Energy-Efficient RIS-Aided Cell-Free Massive MIMO Systems: Application, Opportunities, and Challenges

Fig. 3 delineates the signal transmission processes inherent in a cellular mMIMO system, a CF mMIMO system, and an RIS-aided CF mMIMO system. This includes the uplink channel estimation and downlink signal transmission phases. In the cellular mMIMO system, the BS is responsible for CSI acquisition through pilot signals from UEs and then proceeds with signal transmission based on these estimates. Besides, power control is centralized at the BS to manage transmission power across cells [3]. While effective, this centralized approach can lead to inefficiencies and limitations in dynamic environments. In contrast, the distributed nature of the CF mMIMO system offers a more adaptable framework for uplink and downlink signal transmission. Each AP in the CF mMIMO system estimates the channels locally in uplink transmission, and these local estimates are then aggregated at the CPU to establish a comprehensive view of the UE data [2]. This distributed channel estimation approach allows quicker responses to local channel conditions, reducing latency and enhancing the system’s ability to adapt to changes. For downlink operations, the CPU performs encoding while delegating power allocation to the APs, which can adapt their transmit power based on local conditions. This decentralization of power control not only reduces the computational load at the CPU but also increases the system’s flexibility and scalability, thereby improving adaptability in dynamic wireless environments [3].

In an RIS-aided CF mMIMO system, the RIS serves as a substitute for APs, effectively reducing the system’s AP count and energy consumption. It is important to note that the RIS, depicted in Fig. 3, is a passive component rather than an energy-harvesting device that relies on energy supplied by the APs for its functioning [7]. The design of RIS phase shifts is strategically based on statistical CSI, which is less time-varying than the instantaneous CSI. This approach allows for less frequent updates of the RIS phase shifts, thereby significantly reducing the overhead associated with channel estimation [8]. The complexity of CSI acquisition rises with the growing number of antennas and channels. To enhance the process, methods such as pilot-based estimation and AI-driven machine learning can be utilized to achieve greater system performance. Furthermore, in both CF mMIMO and RIS-aided CF mMIMO systems, distributed power control at APs is optimized for energy efficiency and adaptability to varying conditions through long-term and short-term strategies. Machine learning and AI can also be utilized to improve power control dynamics. The main difference lies in their signal transmission schemes, i.e., direct transmission in CF mMIMO versus RIS reflection transmission in RIS-aided systems [5].

The deployment of reflecting elements in RIS-aided CF systems can be implemented using two strategies: centralized and distributed deployments (as depicted in Fig. 2). Centralized deployment consolidates elements on a single RIS, simplifying power allocation and reducing computational complexity and pilot overhead, with centrally optimized beamforming for better signal strength and coverage. Distributed deployment spreads elements across multiple RIS units, complicating channel estimation but enhancing spatial diversity and resistance to fading. Each RIS unit optimizes beamforming locally while coordinating with a central controller, improving coverage and connectivity. In actual deployment scenarios, distributed configurations typically outperform centralized ones in various aspects. Despite higher power requirements, the distributed setups provide improved independence and space-time diversity with an equivalent number of RIS elements, making them more adaptable to the specific needs of the communication system [9]. In addition, hybrid deployment which combines centralized and distributed strategies is practically applicable. This approach places a central RIS offering broad coverage, while smaller distributed RIS units provide targeted enhancements. This approach balances the simplicity of centralized control with the performance benefits of distributed setups, optimizing coverage, and signal quality, and managing complexity and power consumption effectively.

In summary, cellular mMIMO systems rely on centralized power control by the BS, which can lead to performance degradation at the edges. CF mMIMO systems implement distributed power control at APs, ensuring uniform performance and enhancing system flexibility. RIS-aided CF mMIMO systems combine distributed AP control with passive RIS control, optimizing power allocation and improving system adaptability. Additionally, reflecting elements in RIS can be deployed in centralized, distributed, or hybrid configurations, each offering unique advantages in terms of power management, spatial diversity, and overall system efficiency.

The primary challenge faced by existing CF mMIMO systems is the high energy consumption due to extensive AP deployment. To mitigate this, we propose substituting multiple APs with RIS to reduce overall energy consumption. However, RIS integration introduces interference, exacerbating energy consumption issues as interference energy is associated with transmitted signals. Specifically, mutual coupling effects and EMI during signal transmission, stemming from interference within RIS elements, worsen with spatial correlation between RIS elements. Additionally, RIS integration intensifies inter-operator interference, directly proportional to transmitted signal power. While these challenges are prevalent in other RIS-aided wireless networks, CF mMIMO systems, employing large-scale fading decoding (LSFD) at the APs [2], are less prone to interference but require precise power control. Inconsistent power management among densely deployed APs can degrade system performance. Effective interference mitigation demands varied cooperation strategies among APs, setting RIS-aided CF mMIMO apart from other RIS-aided networks. Subsequent sections explore these adverse effects.

In the RIS-aided CF mMIMO system, beam alignment is intricately challenged by the need for precise coordination among APs and the RIS. Unlike traditional cellular networks where each cell handles beam alignment in isolation, the CF mMIMO architecture relies on a collaborative approach among APs to direct coherent signals toward users [12, 9]. This collaboration is essential for enhancing spatial multiplexing and reducing inter-cell interference, which is a typical issue in cellular systems with limited inter-cell coordination.
Functioning as a passive device, the RIS cannot generate signals but excels in reflecting them. Consequently, when integrated into the CF mMIMO system, the RIS replaces a subset of APs by reflecting the target signal, enhancing the accuracy of channel estimation through its involvement in pilot signal transmission. However, incorporating RIS into the CF mMIMO system inherently complicates the transmission environment. The RIS requires the APs to estimate the cascaded channels it introduces, which increases the computational burden for channel estimation. Furthermore, the integration of RIS in CF mMIMO systems, with their distributed APs, can heighten synchronization challenges, impacting time and frequency domains. Mismatches in AP frequency synthesizers can cause beam misalignments and heightened interference, critical for coherent signal combining at user terminals [1]. To mitigate these issues, distributed synchronization protocols are crucial, which enable the exchange of timing and phase information across the network, ensuring system-wide synchronization and proper alignment. This synchronization is key to optimizing the performance and reliability of RIS-aided CF mMIMO systems.

Optimizing both the RIS reflection phase and AP beamforming is crucial in RIS-aided wireless communication systems to enhance signal strength and system EE by directing beams accurately to target users. However, in CF mMIMO systems, coordinating beamforming at APs and phase design at RIS presents a unique challenge. The dense AP deployment increases complexity, requiring coordination among distributed APs, resulting in higher computational complexity and signaling overhead compared with other RIS-aided communication systems. Thus, developing advanced joint optimization strategies for RIS-aided CF mMIMO systems is essential yet challenging [3].

In summary, implementing RIS-aided CF mMIMO systems faces three main challenges: RIS-induced interference, beam alignment, and joint optimization. RIS-induced interference complicates power management and affects signal quality. Beam alignment requires precise synchronization between APs and RIS to avoid misalignments. Joint optimization of RIS reflection and AP beamforming requires advanced strategies to manage computational and signaling overhead. Addressing these challenges is essential for enhancing the efficiency and performance of RIS-aided CF mMIMO systems.

Cooperative communication strategies for uplink systems with RIS address interference challenges like mutual coupling and EMI. In this distributed framework, APs, a CPU, and RISs collaborate to enhance communication efficiency and reliability. The CPU collects CSI to encode signals that are precisely tailored to current channel conditions and sends them to the APs. These APs then exploit beamforming techniques like maximum ratio (MR) combining to optimize signal reception. The LSFD method dynamically adjusts transmission power for efficiency. The RIS plays a critical role by reflecting signals to enhance strength and coverage without extra power consumption, addressing challenges like mutual coupling and EMI. This two-layer cooperation for interference elimination stands out for its focus on optimizing EE and managing the complexity introduced by RIS integration [6, 12]. Furthermore, to balance the system overhead and performance in RIS-aided CF mMIMO systems, it is crucial to optimize resource allocation, design cooperative communication protocols, and integrate advanced signal processing techniques. This approach ensures efficient operation by balancing system demands with performance capabilities [9]. Different from [10] which proposed an optimization algorithm that leverages the characteristics of RIS-induced interference, our method emphasizes system-specific optimization, focusing on enhancing signal strength and coverage while reducing interference introduced by RIS integration, making it particularly suitable for RIS-aided CF mMIMO systems.

In contrast to traditional cellular systems that often transmit at full power, leading to inefficient power utilization, our RIS-aided CF mMIMO system employs a fractional power control scheme. This assigns specific power control factors to each UE based on its channel condition, ensuring uniform service quality and consistent communication performance regardless of location or channel state. This approach is particularly effective in RIS-aided systems, where the integration of RIS can further achieve high-capacity communication coverage.

Beam tracking in RIS-aided CF mMIMO systems is essential for high-frequency bands, addressing the challenges of precise alignment due to narrow beams and blockage sensitivity. Our research presents an innovative beam tracking approach tailored to these systems, where RIS units are used in place of some APs to lower energy consumption without sacrificing performance. The passive RIS elements, which reflect but do not process signals, shift the computational load to the APs. To manage this, we’ve developed a beam tracking method that dynamically adjusts to the UE’s movement and optimizes RIS element usage based on environmental changes. This method, which combines precise beam direction tracking with RIS element optimization, is distinctive in the field. We utilize advanced beamforming at the APs, including MR combining, for optimal signal reception. Coupled with algorithms like least mean square (LMS) and bidirectional LMS, these approaches dynamically adjust transmission power, ensuring accurate beam tracking in dynamic conditions [13].

To enhance EE in RIS-aided CF mMIMO systems, our research underscores the importance of joint optimization of AP beamforming and RIS phase shifts. This dual-focus strategy is pivotal for maximizing the reflection performance of the RIS and the beamforming at the APs, which are key variables in the system’s optimization. While power control at the AP contributes to improving EE, our focus lies in the iterative optimization of RIS phase shifts and AP beamforming to enhance signal quality and boost EE. Table I summarizes the characteristics of typical methods and machine learning-based optimization.

1. 1.

   Successive convex approximation (SCA): SCA effectively addresses non-convex problems by solving their convex surrogate problems iteratively, suitable for systems needing regular updates with low computational demand and where objectives and constraints are amenable to local convex approximation [6].

2. 2.

   Riemannian manifold optimization: This method leverages Riemannian manifold geometry to efficiently tackle non-linear constraints such as unit-modulus in RIS phase shift design. It is well-suited for systems with the computational capacity to perform complex manifold operations [9].

3. 3.

   Lagrange optimization: This method employs Lagrange multipliers to address non-convex optimization problems with a mix of equality and inequality constraints, optimized for systems with significant initial conditions through KKT conditions [5].

4. 4.

   Deep deterministic policy gradient (DDPG) optimization: DDPG adopts deep neural networks to train an actor-critic model for continuous control tasks, enabling effective learning and convergence in continuous action spaces, ideal for dynamic environments with changing system parameters [14].

In RIS-aided CF mMIMO systems, cooperative communication enhances signal transmission through AP-CPU-RIS collaboration. Beam tracking ensures precise beam alignment using methods such as Kalman filters. Joint optimization maximizes energy efficiency by addressing RIS phases and AP beamforming, utilizing traditional methods and machine learning methods. Together, these strategies significantly boost system performance.