Multiobjective Signal Processing Optimization: The way to balance conflicting metrics in 5G systems

Emil Björnson, Eduard Axel Jorswieck, Mérouane Debbah, Björn E. Ottersten

Abstract

As 5G networks evolve, they must satisfy a heterogeneous set of requirements, including massive connectivity, ultra-high data rates, and extreme energy efficiency. These goals are often in direct conflict, making traditional single-objective optimization insufficient. This paper explores Multi-Objective Optimization (MOO) as a formal framework for navigating these trade-offs. By identifying the Pareto boundary—the set of points where no metric can be improved without sacrificing another—engineers can make informed decisions about network operation. A case study in Massive MIMO illustrates how MOO uncovers optimal configurations that balance user experience with system-wide efficiency.

Introduction: Beyond Single-Objective Design

Previous generations of cellular technology focused primarily on a single metric: peak data rate. In the 5G era, however, designers must simultaneously manage 10–100x higher user rates, 1000x higher area rates, and a 1000x improvement in energy efficiency.

The standard engineering response has been to maximize one primary objective while treating others as fixed constraints. This heuristic approach is limiting; it assumes the "correct" constraint values are known a priori and can blind designers to superior operating points. MOO offers a more rigorous alternative by treating all conflicting goals as equal components of the optimization problem.

The MOO Framework and the Pareto Boundary

The core of MOO is the mapping of available resources (the "resource bundle") to an "attainable objective set." Because objectives conflict, there is rarely a single "global optimum." Instead, the goal is to find the Pareto boundary—the frontier of the attainable set where any further gain in one objective necessitates a loss in another.

Mapping from the resource space to the attainable objective set.

Optimization Strategies

The paper identifies two primary strategies for navigating this boundary:

  1. A Posteriori Methods (Visualization): These involve sampling the Pareto boundary to provide a visual map of trade-offs. This allows designers to subjectively choose an operating point based on current network priorities.
  2. A Priori Methods (Scalarization): These combine multiple objectives into a single scalar "goal function" based on pre-defined preferences.

A significant contribution of this work is the recommendation of the weighted Chebyshev goal function. While many scalarization methods (like weighted sums) can lead to NP-hard problems in complex network environments, the Chebyshev approach often maintains mathematical tractability (convexity) while ensuring the discovery of Pareto optimal points.

Comparison of different scalarization goals on the Pareto boundary.

Case Study: Massive MIMO Trade-offs

To demonstrate the framework, the authors simulated a 16-cell Massive MIMO network. The analysis focused on three key metrics: average user rate, average area rate, and energy efficiency (EE).

Key Findings

  • Energy Efficiency vs. User Rate: These metrics align initially but conflict after a certain threshold. Peak EE (11.1 Mbit/J) was achieved at a user rate of 20.4 Mbit/s; pushing rates higher led to a sharp decline in efficiency.
  • Area Rate vs. Energy Efficiency: These objectives are more compatible. Once peak EE is reached, area rates can be increased further with only marginal impacts on efficiency.
  • The "Many Users" Strategy: High area rates are most efficiently achieved by serving many users in parallel at moderate individual rates, rather than providing extreme rates to a few users.

Conclusion and Future Directions

MOO provides a mathematical foundation for the flexibility required in 5G architectures. It reveals that different network priorities require fundamentally different resource allocations, such as dynamically adjusting antenna counts or precoding strategies.

While powerful, MOO faces challenges in high-dimensional visualization and computational complexity. Future work must focus on developing tractable MOO formulations that account for real-world imperfections like uncertain channel states and hardware limitations, enabling real-time, multi-objective network adaptation.

Summary updated

BibTeX
@article{bjornson2014multiobjective,
  author = {Emil Björnson and Eduard Axel Jorswieck and Mérouane Debbah and Björn E. Ottersten},
  title = {{Multiobjective Signal Processing Optimization: The way to balance conflicting metrics in 5G systems}},
  journal = {IEEE Signal Process. Mag.},
  year = {2014},
  doi = {10.1109/msp.2014.2330661},
}
Figure