Reconfigurable Intelligent Surfaces: Performance Assessment Through a System-Level Simulator

Reconfigurable intelligent surfaces (RISs) constitute an emerging transmission technology that is under evaluation for possible integration in future programmable wireless networks. For example, the European Telecommunications Standards Institute (ETSI) has recently launched a focused industry specification group whose objective is to coordinate current research and development efforts on RISs for facilitating the potential adoption of this technology in future telecommunication standards¹¹1https://www.etsi.org/committee/1966-ris..

An RIS is an engineered (intelligent) surface, which is made of electromagnetic metamaterials and is equipped with simple electronic circuits (e.g., diodes or tunable capacitors), that is capable of intentionally and smartly controlling the propagation of the electromagnetic waves in complex wireless propagation environments, so as to boost the signal quality in a cost-effective and energy-efficient manner [1]. The potential benefits of RISs in wireless networks depend on the carrier frequency. In particular, an RIS deployed in the mid frequency spectrum, e.g., the C-band, of fifth-generation (5G) cellular networks is expected to primarily enhance the capacity of cell-edge users, to improve the channel rank of multiple-input multiple-output (MIMO) transmission links, and to suppress the network interference. An RIS deployed in the high frequency spectrum, e.g., the millimeter-wave (mmWave) band, is expected to primarily enhance the signal-to-noise-ratio (SNR) in weak coverage areas and blind zones, by overcoming non-line-of-sight propagation conditions and by enhancing outdoor-to-indoor communications. In this context, an RIS may constitute a cost-effective, smart, reconfigurable, and easy to deploy alternative to relays, which is able to operate in full-duplex mode without self-interference [2].

While the potential advantages and gains of deploying RISs in wireless networks are under discussion and analysis, it is acknowledged that several important design challenges need to be tackled for possibly integrating RISs in future telecommunication standards. These include the definition of electromagnetically consistent models for metasurfaces, as well as the derivation of channel models for RIS-aided links and their integration into ray tracing simulators [3], the development of scalable algorithms for optimizing the operation, the placement and deployment of RISs [4], the design of efficient channel estimation algorithms, and feedback channels for controlling and configuring RISs [5], as well as the ownership and interference management in multi-operator networks. In addition, the identification of the most suitable use cases and deployment scenarios for RISs is currently being discussed. Example of use cases that were recently showcased at the Docomo Open House 2021 include the deployment of transparent RISs for enabling outdoor-to-indoor communications in the mmWave frequency band²²2https://docomo-openhouse.jp/2021/en/exhibition/022/. by using, e.g., transparent metamaterial films without spoiling the appearance of windows³³3https://www.sekisui.co.jp/electronics/en/application/film.html..

Testbed validations, real-world experiments, and field trials are essential in order to demonstrate, in existing 5G network deployments, the actual value of emerging technologies like RISs. In the recent open technical literature, various testbed platforms have been reported. Examples of hardware prototypes include two-state digital surfaces for anomalous reflection in the sub-6 GHz [6] and mmWave frequency [7] bands, four-state digital surfaces for anomalous reflection in the sub-6 GHz and mmWave frequency bands [8], varactor-controlled surfaces with continuous phase shift capabilities [9], refracting surfaces in the mmWave frequency band [10], and omni-coverage surfaces that are endowed with joint reflection and refraction capabilities [11]. By contrast, a limited number of real-world measurement campaigns have been publicly reported. These include the demonstration of an RIS testbed that was deployed by Docomo and Metawave in a real urban environment in Japan for expanding the coverage of outdoor non-line-of-sight links at 28 GHz [12], and a one-bit digitally-controlled RIS prototype that was successfully tested for data transmission over a 500 m communication link [13].

The available testbed prototypes and the field trials reported to date constitute fundamental proof-of-concept evaluations of the potential gains of RISs in realistic but small-scale settings. In order to evaluate the potential benefits of deploying RISs in existing and future wireless networks, large-scale simulations and field trials in realistic propagation environments and network topologies are needed. This article is a first attempt to shed light on the potential performance gains of RISs when they are deployed in a large-scale urban scenario and when they are integrated into an existing 5G cellular network. The simulation study reported in this paper is the first of its kind and relies on the Coffee Grinder Simulator^©, which is a proprietary system-level platform for fast wireless system exploration that is developed and maintained by Huawei Technologies Sweden – Gothenburg Research Center. The aim of our study is to quantify the performance gains offered by RISs, as a function of their size, deployment density, and frequency of operation, when they are deployed in an existing 5G cellular network. Specifically, we analyze and compare the deployment of RISs in the mid frequency band (the C-band at 3.5 GHz) and in the low mmWave frequency band (at 28 GHz). Several numerical results are illustrated and commented. For example, it is shown that the deployment of RISs enhances the coverage of cell-edge users from 77% to 95% at 3.5 GHz and from 46% to 95% at 28 GHz.

To quantify the performance gains offered by the deployment of RISs in a typical urban environment that is served by a 5G network infrastructure, we employ the Coffee Grinder Simulator^©. Coffee Grinder is a proprietary simulation platform for fast wireless systems exploration that is developed and maintained by Huawei Technologies Sweden – Gothenburg Research Center. Coffee Grinder is an advanced link-level and system-level simulator that is designed for analyzing low and high frequency bands. The simulation platform is characterized by the agility of configuration in space and runtime, as well as the visualization of results.

Coffee Grinder provides realistic urban and rural 5G and beyond 5G network deployments. The simulator integrates the OpenStreetMap API for modeling buildings, streets, vehicles, and it supports map-based 5G channel models based on the METIS 5G channel model, which includes vegetation and atmospheric attenuation models. Hybrid three-dimensional ray tracing models are supported based on the sparse point cloud method, i.e., a deterministic map-based model is employed for modeling surfaces and corners, and environmental small scale features are included through a diffuse scattering model.

The propagation channels are calculated at discrete frequency bins and all components of the channel model are computed for all frequency bins, in order to obtain realistic and accurate results that account for the frequency dependency (e.g., antenna elements, beam squint, material properties, Doppler shift and spread, Fresnel zones, rain attenuation). The ray bundles are explored for all candidate routes and each ray is evaluated by applying Jones’s calculus to the antenna, reflection, diffraction, penetration, and scattering. Several material models are supported, including those in the METIS channel model as well as custom materials. The antenna models of the base station (BS) and user equipment (UE) are fully customizable. High-level physical layer algorithms are supported as well.

In Fig. 1, as an example, we illustrate the grid-based representation of the potential positions of the UEs in an urban environment, as well as an example of the ray bundles that can account for thousands of paths. As far as this article is concerned, the most important feature of the Coffee Grinder Simulator^© is the support of realistic models for RISs, which include anaomalous reflections and the calculation of their locations for an optimized deployment. Coffee Grinder was initially designed in Matlab and it was subsequently migrated to Python with a GPU acceleration support (Nvidia Optics).

In this section, we describe the network simulation environment, the simulation setup, and how the RISs are modeled and incorporated into the system-level simulator.

In Table I, we report the simulation setup. The considered network scenario corresponds to an urban area in the city of Hangzhou, China, which is identified by the geographical coordinates (30° 16’ 58.8” N, 120° 9’ 21.6” E). The area is covered by BSs that are located according to a real life grid. The number of BSs and RISs depends on the operating frequency. In this study, two typical carrier frequencies for 5G networks are analyzed: the C-band at 3.5 GHz (mid frequency band) and the mmWave band at 28 GHz (high frequency band). The focus of our analysis is on downlink (DL) transmissions. A similar analysis for the uplink (UL) can be performed. It is worth mentioning that the DL-UL reciprocity may not necessarily hold in RIS-aided channels, and it depends on the specific characteristics of the RISs [6]. The impact of the DL-UL reciprocity in RIS-aided networks is postponed to a further research work.

To analyze the improvement in terms of coverage and rate, we compute coverage maps across the entire considered geographic area, where the potential UEs may be located in indoors and outdoors. The UEs are assumed to be located on a grid-based layout with a density of 1 UE per 100 m². As far as the locations of the RISs are concerned, we have considered thousands of different three-dimensional candidate locations that are tested for improving the outdoor and indoor coverage. We assume that multiple RISs can assist the transmission of each UE. To better analyze the gains offered by deploying RISs, the BSs are assumed not to be equipped with massive MIMO antennas. The exact array gains of the BSs and UEs that are considered for link budget calculations are reported in Table I. The use of massive MIMO antennas is expected to further improve the coverage and rate. Specifically, a UE is typically considered to be in coverage if the received SNR is greater than 0 dB. Besides this typical case study, we consider an SNR-boosted scenario in which the UEs are in coverage if the SNR is greater than 10 dB for each considered frequency bandwidth.

As far as the design of the RISs is concerned, we consider a typical far-field assumption. Therefore, we assume that the RIS is configured as an anomalous reflector and that it introduces a linear phase shift that depends on the difference between the angle of incidence and the desired angle of reflection. This is the typical geometrical optics approximation [3, Eq. (94)] with unit amplitude constraint. It is known that the configuration of an RIS as an anomalous mirror is suboptimal in the near-field region of the RIS. However, it offers asymptotically the same performance as the optimal design (focusing lens) in the far-field region of the RIS [14]. From the practical point of view, deploying RISs that are configured as anomalous mirrors is advantageous since it is easier to estimate the directions of the UEs as compared with their exact locations. The accuracy of the far-field approximation is further discussed in the sequel (see Fig. 4). The RISs are, in addition, designed according to a local design and no average power conservation is enforced, which usually results in local power amplifications. These assumptions result in a sort of “worst case” design for the RISs, which allows us to better understand the fundamental gains that the RISs can offer in a realistic 5G cellular network deployment.

As an illustrative example, Fig. 1 shows the SNR coverage maps that are obtained with the Coffee Grinder Simulator^© when the carrier frequency is 28 GHz. The maps are obtained by assuming that the SNR target for the cell-edge (5^th percentile) UEs is 10 dB. We see that 46% of the UEs is in coverage (or 77% if the SNR target is only 0 dB) in the absence of RISs, while 95% of the UEs is in coverage when all the RISs in the network are utilized. It is apparent, therefore, that the deployment of RISs is beneficial for boosting the SNR of weak coverage areas and blind spots, which in turn enhances the coverage probability and the network spectral efficiency.

In this section, we report the system-level simulation results obtained with the Coffee Grinder Simulator^© in the RIS-assisted 5G urban network scenario illustrated in Fig. 1. The system-level performance evaluation is conducted in terms of outdoor/indoor coverage and ergodic rate at different carrier frequencies and for different sizes of the RISs.

Specifically, the key performance indicator of interest is the cumulative distribution function (CDF) of the SNR received by the UEs in the absence and in the presence of RISs, i.e., the outage probability. Specifically, the complementary CDF (CCDF) represents the fraction (or percentage) of locations of the UEs where the received SNR is greater than the predefined target SNR threshold, i.e., the UE at the specified location is in coverage. From the obtained curves of CDF and CCDF, the performance gains of deploying RISs are quantified. The following two sections report the results for a 5G network deployment in the C-band and in the mmWave frequency range, respectively. Special focus is put on the impact of the size of the RISs at different carrier frequencies.

RIS is an emerging technology for improving the coverage and the rate of wireless communication systems without requiring active power amplifiers, multiple radio frequency chains, and complex digital signal processing units. In this paper, for the first time in the open technical literature, we have reported a system-level performance evaluation study of an RIS-aided 5G network that is deployed in a typical urban environment for operation at 3.5 GHz (C-band) and 28 GHz (mmWave). The study is performed by utilizing the Coffee Grinder Simulator^© that integrates a specifically designed algorithm for optimizing the locations of the RISs throughout the network, as a function of their size, operating frequency, and reflection capabilities. In particular, the RISs are modeled as anomalous reflectors for serving UEs located in the far-field region.

The benefits of adding RISs have been evaluated in terms of coverage probability and ergodic rate for different user percentiles, which include the typical UEs (the 50^th percentile of UEs) and the cell-edge UEs (the 5^th percentile of UEs), by assuming that a UE is in coverage if the received SNR is greater than 10 dB in order to create coverage-boosted areas. Overall, the system-level simulations reveal that the deployment of RISs enhances the coverage of the cell-edge UEs from 77% to 95% in the C-band and from 46% to 95% in the mmWave band. The deployment of only 5 RISs per BS whose size is 3.8 m x 3.8 m in the C-band and 0.67 m x 0.67 m in the mmWave band ensures that 90% of cell-edge UEs are in coverage. If 12 RISs per BS of the same size are deployed, the ergodic rate of the cell-edge UEs increases by a factor of 3.5 and by a factor of 25 in the C-band and mmWave band, respectively. In the mmWave band, in addition, the SNR of the cell-edge UEs increases by 20 dB if, for each BS, 20 RISs whose size is 0.33 m x 0.33 m or 8 RISs whose size is 0.67 m x 0.67 m are deployed.

These results highlight the potential benefits of deploying RISs in a typical 5G urban network. Additional performance gains may be obtained by amalgamating RISs with other technologies, such as massive MIMO, and by considering more complex implementations of RISs that go beyond the operation as anomalous reflectors.