Identifying and Overcoming Some Operational Limitations of Reconfigurable Intelligent Surfaces in 5G and Beyond Wireless Networks

Abstract

Reconfigurable intelligent surfaces (RIS) can dynamically reshape the propagation environment to enhance signal strength, spectral efficiency and reliability in 5th generation (5G) cellular as well as device to device (D2D) communications. However, to reap such benefits, a range of practical and operational challenges need to be addressed for effectively utilizing RIS in realistic urban environments. This includes maintaining line of sight (LoS) between the RIS and the communicating devices for reliable signal reflection in millimeter wave (mmWave) communication, reducing high channel estimation overhead for communication using multipath rich channels and preventing violation of strict latency constraints due to high complexity of optimal RIS configuration, among others.

We begin by examining the limitations of RIS in mmWave D2D networks, where stationary RISs require a fixed LoS and strategic placement to accommodate user mobility. To overcome this, we first formulate the RIS placement task as a set cover problem and place the fewest possible number of RIS by using a greedy approximation algorithm in a preprocessing stage. Then, once devices are deployed, we select an optimal RIS subset to provide indirect LoS to D2D pairs that have their direct LoS link blocked. Here we have considered that the RIS will be allowed to be placed in any of the desired locations.

Next, the mmWave D2D network scenario where optimal RIS locations may be inaccessible is considered. This is due to third party building ownership, and prohibition from deploying dedicated support structures by regulatory authorities. In these situations, one must leverage the existing RIS deployment which may not be optimal. Moreover, user mobility results in continuous change in RIS-user LoS status. To overcome this challenge, a novel visibility polygon based deterministic RIS selection algorithm is proposed.

Next, the focus of the study moves to 5G cellular networks. In obstacle rich dense urban environments, multipath propagation results in high channel estimation overhead for both sub-$6$ GHz and mmWave channels, leading to outdated channel state information (CSI). Therefore, performing channel estimation at the beginning of every coherence time interval incurs massive pilot overhead. To reduce the pilot overhead, an algorithm is proposed that dynamically schedules the next channel estimation time based on outdated CSI. First, RIS phase shifts are computed based on current CSI. Next, user transmit powers and bandwidth are allocated based on outdated CSI to maximise the aggregate throughput. Using this, the proposed algorithm dynamically adjusts the duration between consecutive channel estimation instants such that pilot overhead is reduced without harming the throughput performance. The primary focus of this study is the enhanced mobile broadband (eMBB) users whose objective is to maximize the throughput.

Pilot overhead and high complexity of RIS configuration increase latency, which is detrimental to the performance of ultra reliable low latency (URLLC) users requiring strict delay constraints. Moreover, mobility of such users results in a frequent change of channel conditions and handovers between networks. To maintain seamless connectivity and strict latency constraints during handovers, a joint base station (BS) and RIS selection algorithm based on contextual multi-armed bandits (C-MAB) has been proposed. The algorithm learns when to take the assistance of RIS while communicating with the BS to maintain latency without losing reliability.

Both resource block (RB) allocation and RIS selection in RIS assisted cellular networks are dependent on the CSI of direct BS-user as well as RIS assisted channels, thereby being affected by channel estimation overhead. This is established by proposing a MAB based RB allocation algorithm to allocate RBs in an RIS assisted network which utilizes both direct and RIS assisted channel information as opposed to using only the direct channel information. After establishing this, an efficient RIS-RB pair allocation algorithm based on adversarial bandit and bipartite graph matching has been proposed for mmWave non orthogonal multiple access networks. This algorithm addresses the challenge of high CSI overhead to find the optimal RIS in the presence of dynamic obstacles and allocate RBs optimally to user groups to maximize throughput while ensuring that users with worse channel conditions are not ignored.