Introduction
The proliferation of Internet of Things (IoT) devices necessitates energy-efficient communication strategies to extend the operational lifespan of battery-constrained sensors and wearables. Reconfigurable Intelligent Surfaces (RIS) have emerged as a transformative technology, enabling the manipulation of the wireless propagation environment through low-cost reflecting elements. This paper addresses the critical challenge of minimizing uplink transmit power in RIS-aided IoT networks through a novel optimization framework.
The Optimization Challenge
Optimizing RIS-aided systems involves a complex interplay between the RIS phase shifts and the Base Station (BS) beamforming vectors. This problem is inherently non-convex due to two primary constraints:
- Unit-Modulus Constraints: RIS elements can only change the phase of a signal, not its amplitude.
- Unit-Norm Constraints: BS beamforming vectors must maintain a specific power scale.
Traditional methods, such as Semidefinite Relaxation (SDR), often suffer from high computational complexity ($O(N^6)$) and performance degradation due to rank-one approximation errors. Furthermore, many existing solutions rely on "alternating optimization," which may fail to find the global optimum in such complex search spaces.
Methodology: Riemannian Manifold Optimization
To overcome these hurdles, the authors propose the Riemannian Conjugate Gradient-based Joint Optimization (RCG-JO) algorithm. The core innovation lies in mapping the optimization variables onto a product Riemannian manifold.
By recognizing that the set of all possible phase shifts forms a "complex circle manifold" and beamforming vectors form a "complex oblique manifold," the researchers transformed the constrained non-convex problem into an unconstrained problem on a curved manifold surface. This allows the use of differential geometry tools—such as Riemannian gradients and retractions—to navigate the search space much more efficiently than Euclidean solvers.
Joint Optimization and Channel Estimation
Unlike conventional techniques, RCG-JO performs joint optimization of both variables simultaneously on the product manifold. To support this, the authors also introduced a channel estimation technique using Low-Rank Matrix Completion (LRMC). By using a few "active" RIS elements to sample the channel, the full channel state is reconstructed by exploiting the fact that RIS channels are often dominated by a few Line-of-Sight (LoS) paths.
Performance Evaluation
The proposed algorithm was evaluated through extensive numerical simulations using Rician fading and standard path loss models. The setup included a BS with 4 antennas and an RIS with up to 80 elements serving multiple IoT devices.
Key Results
- Massive Power Savings: RCG-JO achieved a 94% reduction in uplink transmit power compared to a system without RIS, highlighting the technology's potential for battery-limited environments.
- Superiority over Benchmarks: The algorithm outperformed SDR-based schemes by 44% and Deep Reinforcement Learning (DRL) approaches by 57% in terms of power reduction.
- Computational Efficiency: RCG-JO is approximately 98% faster in terms of floating-point operations than SDR-based schemes.
- Rapid Convergence: Numerical results showed that the algorithm typically converges to a stable solution within 15 to 30 iterations.
Conclusion and Future Outlook
The RCG-JO algorithm provides a robust and efficient framework for power control in next-generation IoT networks. By leveraging the geometric structure of the problem constraints, it achieves significant energy savings with low computational overhead.
Future research directions include addressing the limitations of the current model, such as the assumption of continuous phase shifts (whereas real hardware often uses discrete, quantized levels) and the potential coupling between reflection amplitude and phase shift. Additionally, scaling these manifold techniques to extremely large-scale RIS deployments remains a vital area for exploration.
Summary updated