TL;DR
This paper tackles the challenge of maximizing spectral efficiency in urban smart city wireless networks by jointly optimizing beamforming and Reconfigurable Intelligent Surface (RIS) phase shifts in a Full-Duplex (FD) system. They propose a novel two-step algorithm combining successive convex approximation (SCA) and soft actor-critic (SAC) deep reinforcement learning to efficiently handle complex interference scenarios, outperforming existing methods.
What Problem Does This Paper Address?
The paper addresses the limitations of current wireless communication systems, particularly in the context of 6G and the Internet of Things (IoT) in smart cities. While Full-Duplex (FD) technology promises to double spectral efficiency (SE) by allowing simultaneous transmission and reception, it suffers from self-interference (SI). Reconfigurable Intelligent Surfaces (RIS) offer a promising solution to enhance signal coverage and quality with low power consumption. However, combining FD and RIS introduces new challenges, specifically "loop-interference" (LI) from the RIS reflecting signals back to the FD base station. In complex urban environments with blockages and reflections, this LI, combined with residual SI, can significantly degrade performance, negating the benefits of FD-RIS systems. The paper aims to overcome these interference issues by jointly optimizing system parameters in realistic urban outdoor scenarios.
Methods
The authors propose a two-step optimization approach to tackle the non-convex problem of maximizing spectral efficiency.
- Beamforming Design (Subproblem 1): This subproblem focuses on optimizing the transmitting and receiving beamforming matrices. To handle the non-convexity, they employ a Successive Convex Approximation (SCA) technique. This involves approximating the objective function and constraints with convex lower bounds, transforming the problem into a solvable convex optimization problem that can be addressed using tools like CVX.
- RIS Phase Shifts Design (Subproblem 2): This subproblem optimizes the phase shifts of the RIS elements. Given the discrete nature of RIS phase shifts and the non-convexity, they introduce a Soft Actor-Critic (SAC) deep reinforcement learning (DRL) algorithm. SAC is a model-free DRL approach known for its stability and ability to handle discrete action spaces, making it suitable for optimizing RIS phase shifts. The algorithm learns an optimal policy by interacting with the environment, aiming to maximize the sum SE.
These two subproblems are solved iteratively until convergence, forming a joint optimization strategy.
Datasets
The paper does not explicitly mention training or evaluating on pre-existing datasets. Instead, it relies on simulated data generated based on a defined system model and parameters for two typical urban outdoor scenarios. The performance is evaluated against various benchmarks, including exhaustive search, fixed-point iteration, continuous phase shifts, and random phase shifts methods.
Key Insights and Ideas
- Interference is Key: The paper highlights that in urban environments, the interplay between residual SI and RIS-induced LI is a critical bottleneck for FD-RIS systems, not just the inherent FD SI.
- Scenario-Specific Interference: The authors identify two distinct urban scenarios where either UL-to-DL interference (UDI) or LI becomes the dominant interference factor, necessitating tailored optimization strategies.
- DRL for Discrete Optimization: The successful application of SAC DRL for optimizing discrete RIS phase shifts demonstrates its power in handling complex, real-world communication constraints.
- Joint Optimization is Crucial: Optimizing beamforming and RIS phase shifts jointly, rather than independently, leads to significant performance gains, especially in challenging interference conditions.
- RIS Location Matters: The placement of the RIS significantly impacts performance, with optimal locations found closer to the BS or IoT devices depending on the scenario.
Key Results
- The proposed joint beamforming and phase shift design algorithm achieves performance only slightly lower than the exhaustive search method (considered an upper bound for discrete phase shifts) but with significantly reduced computational complexity.
- The proposed algorithm outperforms the fixed-point iteration scheme and other baseline methods.
- In scenario two, where LI is more pronounced, the proposed algorithm shows superior performance, demonstrating its advantage in mitigating complex interference in urban environments.
- The performance gains from RIS are substantial, with the proposed algorithm achieving significant improvements compared to scenarios without RIS or with random phase shifts. For instance, in scenario one, RIS deployment on the BS side yielded a 135% performance gain.
- The algorithm demonstrates robustness across various parameters, including the number of RIS elements, the number of resolution bits, and residual SI levels.
Limitations and Future Work
- Perfect CSI Assumption: The paper assumes perfect channel state information (CSI) is known, which is idealistic in practice. The authors suggest that considering imperfect CSI would be a valuable future research direction.
- SE vs. EE: While the paper focuses on maximizing spectral efficiency (SE), it acknowledges that the RIS-aided system is also power-saving. Future work could involve formulating an energy efficiency (EE) objective function to further investigate this aspect.
- Model-Based vs. Model-Free: The paper uses a model-free DRL approach for phase shifts. Exploring model-based DRL or hybrid approaches could offer alternative optimization strategies.
- Real-World Deployment: The study is based on simulations. Further research on practical implementation and validation in real-world testbeds would be beneficial.
Summary updated