Abstract
Reconfigurable distributed antenna and reflecting surface (RDARS) is a promising architecture for future sixth-generation (6G) wireless networks. In particular, the dynamic working mode configuration for the RDARS-aided system brings an extra selection gain compared to the existing reconfigurable intelligent surface (RIS)-aided system and distributed antenna system (DAS). In this paper, we consider the RDARS-aided downlink multiple-input multiple-output (MIMO) system and aim to maximize the weighted sum rate (WSR) by jointly optimizing the beamforming matrices at the based station (BS) and RDARS, as well as mode switching matrix at RDARS. The optimization problem is challenging to be solved due to the non-convex objective function and mixed integer binary constraint. To this end, a penalty term-based weight minimum mean square error (PWM) algorithm is proposed by integrating the majorization-minimization (MM) and weight minimum mean square error (WMMSE) methods. To further escape the local optimum point in the PWM algorithm, a model-driven DL method is integrated into this algorithm, where the key variables related to the convergence of PWM algorithm are trained to accelerate the convergence speed and improve the system performance. Simulation results are provided to show that the PWM-based beamforming network (PWM-BFNet) can reduce the number of iterations by half and achieve performance improvements of 26.53% and 103.2% at the scenarios of high total transmit power and a large number of RDARS transmit elements (TEs), respectively.
Introduction: The RDARS Architecture
As we move toward 6G, researchers are looking for ways to overcome the high path loss of millimeter-wave (mmWave) and Terahertz bands. Two popular solutions have emerged: Distributed Antenna Systems (DAS), which use multiple active antennas to boost coverage but are expensive and power-hungry, and Reconfigurable Intelligent Surfaces (RIS), which are cheap, passive reflectors but suffer from "multiplicative fading" (signal loss at every bounce).
The RDARS architecture attempts to find the "Goldilocks" zone by creating a surface where elements can be either active (connected to the base station via cable) or passive (reflecting incident waves). This hybrid approach provides a unique "selection gain" that RIS and DAS lack. By choosing exactly which elements are active based on the current channel state, the system can bypass obstacles more effectively than a purely passive surface.
The Optimization Challenge
While RDARS offers superior flexibility, it creates a complex optimization problem. Engineers must jointly decide which elements should switch modes, how the base station should beamform, and how the passive elements should shift phases. This results in a non-convex, mixed-integer nonlinear programming (MINLP) problem that is notoriously difficult to solve efficiently in real-time.
Proposed Solution: PWM-BFNet
The authors tackle this in two stages. First, they developed the PWM Algorithm, a conventional iterative method combining WMMSE (to manage the sum rate objective), Majorization-Minimization (to handle unit-modulus constraints), and Penalty Methods (to force discrete binary decisions into a continuous framework).
Recognizing that the standard PWM algorithm is slow and prone to getting stuck in local optima, they introduced PWM-BFNet. This model-driven deep learning framework "unfolds" the iterative algorithm into a neural network.
Instead of using a black-box AI, PWM-BFNet maintains the mathematical structure of the PWM algorithm but uses trainable layers to learn optimal penalty parameters ($\rho$) and regularization terms ($\epsilon$) for each iteration. This ensures the AI respects the physics of wireless propagation while using learning to skip the tedious manual tuning of hyper-parameters.
Key Results and Insights
The study utilized 10,000 synthetic channel realizations generated using a Rician fading model to simulate mmWave environments. The setup included a base station with 16 antennas serving 4 users, with an RDARS surface containing 128 elements. Key findings include:
- Efficiency: PWM-BFNet reduced the number of iterations required for convergence by 50% compared to the standard PWM algorithm.
- Performance Gains: In scenarios with high transmit power, the proposed method improved the weighted sum rate by 26.53%. When the number of transmit elements (active modes) on the RDARS was large, the performance improvement jumped to 103.2% over benchmark schemes.
- Robustness: RDARS consistently outperformed both pure RIS and pure DAS architectures, proving that a hybrid approach is more robust to varying user locations and blockages.
Conclusion
PWM-BFNet demonstrates that model-driven deep learning can significantly enhance the performance of hybrid 6G architectures. By unfolding traditional optimization algorithms, researchers can achieve the speed of neural networks without sacrificing the interpretability and physical constraints of wireless communication theory. Future work will focus on multi-cell scenarios and high-mobility environments where channel estimation overhead becomes a more critical factor.
Summary updated