Balancing Efficiency and Fairness in Wireless Resource Allocation

781 concepts531 papers6 proposals

In a perfect wireless network, every user would enjoy lightning-fast data rates, zero latency, and minimal power consumption. In reality, wireless channels are highly unequal. A user standing directly beneath a cellular tower experiences a pristine, high-capacity signal, while a user at the edge of the cell struggles against severe path loss, shadowing, and inter-cell interference.

This physical heterogeneity creates a fundamental tension in network design: the throughput-fairness tradeoff. If a network operator aims purely to maximize aggregate system efficiency (such as total sum-rate or global energy efficiency), the optimal strategy is inherently "greedy." The system will allocate the vast majority of its resources—such as transmit power, frequency subcarriers, and antenna beams—to the users with the best channel conditions, completely starving users at the periphery. Conversely, if the operator demands strict uniformity across all users, the entire network's performance collapses to the level of the least capable device.

Understanding why this tradeoff occurs, how it manifests across different network architectures, and how modern optimization frameworks navigate it is essential for designing next-generation wireless systems.


The Physics of the Greedy Allocation Trap

To understand why efficiency and fairness conflict, we must look at the mathematical structure of wireless capacity. The data rate $R$ achievable by a user is a logarithmic function of their Signal-to-Interference-plus-Noise Ratio (SINR):

$$R = B \log_2(1 + \text{SINR})$$

Because of distance-dependent path loss, a cell-edge user requires exponentially more transmit power to achieve the same SINR as a cell-center user. As a result, if an algorithm is programmed to maximize the total network sum-rate ($\sum R_i$), it encounters a much higher "return on investment" by giving power to close users.

As Jang et al. observe, pure sum-rate maximization inevitably triggers a "winner-takes-all" scenario. Resources concentrate heavily on a privileged subset of users, leaving disadvantaged users with near-zero data rates. This phenomenon is not unique to throughput; it replicates across multiple performance metrics:

  • Energy Efficiency (EE): One might assume that making a network "green" would naturally distribute resources evenly. However, Zappone et al. and He et al. demonstrate that maximizing the aggregate bits-per-joule ratio causes the exact same starvation effect. Because serving a cell-edge user requires a disproportionately large amount of power for a meager handful of bits, an energy-efficient objective function will simply shut off service to the edge to preserve the global power budget.
  • Mobile Edge Computing (MEC): When optimizing computational offloading, minimizing the total energy consumption or task delay across a multi-user system creates a biased allocation. As Zhang et al. and Sardellitti et al. show, systems favor users with strong wireless links for cloud offloading, forcing users with poor connectivity to process tasks locally, draining their local batteries and causing massive task backlogs.

The Cost of Absolute Fairness

The intuitive countermeasure to user starvation is max-min fairness, an objective that seeks to maximize the performance of the worst-off user in the network ($\max \min R_i$). While socially equitable, strict max-min fairness introduces two severe penalties: computational complexity and capacity collapse.

First, enforcing absolute fairness is incredibly difficult. Razaviyayn et al. proved that achieving max-min fairness in multi-antenna (MIMO) interfering broadcast channels is NP-hard, even when devices have as few as two antennas. The coupled nature of multi-user interference means that adjusting the beamforming vector to help one starving user unpredictably degrades the signals of multiple neighbors.

Second, strict fairness imposes a heavy performance tax on the rest of the network. As Proutiere et al. point out, trying to guarantee uniform data rates regardless of location drastically reduces the overall efficiency of the base station. The system spends an excessive amount of time and power transmitting low-rate data to distant users, dragging the entire network down to the lowest common denominator.

Sum-Rate Maximization (Greedy)
[Base Station] ===(High Power/Rate)===> [Close User]
               ---(No Power/Starved)--> [Edge User]
Result: High total throughput, zero fairness.

Strict Max-Min Fairness
[Base Station] ---(Low Power/Rate)----> [Close User]
               ---(Low Power/Rate)----> [Edge User]
Result: Low total throughput, perfect uniformity.

Engineering the Compromise: Algorithmic and Structural Solutions

To escape this binary choice between a greedy system and a crippled one, researchers have developed mechanisms that smoothly bridge the gap between efficiency and fairness.

1. Soft Max-Min and Proportional Fairness

Instead of forcing strict equality, modern frameworks use alternative utility functions. Proportional fairness maximizes the sum of the logarithms of user rates ($\sum \ln R_i$), which naturally penalizes starvation without capping the throughput of high-performing users.

Alternatively, Zhu et al. propose a "soft max-min" optimization strategy for hybrid beamforming in high-frequency bands. By using differentiable mathematical approximations of the minimum rate, their algorithm ensures that cell-edge users meet a viable quality-of-service threshold while still allowing the system to exploit the massive capacity potential of users with excellent channel conditions.

2. Precoding Normalization in Massive MIMO

In multi-user MIMO systems, the way a base station normalizes its transmit power dictates where it sits on the throughput-fairness spectrum. A Large System Analysis of Power Normalization Techniques in Massive MIMO highlights two distinct strategies:

  • Vector Normalization: Normalizing precoding vectors individually maximizes total system throughput by prioritizing users with stronger channel conditions.
  • Matrix Normalization: Scaling the entire precoding matrix as a single unit inherently promotes user fairness, equalizing the quality of experience across the network by preventing strong channels from overpowering weak ones.

3. Power-Domain Non-Orthogonal Multiple Access (NOMA)

Traditional networks separate users by time or frequency to avoid interference. Non-Orthogonal Multiple Access (NOMA) intentionally serves multiple users on the same frequency resource simultaneously, leveraging the throughput-fairness conflict to its advantage.

As Ding et al. explain, NOMA pairs a strong-channel user with a weak-channel user and intentionally allocates a higher power level to the user with the poorer channel. The weak user treats the strong user's signal as noise and decodes their own data directly. Meanwhile, the strong user utilizes Successive Interference Cancellation (SIC) to decode and subtract the weak user's high-power signal, allowing them to decode their own data cleanly. This structural design balances system-wide throughput with individual user fairness by design rather than through secondary optimization.

4. Data-Driven Primal-Dual Learning

As networks grow denser, calculating these trade-offs in real time using traditional optimization becomes intractable. To solve this, Naderializadeh et al. introduced a framework that combines Graph Neural Networks (GNNs) with counterfactual optimization.

By using a primal-dual learning approach, the network dynamically tunes minimum rate constraints through "slack variables." The GNN learns the underlying topology of the network and automatically shifts power to edge users when starvation risks are high, protecting fairness without requiring the heavy computational overhead of traditional iterative algorithms.


Conclusion

The throughput-fairness tradeoff is an inescapable consequence of wireless physics. Because a greedy system maximizes aggregate metrics by abandoning disadvantaged nodes, and a strictly fair system destroys capacity by catering entirely to them, modern wireless design focuses on controlled compromises. Through mathematical reformulations like soft max-min utility, structural innovations like NOMA, and adaptive machine learning, next-generation networks ensure that efficiency does not come at the cost of exclusion.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 509 papers
  2. Version 1 · 2026-10-04 · 538 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_throughput_fairness_tradeoff_7fc10eff,
  title = {throughput-fairness tradeoff},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/throughput-fairness-tradeoff/rdr_7fc10eff9ed8438bb78c2226c8538943},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 2, frozen text):

@misc{lacuna_throughput_fairness_tradeoff_7fc10eff_v2,
  title = {throughput-fairness tradeoff},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/throughput-fairness-tradeoff/rdr_7fc10eff9ed8438bb78c2226c8538943/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake