practice raises privacy concerns since the collected data can disclose sensitive personal information. One proposed mitigation strategy is to obscure the true FoV by mixing actual requests with additional camouflaged ones. By sending both real and decoy tile requests, the server faces greater uncertainty in identifying the genuine FoV. Privacy in this setting can be quantified through the spatial degree of privacy (SDoP), which represents the normalized number of camouflaged requests relative to the actual FoV. A key privacy concern in decentralized IoT services is the unintended exposure of user data during distributed processing. In the Social Internet of Things (SIoT), service discovery has been addressed through locality-sensitive hashing (LSH), an approach that limits dependence on central servers, lowers privacy risks, and enables distributed recommendation systems (Yan et al.,2019). Comparable issues arise in the IoV, where offloading services can reveal sensitive information such as drivers’ locations and personal data. 5.2.3. Communication Overhead, Scalability, and Fairness in Decentralized QoE Monitoring Systems Decentralized QoE monitoring distributes computation across devices, creating challenges in communication efficiency, scalability, and fairness. High communication costs, device heterogeneity, and latency constraints hinder Garcia et al.: Preprint submitted to Elsevier Page 18 of 35 Privacy-Preserving QoE in Next-Gen Networks real-time performance, while unreliable or strategic participants threaten fair contribution and system integrity. These challenges are discussed as follows: •Communication overhead and system performance in FL strategies: This challenge arises from the distributed nature of FL (Porcu et al.,2022). Different studies employed methods to reduce the overhead. One approach introduces a federated multi-agent reinforcement learning framework that explicitly models system cost in terms of service latency, energy consumption, and task drop rate. This approach directly quantifies and aims to enhance multi-user QoE (Li et al.,2022). By aggregating only critic networks (rather than raw data or full models) with an attention mechanism, the framework reduces transmission data and network congestion, enabling decentralized offloading decisions while preserving user privacy and improving convergence. For scalable video streaming, an adaptive FL-based algorithm dynamically adjusts streaming parameters, such as bitrate and resolution, in real time based on user context, network conditions, and device capabilities (Darwich et al.,2024). This decentralized data processing at edge nodes minimizes central data aggregation, leading to a 25% reduction in bandwidth usage and an 18% improvement in user satisfaction compared to traditional methods, while maintaining optimal video quality (Darwich et al.,2024). Device heterogeneity, unstable channel conditions, and user mobility introduce further complications, particularly clients that become inactive or unresponsive during training, for instance, due to weak wireless connections, loss of coverage. Since client availability changes unpredictably, the server must wait for responses, which slows training progress. One approach proposes maintaining multiple global models with distinct trade-offs between computational demand and learning performance. This strategy allows clients to adopt models suited to their current resources and conditions, ensuring adaptability and improved QoE for end users (Bai et al.,2023). •Scalability in decentralized QoE approaches: Table 9presents a comparative evaluation of different challenges and corresponding approaches. Although different strategies have been proposed to reduce overhead, such as aggregating only partial model components like critic networks or exchanging only intermediate representations in vertical FL (vFL), their operational robustness is often insufficiently proven. For instance, the resilience of these models against dynamic network conditions, including intermittent client connectivity or packet loss, is frequently unevaluated. The impact of such disruptions on the iterative convergence of vFL is not quantified, and while synchronous FL can exclude offline clients to prevent stalls (Corcuera Bárcena et al.,2023a), this comes at the cost of data diversity and potential model bias, with the long-term effects on accuracy remaining unexplored. Although decentralized methods advance non–real-time QoE monitoring and personalization, persistent latency issues render current frameworks inadequate for the stringent requirements of B5G/6G Ultra-Reliable LowLatency Communications (URLLC). For example, even in an optimized MEC-deployed FL system, model exchange and aggregation can take on the order of seconds (Corcuera Bárcena et al.,2023a). This fundamental mismatch is further obscured when studies report metrics like cumulative training time or communication overhead reduction instead of the crucial real-time, per-inference latency that defines URLLC. A scalability challenge in blockchain-based solutions is fundamentally tied to the choice of consensus mechanism, which determines the trade-offs between decentralization, scalability, energy efficiency, and security (Jain et al.,2025;Alghamdi et al.,2024). The use of Proof-of-Work (PoW), which provides strong decentralization but is impractical for most network services due to its energy consumption and extremely low transaction throughput, particularly focusing on B5G/6G applications (Jiang and Wu,2022;Zhao et al.,2022). In contrast, alternatives such as Practical Byzantine Fault Tolerance (PBFT)-based consensus mechanisms are designed for high-performance, permissioned environments, where they provide lower latency and higher transaction rates. However, these systems face their own scalability limitations; their performance degrades as the number of participating nodes increases due to the high communication overhead required to reach consensus (Nasir et al., 2022). Therefore, while PBFT-based solutions are suitable for applications with a limited number of trusted nodes (e.g., multi-operator agreements), they remain a challenge for large-scale, decentralized deployments. Another alternative for PoW is the Delegated Proof of Stake (DPoS) protocol (Li and Wan,2021), which concentrates power among a few delegate nodes, which can lead to corruption and less decentralization (Mišić et al.,2025). Garcia et al.: Preprint submitted to Elsevier Page 19 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 9 Comparative evaluation of scalability challenges in decentralized approaches. Scalability Challenge Proposed Strategy / Mechanism Benefits Identified Limitations High Communication Overhead in FL Selective Model Aggregation: partial model component aggregation (Li et al.,2022;Ickin et al.,2021) •Reduced data volume per update. •Lowered network congestion. •Does not address the high latency of the aggregation step. •Real-world resilience to packet loss and client dropout is often unverified. Slow and Energy-Intensive Nature of PoW Blockchain Consensus Adoption of High-Performance Consensus: Replace PoW with faster, voting-based protocols (e.g., PBFT) in permissioned environments (Li and Zhang,2024). •Achieves lower energy consumption and higher transaction rates than PoW. •Fails to scale to large node counts due to high message complexity of PBFT (𝑂(𝑛2)) (Deshmukh et al., 2025). •Scalability claims often lack empirical validation beyond small-scale testbeds. Computational Cost of On-Chain Privacy Layering Cryptographic Privacy: Embed privacy-preserving computations by integrating cryptographic methods such as TEEs and zk-SNARKs to enhance verifiability (He et al.,2022; Zhao et al.,2022). •Provides verifiability and auditable privacy guarantees for multi-domain interactions. •Adds computational overhead that compounds existing consensus latency. •Makes real-time use cases even less feasible due to added processing delays. To address blockchain’s throughput constraints in real-time network orchestration, high-performance BFT-based consensus mechanisms are employed to reduce communication overhead, though claimed scalability often lacks sufficient empirical validation. For example, one such framework was validated with only 18 nodes, yet claimed viability for “hundreds” (He et al.,2022). While demonstrating promising performance at this small scale (latency under 4 seconds), this limited validation fails to address communication complexity in larger BFT-based systems, thus leaving its feasibility for widespread B5G/6G deployment uncertain. Ensuring privacy in multi-domain network slice orchestration is challenging due to the need to balance effective coordination with the proprietary nature of network operators’ data. Privacy-preserving mechanisms, which combine TEEs and blockchain, introduce computational overhead. This stems from TEE-related cryptographic operations like key generation and attestation, along with the processing demands of blockchain consensus protocols. Similarly, the computational cost of zk-SNARK proof generation, including setup and transaction verification, continues to restrict scalability in high-throughput environments. •Fairness and incentivization mechanisms: Fairness andparticipant reliability remain concerns indecentralized QoE systems, as contributors may submit inaccurate resource claims or act strategically. Blockchain-based smart contracts have been proposed to enhance accountability by combining reputation management with incentive schemes that reward reliability and penalize malicious behavior (Zhao et al.,2022). Additionally, game-theoretic approaches have also been integrated into consensus protocols, such as bilateral evaluation strategies that align declared resources with actual performance, thereby reducing dishonest behavior and promoting equilibrium participation (He et al.,2022). However, their scalability and long-term stability remain uncertain, particularly in the event of unstable participation, collusion, or manipulation of reputation scores. 5.2.4. Security Challenges: Cryptographic Schemes and Authentication Breaches in security, loss of data privacy, and weak authentication can damage user trust. In healthcare, for example, such issues can compromise patient safety and confidence in the system (Jayaram and Prabakaran,2021). These problems directly affect how users experience and interact with services, making strong security essential for maintaining safety and satisfaction. In the context of authentication mechanisms, existing MEC schemes remain vulnerable to impersonation, replay, and denial-of-service attacks, and they fail to ensure user anonymity. These weaknesses, which reduce user QoE, can be mitigated by combining password and biometric authentication and incorporating anonymity-preserving mechanisms. These improvements strengthen security and privacy while reducing performance penalties that affect QoE (Jia et al., 2022). Furthermore, hybrid cryptographic schemes that combine symmetric and asymmetric encryption strengthen data security, preserve privacy, and support efficient authentication in resource-constrained environments (Foko Sindjoung Garcia et al.: Preprint submitted to Elsevier Page 20 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 10 A summary of the challenges in privacy-preserving QoE implementation and their impacts. Category Challenge Impact on QoE and System Performance Understanding User Needs and Monitoring Network Performance Loss of Visibility due to End-to-End Encryption Prevents direct QoE measurement via DPI, forcing operators to rely on less accurate inference methods (see Section 5.2.1). Model Generalization and Data Heterogeneity ML models for QoE inference fail to generalize across diverse devices and network conditions. Statistical heterogeneity in FL degrades model accuracy and stability (see Section 5.2.1). Integrating Privacy Considerations (beyond encrypted traffic) The Privacy-Utility Trade-off Stronger privacy guarantees, such as those from DP, often reduce data utility, which directly degrades the accuracy of QoE models and the effectiveness of personalized services (see Section 5.2.2). Privacy Risks in Decentralized Architectures Decentralized systems risk exposing sensitive user data (e.g., location, behavior) during service discovery and offloading, requiring protocols that may introduce additional overhead (see section 5.2). Communication, Scalability, and Fairness in Decentralized Systems Communication Overhead in Decentralized Systems Frequent exchange of model updates in FL leads to high communication overhead, increased latency, and slower convergence, impairing real-time QoE optimization capabilities (see section 5.2.3). Scalability of Decentralized Architectures Both FL and Blockchain face performance bottlenecks at scale. FL latency is often too high for real-time control, while blockchain consensus mechanisms limit throughput, making them unsuitable for latency-sensitive QoE management (see Section 5.2.3). Fairness and Incentivization in Collaborative Systems Ensuring fair participation is difficult, allowing malicious or freeriding behavior to compromise the integrity of the collective model and degrade overall data quality (see Section 5.2.3). Security and Authentication Security and Authentication Vulnerabilities Weak authentication schemes are vulnerable to attacks, while achieving strong user anonymity creates trade-offs with system security and traceability, directly impacting user trust and QoE (see Section 5.2.4). Lack of Service Integration Between SPs and MNOs Inefficient Routing and Resource Underutilization Poor coordination leads to inefficient traffic routing that bypasses available edge resources, resulting in higher latency, network congestion, and degraded QoE for end-users (see Section 5.2.5). et al.,2023;Jayaram and Prabakaran,2021). Integrating lightweight authentication and key management mechanisms reduces reliance on secure channels, maintains user anonymity, and enables revocation of compromised keys (Foko Sindjoung et al.,2023). Overall, hybrid designs enable secure and privacy-preserving communication with lower computational and communication overhead, though their scalability and adaptability in dynamic mobile settings remain unresolved challenges (Jayaram and Prabakaran,2021). 5.2.5. Lack of Service Integration Between SPs and MNOs The absence of coordination between SPs and MNOs limits the on-demand use of core and edge resources, reducing end-user QoE. Service requests typically traverse the full MNO infrastructure to reach central cloud servers or CDNs,bypassingunderusedMulti-access EdgeComputing resources that are often reservedforMNO-specificservices (De Deus et al.,2021). This inefficient routing leads to congestion, delay variation, and out-of-sequence packets, all of which impair QoE for multimedia applications. A unified method to address SP requirements, enable dynamic resource allocation, and protect user privacy is still lacking. This can beaddressed bycombiningautonomous resource managementwith networkslicing to extendservices at the edge while maintaining privacy (De Deus et al.,2021). Within a 5G setting, it leverages 5G-NEF to obtain user context and employs MEC service APIs to gather information from MEC servers. These mechanisms make it possible for MNOs to predict user mobility and allocate computing resources in advance to support module migration. 5.3. Comparison of Privacy Techniques Table 11 summarizes the advantages and limitations of the main privacy-preserving approaches in QoE research, highlighting their comparative features to support design considerations. Additionally, Figure 6compares these Garcia et al.: Preprint submitted to Elsevier Page 21 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 11 Comparison of privacy-preserving techniques and their characteristics. Technique Key Principle Advantages Limitations FL Keeps training data local, preserving privacy (Corcuera Bárcena et al., 2023a,b;Nguyen et al.,2023) •Enables collaborative training without data sharing (Corcuera Bárcena et al., 2023a) •May be less accurate than centralized training (Corcuera Bárcena et al.,2023a) •Outperforms isolated models (Corcuera Bárcena et al.,2023b) •Aggregation and non-IID data issues (Nguyen et al.,2023) •Reduces data transmission (Ickin et al., 2021) •Depends on consistent client participation (Corcuera Bárcena et al.,2023b) •Supports privacy-focused applications (Guo et al.,2023) •Risk of leakage from shared models (Ickin et al.,2021) DP Adds noise to data or outputs to reduce leakage (Zhang et al., 2022;Jin et al.,2024) •Protects data in collaborative learning (Ickin et al.,2021) •Adds complexity in the system design (Jin et al.,2024;Feng et al.,2024a) •Noise may reduce utility or accuracy (Feng et al.,2024a) Blockchain Decentralized, tamper-resistant data security (He et al.,2022) •Ensures data integrity and transparency, contributing to trust (He et al.,2022) •High storage and computation overhead (Nyangaresi,2022) •Removes central points of control (He et al.,2022) •Costly to maintain and scalability issues (Nguyen et al.,2023) •Smart contract support (Zhong et al., 2021) •High transaction fees (Nguyen et al.,2023) Encrypted Traffic Inference Infers QoE metrics from encrypted data flows (Islam et al.,2023) •Applies ML/statistics to traffic patterns (Islam et al.,2023) •DPI fails with full encryption (Orsolic and Skorin-Kapov,2020) •Effective for encrypted Web and VR QoE (Islam et al.,2023) •Obscures application-layer metrics (Orsolic and Skorin-Kapov,2020) •Requires advanced techniques for detection tasks (Nyangaresi,2022) Privacy Strength Communication Overhead Scalability Real-Time Suitability Trust 1 2 3 Federated Learning Blockchain Differential Privacy Encrypted Traffic Inference Figure 6: Comparison of FL, Blockchain, DP, and Encrypted Traffic Inference across four dimensions. Higher values indicate stronger performance in each area. techniques across privacy, communication, scalability, and trust dimensions, highlighting their respective strengths and limitations as introduced in Sections 5.1 and 5.2 . 5.4. Evaluation Methods, Tools and Datasets The evaluation methods employed across the studies reveal a clear preference for experimental validation utilizing real-world testing environments to evaluate their proposed solutions. Table 12 provides an overview of the distribution of evaluation methods across the reviewed works. Experimental validation proves particularly valuable in assessing Garcia et al.: Preprint submitted to Elsevier Page 22 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 12 Distribution of evaluation methods in network privacy and security research. Evaluation Method Works Experiment He et al. (2022), Bordel Sánchez et al. (2024), Le et al. (2023), Zhou et al. (2020), Dillbary et al. (2024), Liu et al. (2020a), Jiang and Wu (2022), Akpinar and Hua (2020), Xu et al. (2023), Bai et al. (2023), Valente et al. (2023), Gutterman et al. (2020), Wassermann et al. (2020a), Wassermann et al. (2020c), Casas et al. (2021), Wehner et al. (2021), Yan et al. (2019), Xu et al. (2021), Renda et al. (2022), Islam et al. (2023), Jayaram and Prabakaran (2021), Gong et al. (2020), Al-Quraan et al. (2024), Ickin et al. (2021), Jin et al. (2024), Zhong et al. (2021), Robitza et al. (2020), Guo et al. (2023), Porcu et al. (2022), Feng et al. (2024b), Nguyen et al. (2023), Ickin et al. (2020), De Deus et al. (2021), Tang et al. (2020) Simulation Wei and Yang (2022), Corcuera Bárcena et al. (2023a), Corcuera Bárcena et al. (2023b), Li et al. (2022), Zhang et al. (2022), Wei et al. (2024), Ickin et al. (2023), Alabbasi et al. (2021), Khanal et al. (2021), Liu et al. (2020b), Feng et al. (2024a), Foko Sindjoung et al. (2023), Esper et al. (2023), Bechini et al. (2023), Shan et al. (2020), Zhao et al. (2024), Maale et al. (2023), Md. Fadlullah and Kato (2022), Li and Wan (2021), Zhao et al. (2022), Jia et al. (2020), Liang et al. (2022), Li and Zhang (2024) Experiments and Simulations Gao et al. (2020), Darwich et al. (2024), Orsolic and Skorin-Kapov (2020) Formal Security Analysis and Experiments Jia et al. (2022), Lin (2022), Nyangaresi (2022) the real-world performance of privacy-preserving techniques, especially in contexts like video streaming services and mobile networks, where actual network conditions impact results. Simulations form the second most common evaluation approach. These typically involve controlled environment testing, particularly beneficial for evaluating complex network architectures or privacy mechanisms where real-world testing might prove impractical or cost-prohibitive. A smaller number of studies combined both experimental and simulation approaches, providing validation through both controlled and real-world testing scenarios. Similarly, some studies incorporated formal analysis alongside experiments, which is particularly valuable in evaluating cryptographic protocols and security mechanisms. Table 13 categorizes the main tools and technologies used for proof-of-concept development. The categories include machine learning, network analysis, simulation environments, and development platforms. Machine learning frameworks, particularly TensorFlow and scikit-learn, are among the most frequently applied, especially for implementing FL and developing predictive models. These tools are used to enhance QoE, protect user privacy, support personalization, identify influencing factors, and estimate perceived quality. They are also applied to assess network performance, including QoE metrics. Network analysistoolssuchasTcpdump areusedtocaptureandexamine networktraffic for performanceevaluation. Specialized simulation platforms allow researchers to test complex network scenarios under controlled conditions. The raw traffic traces gathered through these tools are used to build datasets for training machine learning models that infer Web QoE from encrypted traffic. Simulation platforms like Simu5G also help generate realistic and diverse datasets by linking QoS and application-level metrics to QoE, supporting both the training and evaluation of models for QoE prediction and optimization. Table 14 shows the datasets used across the reviewed studies. Simulation-based datasets were the most common. Several studies used established QoE datasets like the Waterloo Streaming QoE Database III and Poqemon-QoEDataset for video streaming analysis. Domain-specific datasets included YouTube traffic data, VR viewing patterns, and mobile network traces. Machine learning benchmark datasets such as CIFAR-10 and MNIST were used for testing learning algorithms. 5.5. An Analysis of Privacy Threats in QoE Monitoring and Prediction To provide a structured overview, we use a threat-model-based taxonomy to analyze key privacy challenges. The analysis links the actors involved, the privacy risks they pose, and the mitigation techniques covered in this survey. Garcia et al.: Preprint submitted to Elsevier Page 23 of 35 Privacy-Preserving QoE in Next-Gen Networks Table 13 Tools and technologies used in privacy-preserving QoE research. Category Tools Works Primary Application Machine Learning Frameworks TensorFlow Xu et al. (2023), Ickin et al. (2023), Khanal et al. (2021), Maale et al. (2023), Md. Fadlullah and Kato (2022), Le et al. (2023) Federated and Split Learning Implementation scikit-learn Corcuera Bárcena et al. (2023a), Ickin et al. (2020), Gutterman et al. (2020)Model Training and Evaluation Keras Gao et al. (2020), Guo et al. (2023) Deep Learning Applications PyTorch Li and Zhang (2024) Neural Network Implementation Flower Valente et al. (2023), Porcu et al. (2022) FL Framework Network Analysis Tools Tcpdump Islam et al. (2023), Orsolic and Skorin-Kapov (2020)Traffic Analysis WebPageTest and RUMSpeedIndex Wassermann et al. (2020a) Web Performance Measurement Simulation Platforms Simu5G Renda et al. (2022), Bechini et al. (2023) 5G Network Simulation SUMO Esper et al. (2023) Simulate Different User Mobility Patterns MATLAB and Opnet Liu et al. (2020b) Network Performance Testing Ganache Bordel Sánchez et al. (2024) Blockchain Simulation Development Tools Appium and Tcpdump Wehner et al. (2021) Mobile Application Testing FFmpeg and Netfilter De Deus et al. (2021) Multimedia Processing RabbitMQ Ickin et al. (2021) Message Queue Management Specialized Tools Microsoft CCF Platform Xu et al. (2023) Cryptographic Operations Node.js, Express Ickin et al. (2023) AI-based QoE Prediction Simu5G, OMNeT++ Gao et al. (2020) Traffic Simulation Ganache Wehner et al. (2021) Blockchain Applications It uses previous studies as references, including standard adversary classifications and privacy metrics (Wagner and Eckhoff,2018), as well as domain-specific models like the Bhadra framework for mobile systems (Rao et al.,2023). Table 15 outlines the threat model itself. It identifies the key actors in the QoE ecosystem (e.g., Network Operators, SPs), their capabilities, their level of data visibility, and the primary privacy threats they introduce. These capabilities define how the adversary behaves: passive adversaries observe data without interfering, active adversaries can modify data, internal adversaries have access from within the system, external ones attack from the outside, and honest-butcurious adversaries follow the system’s rules but attempt to infer private information. Table 16 maps each threat scenario to a specific privacy-preserving technique. For each case, it clarifies the tradeoff by defining both the privacy metric (how privacy is measured) and the corresponding utility metric (how QoE performance is measured), and points to the relevant survey section for a detailed discussion. 6. Discussion 6.1. Architectural Approaches: Network-centric vs Application-centric Our findings show that the distinction in network scenarios is architectural, defined by the role and data visibility of the monitoring entity. This leads to two approaches to QoE monitoring: a network-centric model for network operators (MNOs/ISPs) and an application-centric model for SPs. The emphasis of 5G and later technologies highlights how privacy and user experience are now directly linked to dense, low-latency deployments. In the network-centric model, operators infer QoE from encrypted traffic rather than application payloads (see Section 5.1.2). This yields real-time detection of coarse QoE events but remains limited for subjective, user-centered Garcia et al.: Preprint submitted to Elsevier Page 24 of 35 Privacy-Preserving QoE in Next-Gen Networks 7.6. Data Availability and Dataset Limitations Limited access to public QoE datasets can constrain the scope of evaluations (Ickin et al.,2020), and relying on a single dataset can limit the applicability of findings (Porcu et al.,2022). Public datasets that combine network traces, ground-truth application KPIs, and subjective user QoE scores (e.g., MOS) remain underdeveloped. As a result, most studies rely on limited, siloed, or synthetic data (Ickin et al.,2020). The reliance on synthetic datasets reflects the persistent lack of publicly available real-world network and application data (Ickin et al.,2023). Many studies focus primarily on YouTube data, highlighting the need for further validation across other videostreaming platforms (Wassermann et al.,2020c). The exclusion of mobile platforms in some passive data collection methods, the low incidence of stalling events in certain datasets limiting insights into severe quality issues, and the limitations of using ITU-T P.1203 for evaluating long-form video content QoE all point to the need for more comprehensive data sources (Robitza et al.,2020). Furthermore, assumptions made due to dataset limitations, such as user equipment not sending feedback in predictive models study (Alabbasi et al.,2021), can influence the scope and applicability of research findings. 7.7. Evaluation Metrics and Scope of Analysis There is no standard framework to jointly evaluate a system across the three key axes: Privacy (e.g., DP budget, kanonymity), QoE/Utility (e.g., prediction accuracy, MOS improvement), and Performance (e.g., latency, computational overhead). This makes it difficult to compare the trade-offs of different approaches objectively. The lack of a QoE evaluation system designed for volumetric video streaming represents a research gap given the unique characteristics and higher dimensionality of volumetric content compared to traditional 2D or 360-degree videos (Tang et al.,2020). The detailed energy consumption of UAVs and edge nodes over extended periods is often overlooked (Md. Fadlullah and Kato,2022), and the secure communication between 5G links and core networks is sometimes considered outside the scope of specific studies (Lin,2022). This research direction includes investigating alternative generative models and approaches on a broader spectrum of QoE datasets (Ickin et al.,2020) and addressing the challenges of multi-source data distributions and model scalability in real-time FL environments (Renda et al.,2022). Collecting more diverse datasets containing a wider range of user behaviors and content types for applications like content caching is also a priority (Md. Fadlullah and Kato,2022). Improving evaluation methodologies involves refining QoE definitions beyond simple positive/negative scales (Guo et al.,2023), exploring more granular application and service-specific QoE slices (Guo et al.,2023), and focusing on predictinguser engagementoverlongervideodurations (Robitzaetal.,2020).This would allowresearchers to plot solutions on a 3D space (e.g., Privacy, Utility, Performance), enabling clear, quantitative comparisons of the trade-offs involved. Expanding current QoE models to handle longer sessions (Robitza et al.,2020), developing better QoE evaluation metrics for volumetric videos (Tang et al.,2020), and understanding the reasons behind video abandonment (e.g., boredom or technical issues) are all crucial areas of focus (Robitza et al.,2020). Addressing open issues related to model training efficiency and a more comprehensive evaluation of network footprint are also important considerations (Ickin et al.,2020). 7.8. The Subjectivity of Experience QoE is inherently subjective. Modern applications, especially in VR/AR, involve factors like cybersickness, presence, and cognitive load that are not captured by traditional network metrics (Tang et al.,2020). Future QoE models must integrate physiological data (e.g., from EEG, ECG, EDA) alongside network QoS as presented in Section2.2. The challenge is to analyze this uniquely sensitive data without creating new privacy risks. A promising research direction is using on-device FL to transform a user’s raw measurements into higher-level insights. This includes processing cardiovascular signals like ECG and heart rate to infer stress (Vijayakumar et al.,2024), gaze and viewport traces from immersive video to measure user attention (Jin et al.,2024), and facial expression metadata is used to assess emotions (Tao et al.,2019), ensuring that the underlying personal data remains private and is not directly exposed. 7.9. User-Centric Privacy and Trust End-users often have little visibility into how their data is used for QoE monitoring by either SPs or Network MNOs, what privacy-utility trade-offs are embedded in application designs or network management policies, or how to exercise their right to opt-out. Future research should therefore focus on combining privacy-preserving techniques with transparent governance mechanisms (Garcia et al.,2022). This would enable systems to explain to users, and potentially to regulators, why certain data access is needed for a specific QoE enhancement (e.g., enabling a low-latency network Garcia et al.: Preprint submitted to Elsevier Page 31 of 35 Privacy-Preserving QoE in Next-Gen Networks slice) and provide intuitive controls to manage these preferences. Furthermore, exploring decentralized identity and verifiable credentials can empower users with sovereign control over their data. For example, a user could grant an SP or MNO a time-limited, verifiable credential to access specific QoE-related data without revealing their full identity, giving users direct control over their data. 7.10. Exploring New Applications and Privacy Techniques Theadvancement of applications and techniquesinprivacy is being pursued throughseveralkeyresearch directions. One direction involves the expansion of existing privacy-enhancing systems to new industrial sectors (Bordel Sánchez et al.,2024), including the application of established frameworks to diverse case studies and services beyond the automotive industry (Corcuera Bárcena et al.,2023a). Furthermore, future research will explore the potential of FL and Explainable AI in domains such as healthcare and smart industry applications. In parallel to application expansion, efforts are in progress to generalize encryption systems to accommodate a broader range of streaming services and investigate their effectiveness across various video-streaming protocols (Wassermann et al.,2020c). The real-time classification capabilities of encrypted traffic inference frameworks are currently being analyzed (Dillbary et al.,2024), with ongoing work focused on refining the contribution of individual features to improve classification accuracy. Exploring the application of federated DRL for video caching and transcoding presents another promising area of research (Li and Wan,2021). Moreover, researchers are investigating the feasibility of directly utilizing QoS to QoE models for XR services within the data plane, specifically by evaluating the use of OpenRAN architectures as x-Apps/rApps (Islam et al.,2023). Integrating computational sentiment analysis to enhance network management strategies is also considered a potential research direction (Guo et al.,2023). Furthermore, studies are examining adaptive privacy level mechanisms based on QoS and mobility patterns (Esper et al.,2023), alongside the evaluation of different MEC selection algorithms under privacy constraints. In the realm of Web QoE, future research includes generalizing Web QoE models for multi-device environments (Wassermann et al.,2020a) and improving flow-level feature extraction techniques to decrease the computational overhead of Web QoE monitoring. Future research must strongly prioritize integrating user-centric QoE metrics in conjunction with privacy protection mechanisms. In these next-generation networks, where sensitive user data may be increasingly utilized to personalize experiences, privacy protection will be crucial to enable user trust and the ethical adoption of these services (Mao et al., 2023). In addition to privacy protection, 6G’s QoE assessments may incorporate emotional states, contextual factors, and physiological data. These elements are becoming necessary for achieving comprehensive, user-centric evaluations that capture an individual’s experience while maintaining privacy in 6G systems. Since massive data collection is required for AI-driven operations and IoT-enabled environments in 6G (Sun et al.,2020;Nguyen et al.,2021). Investigating the use of physiological metrics such as heart rate, EEG, and EDA, already demonstrated as strong indicators in AR/VR contexts (Keighrey et al.,2021;Salgado et al.,2018;Zhang et al.,2024), offers potential for future QoE modeling. Further exploration into subjective and physiological measures, such as vibrotactile feedback perception, overall quality ratings, audio fidelity, ECG, respiration, calmness, and energy levels (as shown in (Vijayakumar et al.,2024;Parvez et al.,2023;Zhang et al.,2024)), to capture the nuanced user experience, especially in multimedia and VR applications. In video streaming and immersive environments, future QoE evaluations should cover a wide range of metrics. Visual fidelity remains central, including resolution, perceptual quality (Tran et al.,2024;Amirpour et al.,2024; Llorente et al.,2024), and SNR (Wang et al.,2024b). Spatial factors, such as FoV in 360°video, are equally relevant (Wang et al.,2024c;Hou et al.,2024). Temporal dimensions, including frame rate and playback smoothness, also play a key role (Liu et al.,2023;Kokiadis et al.,2024). Beyond these, immersive and presence-related measures are needed to capture user experience in VR and 360°video (Elwardy et al.,2024;Althoff et al.,2023). At the same time, system-level metrics such as latency and re-buffering remain critical, as they directly influence user frustration (Farahani et al.,2024;Zeng et al.,2024). Expanding the scope to include responsiveness, usability, and emergingdimensions likecognitive load(Ouardietal.,2024;Elwardy etal.,2024) will beessentialfor a comprehensive understanding of QoE in next-generation interactive and resource-intensive applications. 8. Conclusion The increasing prevalence of user-centric services in next-generation networks has made the user experience a central concern. This development introduces two key concerns: (i) future B5G/6G networks must concentrate on QoE instead of relying only on QoS metrics; and (ii) prioritizing QoE introduces new privacy challenges, as it often requires Garcia et al.: Preprint submitted to Elsevier Page 32 of 35 Privacy-Preserving QoE in Next-Gen Networks collecting sensitive user data. This survey investigated the intersection of QoE and privacy preservation, analyzing the trade-offs between enhancing user-perceived service quality and protecting sensitive user data. QoE comprises multiple dimensions containing not just QoS but also integrates user-centric factors, such as emotions, contextual influences, and physiological responses, into QoE metrics, presenting important privacy challenges for future network services. Key privacy-preserving techniques like FL, Encrypted Traffic Inference, Blockchain, and DP were examined. FL enables collaborative QoE model training without exposing raw data, but faces scalability and data heterogeneity challenges. Traffic analysis infers QoE from encrypted patterns, preserving payload privacy, but it depends on feature engineering since DPI is largely ineffective as encryption hides the application-layer content traditionally used for direct QoE assessment. Blockchain offers a trust and transparency layer, though latency from consensus mechanisms can be a constraint. DP provides privacy guarantees by adding noise, requiring a balance with data utility. Persistent challenges include solution scalability in heterogeneous environments, real-time processing demands coupled with overhead from privacy technologies, and the fundamental privacy-utility trade-off. Data sparsity, heterogeneity, and domain-specific needs (e.g., XR data sensitivity, IoV responsiveness) add complexity. Research directions should focus on enhancing scalability for large network deployments, particularly by managing dynamic and diverse network environments, as 6G networks will support user-centric applications. Key areas include optimizing communication efficiency and latency in FL settings, enhancing resilience against model poisoning attacks, and developing privacy mechanisms that can adapt to varying network conditions. Furthermore, this involves establishingtransparentgovernance frameworksthat grantusers clear control overhowtheir data is utilizedby telecoms and SPs. Research could investigate real-time optimization techniques and improve the generalization capabilities of QoE prediction models. Additional focus areas include reducing computational overhead while maintaining privacy guarantees, exploring quantum-resistant security mechanisms, and developing frameworks that can handle more complex IoT and edge computing scenarios. Furthermore, researchers should address the challenges of standardization and interoperability across different network architectures while ensuring consistent privacy protection and QoE levels. Acknowledgments This work was supported by Ericsson Telecomunicações Ltda, and by the Sao Paulo Research Foundation, grant 2021/00199-8, CPE SMARTNESS, and grant 2023/15919-1. CRediT authorship contribution statement Rodrigo Dutra Garcia: Writing – original draft, Visualization, Validation, Methodology, Conceptualization. Gowri Sankar Ramachandran: Writing – review & editing, Supervision, Conceptualization. Christian Esteve Rothenberg: Writing – review & editing, Supervision. Bhaskar Krishnamachari: Writing – review & editing, Supervision. Jó Ueyama: Writing – review & editing, Supervision, Conceptualization. Declaration of interests ———————————————————————– Rodrigo Dutra Garcia reports financial support was provided by State of Sao Paulo Research Foundation. Christian Esteve Rothenberg reports financial support was provided by State of Sao Paulo Research Foundation. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. ———————————————————————– References Ahmad, A., Mansoor, A.B., Barakabitze, A.A., Hines, A., Atzori, L., Walshe, R., 2021. Supervised-learning-based qoe prediction of video streaming in future networks: A tutorial with comparative study. IEEE Communications Magazine 59, 88–94. URL: http://dx.doi.org/10.1109/ MCOM.001.2100109, doi:10.1109/mcom.001.2100109. Akpinar, K., Hua, K.A., 2020. Ppnet: Privacy protected cdn-isp collaboration for qos-aware multi-cdn adaptive video streaming. ACM Transactions on Multimedia Computing, Communications, and Applications 16, 1–23. URL: http://dx.doi.org/10.1145/3379983, doi:10.1145/3379983. Al-Quraan, M., Zoha, A., Centeno, A., Salameh, H.B., Muhaidat, S., Imran, M.A., Mohjazi, L., 2024. Enhancing reliability in federated mmwave networks: A practical and scalable solution using radar-aided dynamic blockage recognition. IEEE Transactions on Mobile Computing 23, 10146–10160. URL: http://dx.doi.org/10.1109/TMC.2024.3373529, doi:10.1109/tmc.2024.3373529. Garcia et al.: Preprint submitted to Elsevier Page 33 of 35 Privacy-Preserving QoE in Next-Gen Networks Alabbasi, A., Ganjalizadeh, M., Vandikas, K., Petrova, M., 2021. On cascaded federated learning for multi-tier predictive models, in: 2021 IEEE International Conference on Communications Workshops (ICC Workshops), IEEE. p. 1–7. URL: http://dx.doi.org/10.1109/ ICCWorkshops50388.2021.9473881, doi:10.1109/iccworkshops50388.2021.9473881. Alghamdi, T.A., Khalid, R., Javaid, N., 2024. A survey of blockchain based systems: Scalability issues and solutions, applications and future challenges. IEEE Access 12, 79626–79651. URL: http://dx.doi.org/10.1109/access.2024.3408868, doi:10.1109/access.2024. 3408868. Althoff, L.S., Farias, M.C.Q., Silva, A.R., Carvalho, M.M., 2023. Impact of alignment edits on the quality of experience of 360°videos. IEEE Access 11, 108475–108492. URL: http://dx.doi.org/10.1109/ACCESS.2023.3319346, doi:10.1109/access.2023.3319346. Amirpour, H., Zhu, J., Le Callet, P., Timmerer, C., 2024. A real-time video quality metric for http adaptive streaming, in: ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE. p. 3810–3814. URL: http://dx.doi.org/10. 1109/icassp48485.2024.10446839, doi:10.1109/icassp48485.2024.10446839. Amour, L., Boulabiar, M.I., Souihi, S., Mellouk, A., 2018. An improved qoe estimation method based on qos and affective computing, in: 2018 International Symposium on Programming and Systems (ISPS), IEEE. p. 1–6. URL: http://dx.doi.org/10.1109/ISPS.2018.8379009, doi:10.1109/isps.2018.8379009. Arellano-Uson, J., Magaña, E., Morato, D., Izal, M., 2024. Survey on quality of experience evaluation for cloud-based interactive applications. Applied Sciences 14, 1987. URL: http://dx.doi.org/10.3390/app14051987, doi:10.3390/app14051987. Artioli, E., 2024. Generative ai for http adaptive streaming, in: Proceedings of the ACM Multimedia Systems Conference 2024 on ZZZ, ACM. p. 516–519. URL: http://dx.doi.org/10.1145/3625468.3652912, doi:10.1145/3625468.3652912. Bai, Y., Chen, L., Li, J., Wu, J., Zhou, P., Xu, Z., Xu, J., 2023. Multicore federated learning for mobile-edge computing platforms. IEEE Internet of Things Journal 10, 5940–5952. URL: http://dx.doi.org/10.1109/JIOT.2022.3224239, doi:10.1109/jiot.2022.3224239. Barakabitze, A.A., Barman, N., Ahmad, A., Zadtootaghaj, S., Sun, L., Martini, M.G., Atzori, L., 2020. Qoe management of multimedia streaming services in future networks: A tutorial and survey. IEEE Communications Surveys & Tutorials 22, 526–565. URL: http://dx.doi.org/10. 1109/COMST.2019.2958784, doi:10.1109/comst.2019.2958784. Baraković, S., Husić, J.B., 2022. The importance of security matters for quality of experience in mobile web context. International Journal of Human–Computer Interaction 39, 1712–1722. URL: http://dx.doi.org/10.1080/10447318.2022.2072454, doi:10.1080/10447318. 2022.2072454. Bechini, A., Daole, M., Ducange, P., Marcelloni, F., Renda, A., 2023. An application for federated learning of xai models in edge computing environments, in: 2023 IEEE International Conference on Fuzzy Systems (FUZZ), IEEE. p. 1–7. URL: http://dx.doi.org/10.1109/ FUZZ52849.2023.10309783, doi:10.1109/fuzz52849.2023.10309783. Ben Saad, S., Brik, B., Ksentini, A., 2023. Toward securing federated learning against poisoning attacks in zero touch b5g networks. IEEE Transactions on Network and Service Management 20, 1612–1624. URL: http://dx.doi.org/10.1109/TNSM.2023.3278838, doi:10. 1109/tnsm.2023.3278838. Benarous, L., Zeadally, S., Boudjit, S., Mellouk, A., 2025. A review of pseudonym change strategies for location privacy preservation schemes in vehicular networks. ACM Computing Surveys 57, 1–37. URL: http://dx.doi.org/10.1145/3718736, doi:10.1145/3718736. Bingöl, G., Porcu, S., Floris, A., Atzori, L., 2022. Qoe estimation of webrtc-based audiovisual conversations from facial expressions, in: 2022 16th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), IEEE. p. 577–584. URL: http://dx.doi.org/10. 1109/SITIS57111.2022.00092, doi:10.1109/sitis57111.2022.00092. Bingöl, G., Porcu, S., Floris, A., Atzori, L., 2024. Webrtc-qoe: A dataset of qoe assessment of subjective scores, network impairments, and facial & speech features. Computer Networks 244, 110356. URL: http://dx.doi.org/10.1016/j.comnet.2024.110356, doi:10.1016/j. comnet.2024.110356. Boffetti, E., Rago, A., Piro, G., Boggia, G., 2025. 5g-qoera: An integrated dataset for qoe assessment in 5gnr based on user mobility, radio map, scheduling decisions, and application details, in: 2025 IEEE Wireless Communications and Networking Conference (WCNC), IEEE. p. 1–6. URL: http://dx.doi.org/10.1109/WCNC61545.2025.10978283, doi:10.1109/wcnc61545.2025.10978283. Bonawitz, K., Ivanov, V., Kreuter, B., Marcedone, A., McMahan, H.B., Patel, S., Ramage, D., Segal, A., Seth, K., 2017. Practical secure aggregation for privacy-preserving machine learning, in: Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, ACM. p. 1175–1191. URL: http://dx.doi.org/10.1145/3133956.3133982, doi:10.1145/3133956.3133982. Bordel Sánchez, B., Alcarria, R., Ladid, L., Machalek, A., 2024. Using privacy-preserving algorithms and blockchain tokens to monetize industrial data in digital marketplaces. Computers 13, 104. URL: http://dx.doi.org/10.3390/computers13040104, doi:10.3390/ computers13040104. Bouraqia, K., Sabir, E., Sadik, M., Ladid, L., 2020. Quality of experience for streaming services: Measurements, challenges and insights. IEEE Access 8, 13341–13361. URL: http://dx.doi.org/10.1109/access.2020.2965099, doi:10.1109/access.2020.2965099. Casas, P., Wassermann, S., Wehner, N., Seufert, M., Schuler, J., Hossfeld, T., 2021. Mobile web and app qoe monitoring for isps - from encrypted traffic to speed index through machine learning, in: 2021 13th IFIP Wireless and Mobile Networking Conference (WMNC), IEEE. p. 40–47. URL: http://dx.doi.org/10.23919/WMNC53478.2021.9619058, doi:10.23919/wmnc53478.2021.9619058. Chen, W., Lan, F., Wei, H., Zhao, T., Liu, W., Xu, Y., 2024. A comprehensive review of quality of experience for emerging video services. Signal Processing: Image Communication 128, 117176. URL: http://dx.doi.org/10.1016/j.image.2024.117176, doi:10.1016/j.image. 2024.117176. Chiariotti, F., 2021. A survey on 360-degree video: Coding, quality of experience and streaming. Computer Communications 177, 133–155. URL: http://dx.doi.org/10.1016/j.comcom.2021.06.029, doi:10.1016/j.comcom.2021.06.029. Chillotti, I., Gama, N., Georgieva, M., Izabachène, M., 2019. Tfhe: Fast fully homomorphic encryption over the torus. Journal of Cryptology 33, 34–91. URL: http://dx.doi.org/10.1007/s00145-019-09319-x, doi:10.1007/s00145-019-09319-x. Choon, K.M., Lim, T.M., 2024. A critical review on quality of experience for videos and user engagement on social media platforms, in: 2024 3rd International Conference on Digital Transformation and Applications (ICDXA), IEEE. p. 80–85. URL: http://dx.doi.org/10.1109/ Garcia et al.: Preprint submitted to Elsevier Page 34 of 35 Privacy-Preserving QoE in Next-Gen Networks ICDXA61007.2024.10470509, doi:10.1109/icdxa61007.2024.10470509. Corcuera Bárcena, J.L., Ducange, P., Marcelloni, F., Nardini, G., Noferi, A., Renda, A., Ruffini, F., Schiavo, A., Stea, G., Virdis, A., 2023a. Enabling federated learning of explainable ai models within beyond-5g/6g networks. Computer Communications 210, 356–375. URL: http://dx.doi.org/10.1016/j.comcom.2023.07.039, doi:10.1016/j.comcom.2023.07.039. Corcuera Bárcena, J.L., Ducange, P., Marcelloni, F., Renda, A., 2025. Increasing trust in ai through privacy preservation and model explainability: Federated learning of fuzzy regression trees. Information Fusion 113, 102598. URL: http://dx.doi.org/10.1016/j.inffus.2024. 102598, doi:10.1016/j.inffus.2024.102598. Corcuera Bárcena, J.L., Ducange, P., Marcelloni, F., Renda, A., Ruffini, F., Schiavo, A., 2023b. Federated tsk models for predicting quality of experience in b5g/6g networks, in: 2023 IEEE International Conference on Fuzzy Systems (FUZZ), IEEE. p. 1–8. URL: http://dx.doi.org/ 10.1109/FUZZ52849.2023.10309758, doi:10.1109/fuzz52849.2023.10309758. Dang, S., Amin, O., Shihada, B., Alouini, M.S., 2020. What should 6g be? Nature Electronics 3, 20–29. URL: http://dx.doi.org/10.1038/ s41928-019-0355-6, doi:10.1038/s41928-019-0355-6. Darwich, M., Khalil, K., Bayoumi, M., 2024. Scalable video streaming solutions using federated learning, in: 2024 International Conference on Smart Applications, Communications and Networking (SmartNets), IEEE. p. 1–6. URL: http://dx.doi.org/10.1109/SmartNets61466. 2024.10577738, doi:10.1109/smartnets61466.2024.10577738. De Deus, A.M.G., Julio, E.P., Moreno, M.F., 2021. Join-me: A framework for integrating service providers and mobile network operators, in: 2021 IEEE Symposium on Computers and Communications (ISCC), IEEE. p. 1–7. URL: http://dx.doi.org/10.1109/ISCC53001.2021. 9631393, doi:10.1109/iscc53001.2021.9631393. Deshmukh, A., Baiju, D., Atish, A., Alladi, T., Yu, F.R., 2025. An efficient and scalable byzantine fault tolerant consensus for vehicular networks. IEEE Transactions on Vehicular Technology 74, 13480–13491. URL: http://dx.doi.org/10.1109/TVT.2025.3561360, doi:10.1109/tvt.2025.3561360. Dillbary, N., Yozevitch, R., Dvir, A., Dubin, R., Hajaj, C., 2024. Hidden in time, revealed in frequency: Spectral features and multiresolution analysis for encrypted internet traffic classification, in: 2024 IEEE 21st Consumer Communications &; Networking Conference (CCNC), IEEE. p. 266–271. URL: http://dx.doi.org/10.1109/CCNC51664.2024.10454807, doi:10.1109/ccnc51664.2024.10454807. Dobreff, G., Szalay, M., Ladóczki, B., Molnár, M., Varga, L., Báder, A., Pašić, A., 2023. Data collection framework for end-to-end radio and transport network quality monitoring, in: 2023 15th International Conference on Quality of Multimedia Experience (QoMEX), IEEE. p. 127–130. URL: http://dx.doi.org/10.1109/QoMEX58391.2023.10178470, doi:10.1109/qomex58391.2023.10178470. Elwardy, M., Zepernick, H.J., Chu, T.M.C., Hu, Y., 2024. On the consistency of 360 video quality assessment in repeated subjective tests: A pilot study. EAI Endorsed Transactions on Industrial Networks and Intelligent Systems 11. URL: http://dx.doi.org/10.4108/eetinis. v11i1.4323, doi:10.4108/eetinis.v11i1.4323. Esper, J.P., Achir, N., Cardoso, K.V., Almeida, J.M., 2023. Impact of user privacy and mobility on edge offloading, in: 2023 IEEE 34th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), IEEE. p. 1–6. URL: http://dx.doi.org/10. 1109/PIMRC56721.2023.10293816, doi:10.1109/pimrc56721.2023.10293816. Farahani, R., undefinedetinkaya, E., Timmerer, C., Shojafar, M., Ghanbari, M., Hellwagner, H., 2024. Alive: A latencyand cost-aware hybrid p2p-cdn framework for live video streaming. IEEE Transactions on Network and Service Management 21, 1561–1580. URL: http: //dx.doi.org/10.1109/TNSM.2023.3335190, doi:10.1109/tnsm.2023.3335190. Feng, T., Zhang, Z., Wong, W.C., Sun, S., Sikdar, B., 2024a. A framework for tradeoff between location privacy preservation and quality of experience in location based services. IEEE Open Journal of Vehicular Technology 5, 428–439. URL: http://dx.doi.org/10.1109/OJVT. 2024.3364184, doi:10.1109/ojvt.2024.3364184. Feng, Y., Qi, Y., Li, H., Wang, X., Tian, J., 2024b. Leveraging federated learning and edge computing for recommendation systems within cloud computing networks, in: Li, W., Wang, H. (Eds.), Third International Symposium on Computer Applications and Information Systems (ISCAIS 2024), SPIE. p. 28. URL: http://dx.doi.org/10.1117/12.3034773, doi:10.1117/12.3034773. Foko Sindjoung, M.L., Velempini, M., Tayou Djamegni, C., 2023. A data security and privacy scheme for user quality of experience in a mobile edge computing-based network. Array 19, 100304. URL: http://dx.doi.org/10.1016/j.array.2023.100304, doi:10.1016/j.array. 2023.100304. Gangwal, A., Gangavalli, H.R., Thirupathi, A., 2023. A survey of layer-two blockchain protocols. Journal of Network and Computer Applications 209, 103539. URL: http://dx.doi.org/10.1016/j.jnca.2022.103539, doi:10.1016/j.jnca.2022.103539. Gao, Y., Wei, X., Zhou, L., 2020. Personalized qoe improvement for networking video service. IEEE Journal on Selected Areas in Communications 38, 2311–2323. URL: http://dx.doi.org/10.1109/JSAC.2020.3000395, doi:10.1109/jsac.2020.3000395. Garcia, R.D., Ramachandran, G.S., Jurdak, R., Ueyama, J., 2022. Blockchain-aided and privacy-preserving data governance in multi-stakeholder applications. IEEE Transactions on Network and Service Management 19, 3781–3793. URL: http://dx.doi.org/10.1109/TNSM.2022. 3225254, doi:10.1109/tnsm.2022.3225254. Ghafil, A.S., Ali, I.H., 2021. Video streaming forecast quality of experiencea survey, in: 2021 1st Babylon International Conference on Information Technology and Science (BICITS), IEEE. p. 299–304. URL: http://dx.doi.org/10.1109/bicits51482.2021.9509875, doi:10.1109/bicits51482.2021.9509875. Gong, C., Lin, F., Gong, X., Lu, Y., 2020. Intelligent cooperative edge computing in internet of things. IEEE Internet of Things Journal 7, 9372–9382. URL: http://dx.doi.org/10.1109/JIOT.2020.2986015, doi:10.1109/jiot.2020.2986015. Guo, W., Jin, B., Sun, S.C., Wu, Y., Qi, W., Zhang, J., 2023. Federated learning of wireless network experience anomalies using consumer sentiment, in: 2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), IEEE. p. 499–504. URL: http://dx.doi.org/10.1109/ICAIIC57133.2023.10067061, doi:10.1109/icaiic57133.2023.10067061. Gupta, M., Kumar, M., Dhir, R., 2025. Privacy-preserving federated learning in asynchronous environment using homomorphic encryption. Journal of Information Security and Applications 93, 104116. URL: http://dx.doi.org/10.1016/j.jisa.2025.104116, doi:10.1016/j.jisa. 2025.104116. Garcia et al.: Preprint submitted to Elsevier Page 35 of 35 Privacy-Preserving QoE in Next-Gen Networks Gutterman, C., Guo, K., Arora, S., Gilliland, T., Wang, X., Wu, L., Katz-Bassett, E., Zussman, G., 2020. Requet: Real-time qoe metric detection for encrypted youtube traffic. ACM Transactions on Multimedia Computing, Communications, and Applications 16, 1–28. URL: http://dx.doi.org/10.1145/3394498, doi:10.1145/3394498. He, G., Su, W., Gao, S., Liu, N., Das, S.K., 2022. Netchain: A blockchain-enabled privacy-preserving multi-domain network slice orchestration architecture. IEEE Transactions on Network and Service Management 19, 188–202. URL: http://dx.doi.org/10.1109/TNSM.2021. 3110057, doi:10.1109/tnsm.2021.3110057. Hou, B., Yang, S., Li, F., Zhu, L., Chen, X., Wang, Y., Fu, X., 2024. Nova: Neural-optimized viewport adaptive 360-degree video streaming at the edge. IEEE Transactions on Services Computing 17, 4027–4040. URL: http://dx.doi.org/10.1109/TSC.2024.3451237, doi:10.1109/tsc.2024.3451237. Ickin, S., Fiedler, M., Vandikas, K., 2021. Qoe modeling on split features with distributed deep learning. Network 1, 165–190. URL: http://dx.doi.org/10.3390/network1020011, doi:10.3390/network1020011. Ickin, S., Roeland, D., Hall, G., 2023. Independent split model inference at operator network for network performance estimation, in: NOMS 2023-2023 IEEE/IFIP Network Operations and Management Symposium, IEEE. p. 1–9. URL: http://dx.doi.org/10.1109/NOMS56928. 2023.10154389, doi:10.1109/noms56928.2023.10154389. Ickin, S., Vandikas, K., Moradi, F., Taghia, J., Hu, W., 2020. Ensemble-based synthetic data synthesis for federated qoe modeling, in: 2020 6th IEEE Conference on Network Softwarization (NetSoft), IEEE. p. 72–76. URL: http://dx.doi.org/10.1109/NetSoft48620.2020.9165379, doi:10.1109/netsoft48620.2020.9165379. Islam, M.T., Rothenberg, C.E., Gomes, P.H., 2023. Predicting xr services qoe with ml: Insights from in-band encrypted qos features in 360-vr, in: 2023 IEEE 9th International Conference on Network Softwarization (NetSoft), IEEE. p. 80–88. URL: http://dx.doi.org/10.1109/ NetSoft57336.2023.10175481, doi:10.1109/netsoft57336.2023.10175481. Jain, A.K., Gupta, N., Gupta, B.B., 2025. A survey on scalable consensus algorithms for blockchain technology. Cyber Security and Applications 3, 100065. URL: http://dx.doi.org/10.1016/j.csa.2024.100065, doi:10.1016/j.csa.2024.100065. Javeed, D., Saeed, M.S., Ahmad, I., Adil, M., Kumar, P., Islam, A.N., 2024. Quantum-empowered federated learning and 6g wireless networks for iot security: Concept, challenges and future directions. Future Generation Computer Systems 160, 577–597. URL: http://dx.doi.org/10. 1016/j.future.2024.06.023, doi:10.1016/j.future.2024.06.023. Jayaram, R., Prabakaran, S., 2021. Adaptive cost and energy aware secure peer-to-peer computational offloading in the edge-cloud enabled healthcare system. Peer-to-Peer Networking and Applications 14, 2209–2223. URL: http://dx.doi.org/10.1007/s12083-021-01177-4, doi:10.1007/s12083-021-01177-4. Jia, B., Hao, L., Zhang, C., Huang, B., 2020. A privacy-sensitive service selection method based on artificial fish swarm algorithm in the internet of things. Mobile Networks and Applications 26, 1523–1531. URL: http://dx.doi.org/10.1007/s11036-019-01488-0, doi:10.1007/s11036-019-01488-0. Jia, X., Luo, M., Choo, K.K.R., Li, L., He, D., 2022. A redesigned identity-based anonymous authentication scheme for mobile-edge computing. IEEE Internet of Things Journal 9, 10108–10120. URL: http://dx.doi.org/10.1109/JIOT.2021.3119517, doi:10.1109/jiot.2021. 3119517. Jiang, S., Wu, J., 2022. A blockchain-powered data market for multi-user cooperative search. IEEE Transactions on Network and Service Management 19, 203–215. URL: http://dx.doi.org/10.1109/TNSM.2021.3125604, doi:10.1109/tnsm.2021.3125604. Jin, Y., Zhang, W., Xu, Z., Wang, F., Liu, X., 2024. Privacy-preserving gaze-assisted immersive video streaming. IEEE Transactions on Mobile Computing 23, 15098–15113. URL: http://dx.doi.org/10.1109/TMC.2024.3452510, doi:10.1109/tmc.2024.3452510. Katsigiannis, S., Ramzan, N., 2018. Dreamer: A database for emotion recognition through eeg and ecg signals from wireless low-cost off-the-shelf devices. IEEE Journal of Biomedical and Health Informatics 22, 98–107. URL: http://dx.doi.org/10.1109/JBHI.2017.2688239, doi:10.1109/jbhi.2017.2688239. Kazmi, S.H.A., Qamar, F., Hassan, R., Nisar, K., Al-Betar, M.A., 2024. Security of federated learning in 6g era: A review on conceptual techniques and software platforms used for research and analysis. Computer Networks 245, 110358. URL: http://dx.doi.org/10.1016/j.comnet. 2024.110358, doi:10.1016/j.comnet.2024.110358. Keighrey, C., Flynn, R., Murray, S., Murray, N., 2021. A physiology-based qoe comparison of interactive augmented reality, virtual reality and tablet-based applications. IEEE Transactions on Multimedia 23, 333–341. URL: http://dx.doi.org/10.1109/TMM.2020.2982046, doi:10.1109/tmm.2020.2982046. Khan, M.A., Kumar, N., Alsamhi, S.H., Barb, G., Zywiołek, J., Ullah, I., Noor, F., Shah, J.A., Almuhaideb, A.M., 2025. Security and privacy issues and solutions for uavs in b5g networks: A review. IEEE Transactions on Network and Service Management 22, 892–912. URL: http://dx.doi.org/10.1109/TNSM.2024.3487265, doi:10.1109/tnsm.2024.3487265. Khanal, S., Thar, K., Hossain, M.D., Huh, E.N., 2021. Proactive content caching at self-driving car using federated learning with edge cloud, in: 2021 Twelfth International Conference on Ubiquitous and Future Networks (ICUFN), IEEE. p. 129–134. URL: http://dx.doi.org/10. 1109/icufn49451.2021.9528638, doi:10.1109/icufn49451.2021.9528638. Kokiadis, G., Protopsaltis, A., Morfiadakis, M., Lydatakis, N., Papagiannakis, G., 2024. Decoupled edge physics algorithms for collaborative xr simulations. Computer Animation and Virtual Worlds 35. URL: http://dx.doi.org/10.1002/cav.2294, doi:10.1002/cav.2294. Kougioumtzidis, G., Poulkov, V., Zaharis, Z.D., Lazaridis, P.I., 2022. A survey on multimedia services qoe assessment and machine learning-based prediction. IEEE Access 10, 19507–19538. URL: http://dx.doi.org/10.1109/access.2022.3149592, doi:10.1109/access.2022. 3149592. Kougioumtzidis,G., Vlahov,A.,Poulkov,V.K.,Lazaridis, P.I.,Zaharis,Z.D.,2023. Deeplearning-aidedqoe prediction for virtualrealityapplications over open radio access networks. IEEE Access 11, 143514–143529. URL: http://dx.doi.org/10.1109/ACCESS.2023.3343846, doi:10.1109/access.2023.3343846. Kougioumtzidis, G., Vlahov, A., Poulkov, V.K., Lazaridis, P.I., Zaharis, Z.D., 2024. Qoe prediction for gaming video streaming in o-ran using convolutional neural networks. IEEE Open Journal of the Communications Society 5, 1167–1181. URL: http://dx.doi.org/10.1109/ Garcia et al.: Preprint submitted to Elsevier Page 36 of 35 Privacy-Preserving QoE in Next-Gen Networks OJCOMS.2024.3362275, doi:10.1109/ojcoms.2024.3362275. Laghari, A.A., Zhang, X., Shaikh, Z.A., Khan, A., Estrela, V.V., Izadi, S., 2023. A review on quality of experience (qoe) in cloud computing. Journal of Reliable Intelligent Environments 10, 107–121. URL: http://dx.doi.org/10.1007/s40860-023-00210-y, doi:10.1007/ s40860-023-00210-y. Laiche, F., Ben Letaifa, A., Aguili, T., 2020. Qoe influence factors (ifs) classification survey focusing on user behavior/engagement metrics, in: 2020 IEEE 29th International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE), IEEE. p. 143–146. URL: http://dx.doi.org/10.1109/WETICE49692.2020.00036, doi:10.1109/wetice49692.2020.00036. Le, K.H., Le-Minh, K.H., Thai, H.T., 2023. Brainyedge: An ai-enabled framework for iot edge computing. ICT Express 9, 211–221. URL: http://dx.doi.org/10.1016/j.icte.2021.12.007, doi:10.1016/j.icte.2021.12.007. Li, X., Sun, C., Wen, J., Wang, X., Guizani, M., Leung, V.C., 2022. Multi-user qoe enhancement: Federated multi-agent reinforcement learning for cooperative edge intelligence. IEEE Network 36, 144–151. URL: http://dx.doi.org/10.1109/MNET.001.2200194, doi:10.1109/ mnet.001.2200194. Li, Y., Wan, Z., 2021. Blockchain-enabled intelligent video caching and transcoding in clustered mec networks. Security and Communication Networks 2021, 1–17. URL: http://dx.doi.org/10.1155/2021/7443260, doi:10.1155/2021/7443260. Li, Y., Zhang, W., 2024. Task-offloading strategy of mobile edge computing for wbans. Electronics 13, 1422. URL: http://dx.doi.org/10. 3390/electronics13081422, doi:10.3390/electronics13081422. Liang, Q., Zhao, H., Hou, L., 2022. Federated reinforcement learning-based radio resource scheduling for short video applications, in: 2022 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), IEEE. p. 1–5. URL: http://dx.doi.org/10.1109/ BMSB55706.2022.9828777, doi:10.1109/bmsb55706.2022.9828777. Lin, T.W., 2022. A privacy-preserved id-based secure communication scheme in 5g-iot telemedicine systems. Sensors 22, 6838. URL: http://dx.doi.org/10.3390/s22186838, doi:10.3390/s22186838. Liu, S., Kuwahara, A., Scovell, J.J., Claypool, M., 2023. The effects of frame rate variation on game player quality of experience, in: Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, ACM. p. 1–10. URL: http://dx.doi.org/10.1145/3544548. 3580665, doi:10.1145/3544548.3580665. Liu, Y., Ma, Z., Yan, Z., Wang, Z., Liu, X., Ma, J., 2020a. Privacy-preserving federated k-means for proactive caching in next generation cellular networks. Information Sciences 521, 14–31. URL: http://dx.doi.org/10.1016/j.ins.2020.02.042, doi:10.1016/j.ins.2020.02. 042. Liu, Z., Qiao, B., Fang, K., 2020b. Joint optimization strategy for qoe-aware encrypted video caching and content distributing in multi-edge collaborative computing environment. Journal of Cloud Computing 9. URL: http://dx.doi.org/10.1186/s13677-020-00204-8, doi:10.1186/s13677-020-00204-8. Llorente, A., Perez, J.G., Rodrigo, J.A., Jimenez, D., Menendez, J.M., 2024. Efficient computational cost saving in video processing for qoe estimation. IEEE Access 12, 34846–34862. URL: http://dx.doi.org/10.1109/ACCESS.2024.3373193, doi:10.1109/access.2024. 3373193. Maale, G.T., Sun, G., Kuadey, N.A.E., Kwantwi, T., Ou, R., Liu, G., 2023. Deepfesl: Deep federated echo state learning-based proactive content caching in uav-assisted networks. IEEE Transactions on Vehicular Technology 72, 12208–12220. URL: http://dx.doi.org/10.1109/TVT. 2023.3268541, doi:10.1109/tvt.2023.3268541. Mao, B., Liu, J., Wu, Y., Kato, N., 2023. Security and privacy on 6g network edge: A survey. IEEE Communications Surveys & Tutorials 25, 1095–1127. URL: http://dx.doi.org/10.1109/comst.2023.3244674, doi:10.1109/comst.2023.3244674. Md. Fadlullah, Z., Kato, N., 2022. Hcp: Heterogeneous computing platform for federated learning based collaborative content caching towards 6g networks. IEEE Transactions on Emerging Topics in Computing 10, 112–123. URL: http://dx.doi.org/10.1109/TETC.2020.2986238, doi:10.1109/tetc.2020.2986238. Mišić, J., Mišić, V.B., Chang, X., 2025. Qpos: Decentralized stake-based leader and voter selection in a pbft system with mobile voters. IEEE Transactions on Network Science and Engineering 12, 653–668. URL: http://dx.doi.org/10.1109/TNSE.2024.3507545, doi:10.1109/ tnse.2024.3507545. Nasir, M.H., Arshad, J., Khan, M.M., Fatima, M., Salah, K., Jayaraman, R., 2022. Scalable blockchains — a systematic review. Future Generation Computer Systems 126, 136–162. URL: http://dx.doi.org/10.1016/j.future.2021.07.035, doi:10.1016/j.future.2021.07. 035. Nasralla, M.M., Khattak, S.B.A., Ur Rehman, I., Iqbal, M., 2023. Exploring the role of 6g technology in enhancing quality of experience for m-health multimedia applications: A comprehensive survey. Sensors 23, 5882. URL: http://dx.doi.org/10.3390/s23135882, doi:10.3390/s23135882. Nguyen, M.D., Tong, V., Souihi, S., Mellouk, A., 2023. Fully-decentralized federated learning for qoe estimation, in: GLOBECOM 2023 - 2023 IEEE Global Communications Conference, IEEE. p. 5859–5864. URL: http://dx.doi.org/10.1109/GLOBECOM54140.2023.10437914, doi:10.1109/globecom54140.2023.10437914. Nguyen, V.L., Lin, P.C., Cheng, B.C., Hwang, R.H., Lin, Y.D., 2021. Security and privacy for 6g: A survey on prospective technologies and challenges. IEEE Communications Surveys & Tutorials 23, 2384–2428. URL: http://dx.doi.org/10.1109/COMST.2021.3108618, doi:10.1109/comst.2021.3108618. Nyangaresi, V.O., 2022. Terminal independent security token derivation scheme for ultra-dense iot networks. Array 15, 100210. URL: http://dx.doi.org/10.1016/j.array.2022.100210, doi:10.1016/j.array.2022.100210. Orsolic, I., Skorin-Kapov, L., 2020. A framework for in-network qoe monitoring of encrypted video streaming. IEEE Access 8, 74691–74706. URL: http://dx.doi.org/10.1109/ACCESS.2020.2988735, doi:10.1109/access.2020.2988735. Ouadrhiri, A.E., Abdelhadi, A., 2022. Differential privacy for deep and federated learning: A survey. IEEE Access 10, 22359–22380. URL: http://dx.doi.org/10.1109/ACCESS.2022.3151670, doi:10.1109/access.2022.3151670. Garcia et al.: Preprint submitted to Elsevier Page 37 of 35 Privacy-Preserving QoE in Next-Gen Networks Ouardi, K.E., Rafiei, S., Singhal, C., Brunnström, K., 2024. User performance and quality of experience for a remote-controlled lab based moving platform. Electronic Imaging 36, 237–1–237–7. URL: http://dx.doi.org/10.2352/EI.2024.36.11.HVEI-237, doi:10.2352/ei. 2024.36.11.hvei-237. Panahi, P.H.S., Jalilvand, A.H., Diyanat, A., 2024. Enhancing quality of experience in telecommunication networks: A review of frameworks and machine learning algorithms. URL: https://arxiv.org/abs/2404.16787, doi:10.48550/ARXIV.2404.16787. Parvez, A., Rao, S., Khan, A., 2023. Physiological responsiveness to vr and non-vr environments, in: 2023 International Conference on Computational Science and Computational Intelligence (CSCI), IEEE. p. 533–539. URL: http://dx.doi.org/10.1109/CSCI62032.2023. 00095, doi:10.1109/csci62032.2023.00095. Porcu, S., Floris, A., Atzori, L., 2022. Cb-fl: Cluster-based federated learning applied to quality of experience modelling, in: 2022 16th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), IEEE. p. 585–591. URL: http://dx.doi.org/10.1109/ SITIS57111.2022.00093, doi:10.1109/sitis57111.2022.00093. Rachakonda, L.P., Siddula, M., Sathya, V., 2024. A comprehensive study on iot privacy and security challenges with focus on spectrum sharing in next-generation networks (5g/6g/beyond). High-Confidence Computing 4, 100220. URL: http://dx.doi.org/10.1016/j.hcc.2024. 100220, doi:10.1016/j.hcc.2024.100220. Rao, B., Zhang, J., Wu, D., Zhu, C., Sun, X., Chen, B., 2025. Privacy inference attack and defense in centralized and federated learning: A comprehensive survey. IEEE Transactions on Artificial Intelligence 6, 333–353. URL: http://dx.doi.org/10.1109/TAI.2024.3363670, doi:10.1109/tai.2024.3363670. Rao, I.S., Kiah, M.L.M., Hameed, M.M., Memon, Z.A., 2024. Scalability of blockchain: a comprehensive review and future research direction. Cluster Computing 27, 5547–5570. URL: http://dx.doi.org/10.1007/s10586-023-04257-7, doi:10.1007/s10586-023-04257-7. Rao, S.P., Chen, H.Y., Aura, T., 2023. Threat modeling framework for mobile communication systems. Computers & Security 125, 103047. URL: http://dx.doi.org/10.1016/j.cose.2022.103047, doi:10.1016/j.cose.2022.103047. Renda, A., Ducange, P., Marcelloni, F., Sabella, D., Filippou, M.C., Nardini, G., Stea, G., Virdis, A., Micheli, D., Rapone, D., Baltar, L.G., 2022. Federated learning of explainable ai models in 6g systems: Towards secure and automated vehicle networking. Information 13, 395. URL: http://dx.doi.org/10.3390/info13080395, doi:10.3390/info13080395. Robitza, W., Dethof, A.M., Goring, S., Raake, A., Beyer, A., Polzehl, T., 2020. Are you still watching? streaming video quality and engagement assessment in the crowd, in: 2020 Twelfth International Conference on Quality of Multimedia Experience (QoMEX), IEEE. p. 1–6. URL: http://dx.doi.org/10.1109/QoMEX48832.2020.9123148, doi:10.1109/qomex48832.2020.9123148. Rossi, H.S., Mitra, K., Larsson, S., Åhlund, C., Cotanis, I., 2024. Subjective qoe assessment for virtual reality cloud-based first-person shooter game, in: ICC 2024 - IEEE International Conference on Communications, IEEE. p. 4698–4703. URL: http://dx.doi.org/10.1109/ICC51166. 2024.10622467, doi:10.1109/icc51166.2024.10622467. Salgado, D.P., Martins, F.R., Rodrigues, T.B., Keighrey, C., Flynn, R., Naves, E.L.M., Murray, N., 2018. A qoe assessment method based on eda, heart rate and eeg of a virtual reality assistive technology system, in: Proceedings of the 9th ACM Multimedia Systems Conference, ACM. p. 517–520. URL: http://dx.doi.org/10.1145/3204949.3208118, doi:10.1145/3204949.3208118. Sandeepa, C., Siniarski, B., Kourtellis, N., Wang, S., Liyanage, M., 2024. A survey on privacy of personal and non-personal data in b5g/6g networks. ACM Computing Surveys 56, 1–37. URL: http://dx.doi.org/10.1145/3662179, doi:10.1145/3662179. Sanyour, R., Abdullah, M., Abdullah, S., 2022. Quality of experience in internet of things: A systematic literature review. Journal on Internet of Things 4, 263–282. URL: http://dx.doi.org/10.32604/jiot.2022.040966, doi:10.32604/jiot.2022.040966. Setayesh, M., Wong, V.W.S., 2025. Viewport prediction, bitrate selection, and beamforming design for thz-enabled 360°video streaming. IEEE Transactions on Wireless Communications 24, 1849–1865. URL: http://dx.doi.org/10.1109/TWC.2024.3513221, doi:10.1109/twc. 2024.3513221. Shan,N.,Cui,X., Gao,Z., 2020. “drl+fl”: Anintelligentresourceallocation model based on deepreinforcementlearningformobile edgecomputing. Computer Communications 160, 14–24. URL: http://dx.doi.org/10.1016/j.comcom.2020.05.037, doi:10.1016/j.comcom.2020. 05.037. da Silveira, A.C., Lima de Souza, M., Ghinea, G., Saibel Santos, C.A., 2024. Physiological data for user experience and quality of experience: A systematic review (2018–2022). International Journal of Human–Computer Interaction 41, 664–693. URL: http://dx.doi.org/10.1080/ 10447318.2024.2311972, doi:10.1080/10447318.2024.2311972. Sirohi, D., Kumar, N., Rana, P.S., Tanwar, S., Iqbal, R., Hijjii, M., 2023. Federated learning for 6g-enabled secure communication systems: a comprehensive survey. Artificial Intelligence Review 56, 11297–11389. URL: http://dx.doi.org/10.1007/s10462-023-10417-3, doi:10.1007/s10462-023-10417-3. Skaka - Čekić, F., Baraković Husić, J., 2023. A feature selection for video quality of experience modeling: A systematic literature review. WIREs Data Mining and Knowledge Discovery 13. URL: http://dx.doi.org/10.1002/widm.1497, doi:10.1002/widm.1497. Skowronek, J., Raake, A., Berndtsson, G.H., Rummukainen, O.S., Usai, P., Gunkel, S.N.B., Johanson, M., Habets, E.A.P., Malfait, L., Lindero, D., Toet, A., 2022. Quality of experience in telemeetings and videoconferencing: A comprehensive survey. IEEE Access 10, 63885–63931. URL: http://dx.doi.org/10.1109/access.2022.3176369, doi:10.1109/access.2022.3176369. Sun, Y., Liu, J., Wang, J., Cao, Y., Kato, N., 2020. When machine learning meets privacy in 6g: A survey. IEEE Communications Surveys & Tutorials 22, 2694–2724. URL: http://dx.doi.org/10.1109/COMST.2020.3011561, doi:10.1109/comst.2020.3011561. Tang, Z., Feng, X., Xie, Y., Phan, H., Guo, T., Yuan, B., Wei, S., 2020. Vvsec: Securing volumetric video streaming via benign use of adversarial perturbation, in: Proceedings of the 28th ACM International Conference on Multimedia, ACM. p. 3614–3623. URL: http: //dx.doi.org/10.1145/3394171.3413639, doi:10.1145/3394171.3413639. Tao, X., Liu, Y., Jiang, C., Wang, Z., Qin, X., 2019. Qoe-oriented multimedia assessment: A facial expression recognition approach. IEEE MultiMedia 26, 41–50. URL: http://dx.doi.org/10.1109/MMUL.2018.2879596, doi:10.1109/mmul.2018.2879596. Tondwalkar, A., Andres-Maldonado, P., Chandramouli, D., Liebhart, R., Sanchez Moya, F., Kolding, T., Perez, P., 2024. Provisioning quality of experience in 6g networks. IEEE Access 12, 127007–127017. URL: http://dx.doi.org/10.1109/ACCESS.2024.3455938, doi:10.1109/ Garcia et al.: Preprint submitted to Elsevier Page 38 of 35 Privacy-Preserving QoE in Next-Gen Networks access.2024.3455938. Tran, N.P., Truong, T.P., Do, Q.T., Dao, N.N., Cho, S., 2024. Joint wireless resource allocation and bitrate adaptation for qoe improvement in irs-aided rsma-enabled iomt streaming systems. Internet of Things 25, 101145. URL: http://dx.doi.org/10.1016/j.iot.2024.101145, doi:10.1016/j.iot.2024.101145. Tripathy, R., Meshram, J., Bera, P., 2025. Halffedlearn: A secure federated learning with local data partitioning and homomorphic encryption. Future Generation Computer Systems 171, 107858. URL: http://dx.doi.org/10.1016/j.future.2025.107858, doi:10.1016/j.future. 2025.107858. Valente, R., Senna, C., Rito, P., Sargento, S., 2023. Embedded federated learning for vanet environments. Applied Sciences 13, 2329. URL: http://dx.doi.org/10.3390/app13042329, doi:10.3390/app13042329. Vijayakumar, S., Flynn, R., Corcoran, P., Murray, N., 2024. The role of ecg and respiration in predicting quality of experience, in: 2024 16th International Conference on Quality of Multimedia Experience (QoMEX), IEEE. p. 118–124. URL: http://dx.doi.org/10.1109/ QoMEX61742.2024.10598275, doi:10.1109/qomex61742.2024.10598275. Vlahovic, S., Suznjevic, M., Skorin-Kapov, L., 2022. A survey of challenges and methods for quality of experience assessment of interactive vr applications. Journal on Multimodal User Interfaces 16, 257–291. URL: http://dx.doi.org/10.1007/s12193-022-00388-0, doi:10.1007/s12193-022-00388-0. Wagner, I., Eckhoff, D., 2018. Technical privacy metrics: A systematic survey. ACM Computing Surveys 51, 1–38. URL: http://dx.doi.org/ 10.1145/3168389, doi:10.1145/3168389. Wang, C.H., Huang, K.Y., Yao, Y., Chen, J.C., Shuai, H.H., Cheng, W.H., 2024a. Lightweight deep learning: An overview. IEEE Consumer Electronics Magazine 13, 51–64. URL: http://dx.doi.org/10.1109/MCE.2022.3181759, doi:10.1109/mce.2022.3181759. Wang, S., Yang, J., Bi, S., 2024b. Adaptive video streaming in multi-tier computing networks: Joint edge transcoding and client enhancement. IEEE Transactions on Mobile Computing 23, 2657–2670. URL: http://dx.doi.org/10.1109/TMC.2023.3263046, doi:10.1109/tmc.2023. 3263046. Wang, Y., Li, J., Li, Z., Shang, S., Liu, Y., 2024c. Synergistic temporal-spatial user-aware viewport prediction for optimal adaptive 360degree video streaming. IEEE Transactions on Broadcasting 70, 453–467. URL: http://dx.doi.org/10.1109/TBC.2024.3374119, doi:10.1109/tbc.2024.3374119. Wassermann, S., Casas, P., Houidi, Z.B., Huet, A., Seufert, M., Wehner, N., Schuler, J., Cai, S., Shi, H., Xu, J., Hossfeld, T., Rossi, D., 2020a. Are you on mobile or desktop? on the impact of end-user device on web qoe inference from encrypted traffic, in: 2020 16th International Conference on Network and Service Management (CNSM), IEEE. p. 1–9. URL: http://dx.doi.org/10.23919/CNSM50824.2020.9269095, doi:10. 23919/cnsm50824.2020.9269095. Wassermann, S., Casas, P., Houidi, Z.B., Huet, A., Seufert, M., Wehner, N., Schuler, J., Cai, S., Shi, H., Xu, J., Hossfeld, T., Rossi, D., 2020b. Are you on mobile or desktop? on the impact of end-user device on web qoe inference from encrypted traffic, in: 2020 16th International Conference on Network and Service Management (CNSM), IEEE. p. 1–9. URL: http://dx.doi.org/10.23919/cnsm50824.2020.9269095, doi:10. 23919/cnsm50824.2020.9269095. Wassermann, S., Seufert, M., Casas, P., Gang, L., Li, K., 2020c. Vicrypt to the rescue: Real-time, machine-learning-driven video-qoe monitoring for encrypted streaming traffic. IEEE Transactions on Network and Service Management 17, 2007–2023. URL: http://dx.doi.org/10. 1109/TNSM.2020.3036497, doi:10.1109/tnsm.2020.3036497. Wehner, N., Seufert, M., Schuler, J., Casas, P., Hossfeld, T., 2021. How are your apps doing? qoe inference and analysis in mobile devices, in: 2021 17th International Conference on Network and Service Management (CNSM), IEEE. p. 49–55. URL: http://dx.doi.org/10.23919/ CNSM52442.2021.9615549, doi:10.23919/cnsm52442.2021.9615549. Wei, X., Hellge, C., Yang, C., Son, J., 2024. Improving qoe-privacy tradeoff in xr streaming. IEEE Signal Processing Letters 31, 1504–1508. URL: http://dx.doi.org/10.1109/LSP.2024.3401616, doi:10.1109/lsp.2024.3401616. Wei, X., Yang, C., 2022. Fov privacy-aware vr streaming, in: 2022 IEEE Wireless Communications and Networking Conference (WCNC), IEEE. p. 1515–1520. URL: http://dx.doi.org/10.1109/WCNC51071.2022.9771832, doi:10.1109/wcnc51071.2022.9771832. Wei, Y.S., Wei, X., Zheng, S.Y., Hsu, C.H., Yang, C., 2023. A 6dof vr dataset of 3d virtualworld for privacy-preserving approach and utilityprivacy tradeoff, in: Proceedings of the 14th ACM Multimedia Systems Conference, ACM. p. 444–450. URL: http://dx.doi.org/10. 1145/3587819.3592557, doi:10.1145/3587819.3592557. Wu, M., Xiao, Y., Gao, Y., Lei, X., 2022. Design of quality-of-experience criteria for resource allocation toward 6g wireless networks: A review and new directions, in: 2022 IEEE 96th Vehicular Technology Conference (VTC2022-Fall), IEEE. p. 1–7. URL: http://dx.doi.org/10. 1109/VTC2022-Fall57202.2022.10013024, doi:10.1109/vtc2022-fall57202.2022.10013024. Xu, X., Huang, Q., Zhu, H., Sharma, S., Zhang, X., Qi, L., Bhuiyan, M.Z.A., 2021. Secure service offloading for internet of vehicles in sdn-enabled mobile edge computing. IEEE Transactions on Intelligent Transportation Systems 22, 3720–3729. URL: http://dx.doi.org/10.1109/ TITS.2020.3034197, doi:10.1109/tits.2020.3034197. Xu, Y., Li, X., Yang, Y., Lin, Z., Wang, L., Li, W., 2023. Fedabr: A personalized federated reinforcement learning approach for adaptive video streaming,in:2023 IFIPNetworkingConference (IFIPNetworking),IEEE.p. 1–9. URL:http://dx.doi.org/10.23919/IFIPNetworking57963. 2023.10186404, doi:10.23919/ifipnetworking57963.2023.10186404. Yan, B., Yu, J., Yang, M., Jiang, H., Wan, Z., Ni, L., 2019. A novel distributed social internet of things service recommendation scheme based on lsh forest. Personal and Ubiquitous Computing 25, 1013–1026. URL: http://dx.doi.org/10.1007/s00779-019-01283-4, doi:10.1007/s00779-019-01283-4. Yeznabad, Y.F., Helfert, M., Muntean, G.M., 2024. Adaptive bitrate allocation in mec-enabled networks: A collaborative approach to enhance user qoe, in: 2024 IEEE Wireless Communications and Networking Conference (WCNC), IEEE. p. 1–6. URL: http://dx.doi.org/10.1109/ WCNC57260.2024.10571237, doi:10.1109/wcnc57260.2024.10571237. Zeng, J., Zhou, X., Li, K., 2024. Toward high-quality low-latency 360°video streaming with edge–client collaborative caching and super-resolution. IEEE Internet of Things Journal 11, 29020–29034. URL: http://dx.doi.org/10.1109/JIOT.2024.3405493, doi:10.1109/jiot.2024. Garcia et al.: Preprint submitted to Elsevier Page 39 of 35 Privacy-Preserving QoE in Next-Gen Networks 3405493. Zhang, C., Xie, Y., Bai, H., Yu, B., Li, W., Gao, Y., 2021. A survey on federated learning. Knowledge-Based Systems 216, 106775. URL: http://dx.doi.org/10.1016/j.knosys.2021.106775, doi:10.1016/j.knosys.2021.106775. Zhang, H., Dong, L., Gao, G., Hu, H., Wen, Y., Guan, K., 2020. Deepqoe: A multimodal learning framework for video quality of experience (qoe) prediction. IEEE Transactions on Multimedia 22, 3210–3223. URL: http://dx.doi.org/10.1109/tmm.2020.2973828, doi:10.1109/ tmm.2020.2973828. Zhang, Y., Su, Y., Sun, X., 2024. A qoe physiological measure of vr with vibrotactile feedback based on frontal lobe power asymmetry. IEEE Transactions on Multimedia 26, 2932–2942. URL: http://dx.doi.org/10.1109/TMM.2023.3305813, doi:10.1109/tmm.2023. 3305813. Zhang, Z., Cao, T., Wang, X., Xiao, H., Guan, J., 2022. Vc-ppq: Privacy-preserving q-learning based video caching optimization in mobile edge networks. IEEE Transactions on Network Science and Engineering 9, 4129–4144. URL: http://dx.doi.org/10.1109/TNSE.2022. 3195926, doi:10.1109/tnse.2022.3195926. Zhao, P., Li, C., Fu, Y., Hui, Y., Zhang, Y., Cheng, N., 2022. Blockchain-enabled conditional decentralized vehicular crowdsensing system. IEEE Transactions on Intelligent Transportation Systems 23, 18937–18950. URL: http://dx.doi.org/10.1109/TITS.2022.3166216, doi:10.1109/tits.2022.3166216. Zhao, T., Li, F., He, L., 2024. Secure video offloading in mec-enabled iiot networks: A multicell federated deep reinforcement learning approach. IEEE Transactions on Industrial Informatics 20, 1618–1629. URL: http://dx.doi.org/10.1109/TII.2023.3280314, doi:10.1109/tii. 2023.3280314. Zhong, Y., Gu, J., 2024. Lightweight block ciphers for resource-constrained environments: A comprehensive survey. Future Generation Computer Systems 157, 288–302. URL: http://dx.doi.org/10.1016/j.future.2024.03.054, doi:10.1016/j.future.2024.03.054. Zhong, Y., Li, Z., Liao, L., 2021. A privacy-preserving caching scheme for device-to-device communications. Security and Communication Networks 2021, 1–8. URL: http://dx.doi.org/10.1155/2021/6696149, doi:10.1155/2021/6696149. Zhou, E., Turcotte, J., De Carli, L., 2020. Enabling security analysis of iot device-to-cloud traffic, in: 2020 IEEE 19th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom), IEEE. p. 1888–1894. URL: http://dx.doi.org/10.1109/ TrustCom50675.2020.00258, doi:10.1109/trustcom50675.2020.00258. Zhou, L., Wu, D., Wei, X., Dong, Z., 2019. Seeing isn’t believing: Qoe evaluation for privacy-aware users. IEEE Journal on Selected Areas in Communications 37, 1656–1665. URL: http://dx.doi.org/10.1109/JSAC.2019.2916452, doi:10.1109/jsac.2019.2916452. Zuo, Y., Guo, J., Gao, N., Zhu, Y., Jin, S., Li, X., 2023. A survey of blockchain and artificial intelligence for 6g wireless communications. IEEE Communications Surveys & Tutorials 25, 2494–2528. URL: http://dx.doi.org/10.1109/COMST.2023.3315374, doi:10.1109/comst. 2023.3315374. Garcia et al.: Preprint submitted to Elsevier Page 40 of 35