Traditional network engineering operates on a simple premise: more is better. If a system can deliver higher throughput, lower packet loss, or greater spectral efficiency, it is assumed to be superior. However, this device-centric approach overlooks a critical reality: digital networks ultimately serve human users, and human perception does not scale linearly with raw technical metrics.
A network might achieve massive aggregate bandwidth, yet still frustrate a user if a video stalls during its first few seconds or if an interactive application lags. This discrepancy is known as the capacity-experience gap. To bridge it, a growing body of research advocates for a paradigm shift from objective Quality of Service (QoS) to subjective Quality of Experience (QoE)—directly embedding human sensory and cognitive limits into the network optimization loop.
The Non-Linearity of Perceived Quality
Why do traditional engineering metrics fail to guarantee a good user experience? The answer lies in the non-linear relationship between physical resources and human satisfaction.
Consider bandwidth. Beyond a certain application-specific threshold, providing additional bitrate yields diminishing returns. As Jingjing Cui et al. demonstrate in QoE-Based Resource Allocation for Multi-Cell NOMA Networks, maximizing raw data throughput across a network often leads to an unfair and inefficient distribution of resources. A user streaming a high-definition video requires strict performance guarantees, while a user downloading a background file gains little immediate satisfaction from a temporary spike in speed.
When systems chase high technical benchmarks blindly, they waste valuable resources. Geoff V. Merrett, in his work on Managing power in heterogeneous multicore systems, points out that computing and communication systems frequently expend significant energy to achieve throughput gains that are entirely imperceptible to the end user.
To quantify this mathematically, researchers use utility functions that map physical resources to user "happiness." For example, Huy A. Nguyen et al. developed a bandwidth allocation scheme in How to Maximize User Satisfaction Degree in Multi-service IP Networks that replaces raw throughput maximization with a metric called "Network Worth." By weighting application utility based on user priority and application type, their system ensures that scarce bandwidth is directed where it will generate the highest marginal increase in human satisfaction.
[Raw Network Metrics] ---> (Non-Linear Mapping/MOS) ---> [Perceived QoE]
(Throughput, Loss) (Diminishing Returns) (User Satisfaction)
Mapping Senses to Network Knobs
To optimize a network for human perception, designers must first translate subjective human experiences into concrete, actionable engineering constraints. This requires a deep understanding of human cognitive and sensory thresholds.
In Perception-Based Media Processing [Scanning the Issue], Lina J. Karam et al. argue for integrating how we see, hear, and feel directly into algorithmic design. For highly immersive applications like wireless virtual reality (VR), this mapping must be incredibly precise. As Fenghe Hu et al. lay out in Cellular-Connected Wireless Virtual Reality: Requirements, Challenges, and Solutions, human sensory thresholds—such as the number of pixels per degree of visual field and the refresh rates required to prevent motion sickness—can be mathematically translated into strict network parameters like minimum throughput and maximum round-trip time (RTT).
This user-centric framing applies beyond media consumption to complex digital ecosystems and service retrieval. Hai Dong et al. show in A QoS-based service retrieval methodology for digital ecosystems that combining subjective user perceptions with objective domain-specific metrics yields far more accurate service rankings than technical data alone. Ultimately, whether evaluating a web service or a network protocol, the system's effectiveness is determined by the user's ability to complete tasks without perceived disruption, a principle emphasized by Xianglin Wei in Online System QoS Enhancement of Flipped Classroom for Intelligent Manufacturing Specialty Based on Intelligent Information Platform.
Closing the Optimization Loop
Once human perception is quantified, the next challenge is embedding these metrics directly into the network's control loops and resource allocation algorithms. Traditional physical-layer optimization relies heavily on Channel State Information (CSI), but optimizing for the channel state alone can degrade user experience if it ignores network-layer dynamics.
For instance, Jian Li 0025 et al. demonstrate in Queue-Aware Energy-Efficient Joint Remote Radio Head Activation and Beamforming in Cloud Radio Access Networks that failing to account for operational parameters like queuing delay leads to poor user-perceived performance, even when the physical link looks optimal. To resolve this, researchers use cross-layer optimization frameworks. Yuanwei Liu et al. propose a cross-layer design in Toward Cross-Layer Design for Non-Orthogonal Multiple Access: A Quality-of-Experience Perspective that maps high-level application demands directly down to physical-layer resource allocation, dynamically scheduling traffic to eliminate application-level glitches like video stalls.
+--------------------------------------------------+
| Application Layer |
| Perceived QoE / Mean Opinion Score (MOS) Target |
+------------------------+-------------------------+
| (Dynamic Mapping)
v
+--------------------------------------------------+
| Physical Layer |
| Cross-Layer Resource Allocation & Beamforming |
+--------------------------------------------------+
This optimization can also be achieved using data-driven approaches. In the ACM MMSys 2024 Bandwidth Estimation in Real Time Communications Challenge, Sami Khairy et al. highlight how maximizing the mathematical accuracy of link capacity predictions does not inherently improve real-time communication. By training offline Reinforcement Learning (RL) models on massive real-world datasets and using objective audio/video Mean Opinion Scores (MOS) as the reward function, they successfully aligned real-time bandwidth estimation with actual human satisfaction.
Targeting Application-Specific Bottlenecks
Because different applications trigger different human cognitive responses, user-centric network optimization must target the specific bottlenecks that drive human dissatisfaction.
- Video Streaming: In video delivery, the primary culprits of a poor experience are initial buffering and mid-stream stalling. Qingyong Wang et al. analyze this in Data analysis on video streaming QoE over mobile networks, proving that reducing the time-to-first-frame is far more critical to user retention than maintaining an unnecessarily high average bitrate. To combat this, Saeed Ullah et al. design a cache management scheme in Management of scalable video streaming in information centric networking that strategically places essential video "base layers" close to the user to eliminate stalling, while pushing optional enhancement layers further back into the network.
- Virtual Desktops: In interactive environments, system-level throughput metrics often mask severe usability issues. Peiyu Liu et al. point out in Prudent Practices for Designing Virtual Desktop Experiments that virtual desktop infrastructure evaluations frequently focus on aggregate data rates while ignoring highly perceptible factors like local frame rates and interactive response times.
- Network Operations: Even network monitoring can benefit from this prioritization. Instead of reacting to an overwhelming "alarm storm" of isolated hardware events, Runhua Zhang demonstrates in Research on Management of Mobile Communication Network Centralized Alarm that filtering and prioritizing network alarms based on user-impacting performance metrics (like call success rates) allows operators to resolve critical user-facing issues much faster.
The Paradigm Shift
Designing systems around human perception requires a holistic view of service success. As Jie Lu et al. argued in their foundational analysis, Cost and Benefit Analysis for E-Service Applications, evaluating digital services purely from the provider's technical perspective creates a one-sided, flawed view of value; true success requires integrating external consumer feedback and satisfaction.
Looking forward, this philosophy is becoming central to the architecture of next-generation networks. As Shuping Dang et al. note in Pre-6G Graduate Education of Communications Engineering, the upcoming 6G era will explicitly move away from raw throughput as the primary key performance indicator, focusing instead on human-centric metrics and psychological impact.
However, fully realizing this vision remains an active challenge. For instance, Joongheon Kim et al. observe in Joint Mobile Charging and Coverage-Time Extension for Unmanned Aerial Vehicles that while we can optimize physical infrastructure—like drone base stations—for energy and queue stability, seamlessly integrating dynamic human QoE metrics directly into these automated control loops remains a vital frontier for modern network research.
Key Foundational Work
A QoS-based service retrieval methodology for digital ecosystems (2009) — contributed 1 concept elements to this direction.
Cost and Benefit Analysis for E-Service Applications (2002) — contributed 1 concept elements to this direction.