Reconfigurable intelligent surfaces (RISs) enhance communication performance by adjusting the propagation directions of incident signals. However, joint beamforming design requires the acquisition of channel state information, often leading to significant pilot overhead in RIS-assisted systems, particularly when the number of reflective elements is large. In this study, we analyze the characteristics of the cascaded channel and propose a method that combines amplitude correlation with existing structured sparsity. Leveraging these characteristics, we first derive an on-grid channel estimation method, demonstrating the effectiveness of incorporating additional characteristics in cascaded channel estimation. We then extend the proposed algorithm to off-grid channel estimation by refining the coarsely estimated channel using alternating optimization and gradient descent. Furthermore, we adapt the algorithm to enhance estimation accuracy with the support of digital twin (DT) technology, utilizing a few pilots to refine the channel generated by DT. Simulation results show up to a 5 dB improvement in normalized mean squared error compared to state-of-the-art channel estimation algorithms that employ structured sparsity. Additionally, with DT assistance, the proposed algorithm achieves nearly a two-fold performance improvement over traditional algorithms that do not incorporate amplitude correlation and structured sparsity.

Reconfigurable intelligent surfaces (RIS) have emerged as a transformative technology for enhancing wireless coverage and transmission rates while reducing hardware costs and power consumption. This work addresses the limitations of separately optimizing RIS deployment and beamforming by proposing a unified joint deployment and beamforming framework tailored for multi-user multi-input single-output systems. By formulating RIS control as a Markov decision process, we develop a deep reinforcement learning framework that integrates a graph neural network to exploit the inherent topology of wireless communication networks. To reduce the action space and improve learning efficiency, the framework leverages discrete Fourier transform codebooks. Simulation results demonstrate that the proposed approach achieves up to a twofold improvement in weighted sum rate compared to fixed RIS deployment strategies, all while eliminating the need for explicit cascaded channel estimation and accurate channel model.

The deployment of multiple reconfigurable intelligent surfaces (RIS) is a promising strategy to enhance wireless system performance. However, joint beamforming in multi-RIS assisted systems faces significant challenges due to the increased number of optimization variables, non-convex objective functions, and constraints. In this study, we propose an algorithm based on weighted minimum mean square error optimization and the successive convex approximation algorithm, maximizing the weighted sum rate in a double-RIS assisted downlink multi-user multiple-input single-output system. We also present a general framework for model-driven deep learning that addresses the limitations of existing methods, which often lack flexibility to different channels and suffer from a large training burden due to the high-dimensional action space of deep reinforcement learning (DRL). Initially, we configure the step size in the proposed algorithm as trainable, accelerating convergence. Then, a recurrent neural network generates the step size for iterations, allowing dynamic iteration extension in varying environmental conditions. We enhance the neural network’s self-adaptability by introducing a model-driven DRL algorithm, integrating expert knowledge into the DRL actor network’s design. Simulation results demonstrate up to 30% performance improvement over traditional algorithms, achieved by our model-driven framework. The proposed model-driven DRL shows higher capacity for dynamic extension and rapid adaptation to new environments.

Thanks to the ability to customize the propagation of wireless signals, reconfigurable intelligent surfaces (RISs) have great potential in enhancing the performance of future wireless communication systems. While the majority of papers in the literature considers single-RIS scenarios, the potential deployment of multiple RISs, that offer ubiquitous connectivity for diverse user demands, calls for further investigation. This paper considers a downlink multi-group multicast system underpinned by multiple RISs and aims to maximize the sum spectral efficiency subject to an overall transmit power constraint. This optimization problem is highly challenging due to the non-convex, non-smooth, and non-differentiable properties of the objective function, as well as the non-convex unit modulus constraint. To address this complex problem, we propose a model-driven deep learning (DL) approach. This involves first solving the joint active and passive beamforming design through an alternating projected gradient (APG) algorithm with an approximate objective function. The APG algorithm is then unfolded into an iterative procedure using multiple layers with trainable parameters. A network training method is proposed to ensure that the performance improves with the number of iterations. Remarkably, our model is also nicely generalizable to the imperfect channel state information (CSI) scenario, without any change to the network architecture, by simply combining the recursive approximation method and adding some long/short-term trainable parameters to accommodate the two-timescale transmission protocol. Our simulation results demonstrate the superiority of our proposed DL method over existing algorithms in terms of both complexity and performance. Specifically, the proposed model-driven DL method reduces the runtime by approximately 80% compared to the APG algorithm and 99.97% compared to the majorization-minimization algorithm, while it also achieves comparable performance. Furthermore, our proposed method for imperfect CSI scenarios reduces the performance loss by 5%-10% compared to the proposed method without considering the influence of imperfect CSI.

Reconfigurable intelligent surfaces (RIS) can improve signal propagation environments by adjusting the phase of the incident signal. However, optimizing the phase shifts jointly with the beamforming vector at the access point is challenging due to the non-convex objective function and constraints. In this study, we propose an algorithm based on weighted minimum mean square error optimization and power iteration to maximize the weighted sum rate (WSR) of a RIS-assisted downlink multi-user multiple-input single-output system. To further improve performance, a model-driven deep learning (DL) approach is designed, where trainable variables and graph neural networks are introduced to accelerate the convergence of the proposed algorithm. We also extend the proposed method to include beamforming with imperfect channel state information and derive a two-timescale stochastic optimization algorithm. Simulation results show that the proposed algorithm outperforms state-of-the-art algorithms in terms of complexity and WSR. Specifically, the model-driven DL approach has a runtime that is approximately 3% of the state-of-the-art algorithm to achieve the same performance. Additionally, the proposed algorithm with 2-bit phase shifters outperforms the compared algorithm with continuous phase shift.

The use of a hybrid analog-digital architecture that connects one RF chain to multiple antennas through phase shifters is an energy-efficient solution for multiuser multiple-input multiple-output (MU-MIMO) systems. However, designing the hybrid precoder is challenging due to its nonconvex objective functions and constraints. Existing algorithms struggle with high computational complexity or poor performance, which often result from slow or no convergence. This study proposes a solution that leverages model-driven deep learning (DL) to maximize the spectral efficiency of MU-MIMO systems through hybrid precoding. The optimization problem is first transformed into a weighted minimum mean square error optimization. Then, it is combined with manifold optimization and DL to improve performance and simplify the process. The algorithm is designed to be robust in changing environments and utilizes DL to address imperfect channel state information. Simulation results show that the proposed method outperforms existing algorithms, is robust in changing system parameters, and can even outperforms fully digital precoding with the same number of antennas.

Millimeter-wave (mmWave) communications have been one of the promising technologies for future wireless networks that integrate a wide range of data-demanding applications. To compensate for the large channel attenuation in mmWave band and avoid high hardware cost, a lens-based beamspace massive multiple-input multiple-output (MIMO) system is considered. However, the spatial-wideband effect in wideband mmWave systems makes channel estimation very challenging, especially when the receiver is equipped with a limited number of radio-frequency (RF) chains. Furthermore, the real channel data cannot be obtained before the mmWave system is used in a new environment, which makes it impossible to train a deep learning (DL)-based channel estimator using real data set beforehand. To solve the problem, we propose a model-driven unsupervised learning network, named learned denoising-based generalized expectation consistent (LDGEC) signal recovery network. By utilizing the Stein’s unbiased risk estimator loss, the LDGEC network can be trained only with limited measurements corresponding to the pilot symbols, instead of the real channel data. Even if designed for unsupervised learning, the LDGEC network can be supervisingly trained with the real channel via the denoiser-by-denoiser way. The numerical results demonstrate that the LDGEC-based channel estimator significantly outperforms state-of-the-art compressive sensing-based algorithms when the receiver is equipped with a small number of RF chains and low-resolution ADCs.

Channel estimation in wideband millimeter-wave (mmWave) systems is very challenging due to the beam squint effect. To solve the problem, we propose a learnable iterative shrinkage thresholding algorithm-based channel estimator (LISTA-CE) based on deep learning. The proposed channel estimator can learn to transform the beam-frequency mmWave channel into the domain with sparse features through training data. The transform domain enables us to adopt a simple denoiser with few trainable parameters. We further enhance the adaptivity of the estimator by introducing hypernetwork to automatically generate learnable parameters for LISTA-CE online. Simulation results show that the proposed approach can significantly outperform the state-of-the-art deep learning-based algorithms with lower complexity and fewer parameters and adapt to new scenarios rapidly.