Abstract
This letter considers a point-to-point link where a reconfigurable intelligent surface assists the communication between a transmitter and a receiver. The system rate, energy efficiency, and their trade-off are optimized by tuning the number of reflecting elements to be activated and the phase shifts that they apply. Unlike most previous works, the considered resource allocation problem explicitly accounts, at the design stage, for the time and energy that are necessary for channel estimation and for reporting the optimal configuration of the phase shifts to the reconfigurable intelligent surface. Numerical results confirm the optimality of the proposed methods and show the potential gains of reconfigurable intelligent surfaces.
Introduction
Reconfigurable Intelligent Surfaces (RIS) are a cornerstone of 6G research, consisting of arrays of "meta-atoms" that can steer electromagnetic waves to improve wireless links. Conventional wisdom suggests that more elements are always better: more elements mean a narrower beam and higher gain. However, in a real-world system, you cannot use an RIS without first knowing the channel state. To optimize $N$ reflecting elements, the transmitter must send pilots to estimate $N$ different channels, and the controller must then "report" the optimized phase shifts back to the RIS. This creates a fundamental trade-off: increasing $N$ improves the signal-to-noise ratio (SNR) during data transmission but also increases the time spent on channel estimation ($T_E$) and feedback ($T_F$), leaving less time for actual data transfer within a coherence time slot.
Methodology
The authors model a point-to-point link where the direct path is blocked, forcing communication through an RIS. They formulate optimization problems for three objectives: the system rate, the energy efficiency (EE), and a bi-objective trade-off between the two.
Key technical contributions include:
- Overhead-Aware Modeling: Explicitly includes the time and power costs of channel estimation and resource allocation reporting.
- Unimodality Proofs: Mathematical proof that the rate and EE functions are unimodal with respect to $N$, meaning they have a single global maximum.
- Greedy Optimization: A simple greedy algorithm is shown to be guaranteed to find the global optimum due to the unimodal nature of the objective functions.
- Physical Constraints: Derivation of a physical upper bound for $N$ to ensure the path-loss model remains consistent with the law of conservation of energy.
Key Insights: The "More is Less" Paradox
In RIS-aided systems, there is a point where adding one more reflecting element actually decreases the total throughput because the extra time needed to estimate its channel outweighs the marginal gain in signal strength.
- Power Sensitivity: The optimal number of elements ($N^*$) is sensitive to the transmit power. As transmit power increases, the optimal $N$ for rate maximization actually decreases slightly. At high SNR, the system is less desperate for the gain provided by the RIS and more sensitive to the loss of transmission time.
- Hardware Power Impact: For Energy Efficiency, the "static" power consumed by each active RIS element ($P_{c,n}$) is a major factor. High hardware costs per element significantly drop the optimal $N$ to save power.
Performance Evaluation
Numerical results validated using Monte Carlo simulations with $10^4$ realizations under Ricean fading confirm the theoretical findings.
As shown in Figure 1, the proposed optimization of both phase shifts ($\Phi$) and the number of elements ($N$) significantly outperforms schemes that use a random number of elements or random phase shifts. Even though the optimized scheme pays a "time penalty" for channel estimation, the resulting beamforming gain is so much higher that it yields a massive net benefit.
The authors also successfully mapped the trade-off between Rate and EE (Figure 2), showing a Pareto frontier. This allows designers to choose a balance point based on whether they prioritize raw speed or battery life. The impact of hardware power consumption ($P_{cn}$) is also evident, shifting the EE curves significantly and demonstrating that the optimal size of the surface is highly dependent on hardware efficiency.
Conclusion
This work reveals that there is an optimal, finite number of elements for an RIS when accounting for realistic overheads. For a typical setup, the optimal number of elements for rate maximization was found to be around 170 to 200. Future work will extend this to multi-user MIMO scenarios where overhead scales even more aggressively, making the choice of $N$ even more critical.
Summary updated