In wireless communications, the traditional design paradigm focused almost exclusively on maximizing spectral efficiency—squeezing as many bits per second as possible out of a limited frequency band. To achieve this, systems scaled up transmit power and deployed increasingly dense infrastructure. However, as the industry shifts toward green computing and energy-constrained Internet of Things (IoT) networks, the primary metric has evolved from raw throughput to energy efficiency (EE), measured in bits per Joule.
The core challenge of optimizing for energy efficiency is that it does not scale monotonically. While spectral efficiency always increases with more power or more hardware elements, energy efficiency exhibits a distinct saturation behavior. It rises to an optimal peak and then sharply declines. Understanding why this saturation occurs requires looking at two distinct dimensions of wireless networks: transmit power dynamics and hardware infrastructure scaling.
The Power Efficiency Tradeoff: Logarithmic Gain vs. Linear Cost
To understand why energy efficiency saturates as transmit power increases, we must look at the mathematical structure of the bits-per-Joule metric. Energy efficiency is fundamentally defined as the ratio of the achievable data rate to the total power consumed by the system:
$$\text{Energy Efficiency} = \frac{\text{Data Rate}}{\text{Transmit Power} + \text{Static Circuit Power}}$$
This formulation reveals an inherent structural mismatch. According to Shannon’s capacity formula, the achievable data rate scales logarithmically with transmit power ($\log(1 + \text{SNR})$). Meanwhile, the total power consumed by the hardware scales linearly with the transmit power, anchored by a fixed overhead cost ($P_{\text{static}}$) required to keep the internal circuitry running.
Energy Efficiency (bits/Joule)
^
| Optimal Peak
| / \
| / \
| / \
| / \
| / \
+---------------------> Transmit Power
At very low transmit power levels, the static circuit power dominates the denominator, rendering the system highly inefficient because energy is being spent just to keep the device turned on without generating meaningful throughput. As transmit power increases, the data rate initially climbs quickly, outpacing the power consumption and driving energy efficiency upward.
However, as detailed by Ming Zeng et al. in Energy-Efficient Resource Allocation for IRS-Assisted Multi-Antenna Uplink Systems, the system eventually hits a critical threshold. Beyond this point, the logarithmic growth of the data rate flattens out due to diminishing returns, while the power consumption continues to rise linearly.
As a result, the overall energy efficiency curve is strictly concave. As Stefano Buzzi et al. outline in A Survey of Energy-Efficient Techniques for 5G Networks and Challenges Ahead, pushing transmit power past this optimal peak actively degrades the bits-per-Joule performance. This non-monotonic behavior means that standard power-maximization strategies are fundamentally counterproductive for green networks, a reality that requires joint optimization of transmission time and power boundaries, as shown by Songtao Lu et al. in Joint Optimization of Power Allocation and Training Duration for Uplink Multiuser MIMO Communications.
The Hardware Scaling Limit: The Cost of Control Circuits
A parallel saturation phenomenon occurs when scaling up physical hardware infrastructure, particularly with the introduction of Reconfigurable Intelligent Surfaces (RIS) or Large Intelligent Surfaces (LIS). An RIS is an array of passive reflecting elements that can dynamically alter the phase of electromagnetic waves, focusing wireless signals toward a receiver without requiring power-hungry radio frequency (RF) chains.
Intuitively, adding more reflecting elements to a surface should continuously improve energy efficiency by sharpening the signal beam and reducing the required transmit power at the base station. However, this assumption ignores the hidden energy cost of the hardware itself. While the elements are "passive" in terms of signal transmission, each individual element requires active control electronics (such as PIN diodes or varactors) to maintain its phase configuration.
The total power model for an RIS-assisted system must therefore account for this hardware scaling:
$$P_{\text{total}} = P_{\text{transmit}} + P_{\text{static}} + N \cdot P_{\text{element}}$$
where $N$ is the number of reflecting elements and $P_{\text{element}}$ is the power consumed by the control circuit of a single element.
Chongwen Huang et al. demonstrated in Reconfigurable Intelligent Surfaces for Energy Efficiency in Wireless Communication that this linear accumulation of hardware power dissipation eventually triggers a severe saturation point. Initially, increasing $N$ provides massive beamforming gains that drastically reduce the required transmit power. But as $N$ grows large, the marginal spectral efficiency gains begin to plateau, while the cumulative power consumption of the control circuits ($N \cdot P_{\text{element}}$) continues to grow linearly.
In Energy Efficient Multi-User MISO Communication using Low Resolution Large Intelligent Surfaces, the authors formalized this tradeoff, proving that adding hardware elements beyond a specific threshold becomes entirely counterproductive. This hardware-driven efficiency ceiling has been observed across a variety of emerging network deployments, from vehicle-to-infrastructure safety links studied by Kangwei Qi et al. in Deep-Reinforcement-Learning-Based AoI-Aware Resource Allocation for RIS-Aided IoV Networks to edge-computed autonomous driving networks analyzed by Bo Yang et al. in Reconfigurable Intelligent Computational Surfaces for MEC-Assisted Autonomous Driving Networks: Design Optimization and Analysis.
Designing Around the Critical Threshold
Because energy efficiency naturally saturates and declines, modern resource allocation frameworks cannot simply operate on a "more is better" philosophy. Instead, optimization algorithms must be designed to explicitly locate and adapt to these critical thresholds.
Power Capping Strategies
Rather than exhausting the entire available power budget to maximize signal strength, energy-efficient protocols enforce strict power caps. For example, Itsikiantsoa Randrianantenaina et al. developed a framework in Consumption Factor Optimization for Multihop Relaying over Nakagami-m Fading channels that mathematically derives the exact point where the marginal gain in data rate no longer justifies the energy expenditure. Once this critical power threshold is reached, the allocation is capped, prioritizing the bits-per-Joule ratio over raw capacity. This approach prevents the system from drifting into the inefficient high-power zone, a strategy also mirrored by Hoang Duong Tuan et al. in Joint Power Allocation and Beamforming for Energy-Efficient Two-Way Multi-Relay Communications through the use of iterative path-following algorithms.
Modular Activation
To combat hardware-induced saturation in massive surface arrays, researchers have turned to modular architectures. Instead of treating an intelligent surface as an "always-on" utility, systems can dynamically activate only a subset of the available elements.
Yulan Gao et al. explored this in Reflection Resource Management for Intelligent Reflecting Surface Aided Wireless Networks, proposing a modular architecture where groups of reflecting elements are selectively triggered based on real-time channel conditions. If the channel is already sufficient, outstanding modules remain deactivated. This prevents the control circuit power overhead from ballooning, ensuring that the hardware footprint expands and contracts to remain precisely at the peak of the energy efficiency curve.
Key Foundational Work
Transmit power optimization for green multihop relaying over Nakagami-m fading channels (2014) — contributed 3 concept elements to this direction.
Consumption Factor Optimization for Multihop Relaying over Nakagami-m Fading channels (2014) — contributed 3 concept elements to this direction.
Coordinated Beamforming for Energy Efficient Transmission in Multicell Multiuser Systems (2013) — contributed 1 concept elements to this direction.