Evaluation of Beamsteering Performance in MultiuserMIMO Unmanned Aerial Base Stations Networks

Future wireless communication networks can benefit from Unmanned Aerial Base Stations (UABSs) to provide enhanced capacity to ground users (GU) in large and remote locations. Connecting UABSs to the terrestrial network presents several challenges, such as the limited gain traditional antennas need to maintain suitable wireless links between the core network, UABS, and GU. A convenient solution is to use MaMIMO (Massive Multiple Input Multiple Output) since it improves spectral and energy efficiency, thus providing high data rates while reducing power consumption. This paper proposes a Multi-user MIMO (MuMIMO) model for UABS aided networks to increase service range and served capacity. It uses hybrid beamforming and beamsteering on Terrestrial Base Stations (TBSs) and UABSs to grant access to mobile GU in a bicycle race scenario. Results show that using the mobile operators’ locations will benefit the backhaul network’s performance by duplicating the capacity compared to using a private operator. Furthermore, user coverage increases by 400% if MuMIMO is used, compared to a single beam MaMIMO network. The proposed scenario could achieve a channel efficiency of 6.5 bit/s/Hz in the access network and 8.6 bit/s/Hz in the backhaul network. Finally, the average UABS transmitted power is reduced by 2/3, increasing the number of used beams.

the aerial network. Previous work proposed some solutions to mitigate the effects like optimising the UAV altitude, using more directive transmission, or optimising path planning for the UAV [4]- [6]. The research community is actively trying to address these concerns. A promising enabling technology for UABSs is Massive Multiple Input Multiple Output (MaMIMO) due to its scalability and potential to deliver high and stable throughputs. In a MaMIMO cellular system, base stations are equipped with many antennas simultaneously serving multiple single-antenna terminals. By using beamforming, the power is focused into a small region of space, thus reducing interference and improving energy efficiency [9], [10]. Enabling digital beamforming leads to the use of Spatial Duplexing Multiple Access (SDMA) and Multi-user MIMO (MuMIMO). Due to this unique set of features, MaMIMO and MuMIMO attracted significant attention from the wireless and UAV communities, [7], [10]- [15] with the limited attention paid to backhaul links [16], [17].
In our previous work [18], the UABS is connected to the Core Network (CN) using a facility in the supported area centre. A facility was composed of a crane with a high antenna connected to the CN and a truck that stores the UABS for future service. However, these fast deployable facilities raise several connectivity challenges for larger areas. As the UAV's battery lifetime increases, the UABS can travel over a more considerable distance; hence a higher communication range of the backhaul (BH) network is required. Considering TBS and using them to connect UABS will benefit the connectivity of the BH link. Using the sub-6 GHz band for backhaul connection will require a very high spectrum efficiency to achieve significant bit rates due to the limited bandwidth available in this band. As a result, the main challenge here is to optimise the channel efficiency of the backhaul part through the optimal placement of UABS and the maximisation of the antenna gains.
Using technologies like Massive MIMO (MaMIMO) in both the access and backhaul links for UABS will provide several advantages. By using beamforming with 256-antenna elements, the antenna gain could increase up to 24 dB, increasing the UABS connectivity range up to 8 times, thus extending the coverage of UABS-aided networks. In addition, the capacity will be increased too because, for higher antenna gains, higher SNR values at the receiver will be found, resulting in better Modulation and Coding Schemes (MCS) that support higher bit rates [19].

A. CONTRIBUTIONS
We consider a UABS-aided cellular network, where MuMIMO techniques are used at the access (UABS -user) and backhaul links (TBS -UABS). UABSs are mobile so that the network can be dynamically densified as it is envisioned for future beyond-5G and 6G networks. To the best of the authors' knowledge, MuMIMO evaluation on long-range UABS-aided networks, accounting for the access and BH network in realistic scenarios, has not been done yet.
The major contribution of this work is the performance evaluation of the beamforming for MuMIMO with movable equipment mounted on UABS. To this end, we upgrade the capacity-based tool from [18] to include realistic users' mobility and MuMIMO physical layer capabilities through the usage of MaMIMO technology. The simulator takes into account several real factors such as actual 3D map, semi-deterministic channel models, ground users' movable locations and their traffic demands, the legal beam power limitation, realistic location and parameters of terrestrial infrastructure, and drone performance; to optimally allocate UABS and ground users under realistic constraints. Using our simulator, we answer the following questions: • Is the UABS range extended by using MuMIMO? • Is the capacity of the UABS-aided network increased by using MuMIMO?

B. PAPER ORGANISATION
This paper is organised as follows. Section II-B describes the related work reviewing the use of beamforming in UABS networks. Section III-G presents the network architecture and defines the characteristics of the considered MuMIMO technology. In Section IV-C, we describe the methodology with particular attention paid to the scenario definition, the simulation tool, and the evaluation parameters used in this work. Section V-D2 discusses the results and, finally, in Section VI the paper is concluded, and the directions of the future work are given.

II. BEAMFORMING IN UABS NETWORKS
This section describes the recent works in UABS networks and beamforming technology that can support vast bandwidths on beyond 5G networks (B5G).

A. UNMANNED AERIAL BASE STATIONS
Several reviews and tutorials [2], [20]- [22] have described the benefits of using UABSs in 5G networks. Lin et al. [22] present a review of the cellular communications challenges for UAVs, such as the aerial channel and the antenna patterns for 3D communications. However, their approach only evaluates 8 × 1 antenna arrays, and vertical down-tilted is not compensated. Mozaffari et al. [2], [21], [23] introduced the concept of a 3D wireless network. They provide the mathematical background for the user-UAV association in this type of network and discuss the importance of beamforming to maximise the spectral efficiency of the network. They propose an antenna array based on multiple drones and beam-steering capabilities through the location and spacing of the drones. In addition, our work [20] offers a holistic overview of 3D wireless communication networks. We describe channel models relevant for simulation-based works and provide guidelines on their appropriate use. Additionally, we summarise state of the art related to UAVs as aerial users and UAVs as UABS. The content of [20] is based on analytical, simulation, and measurement-based publications.
In [24], [25] Xiao, et al. present a review of the scenarios and challenges for UAV mmWave communications with beamforming capabilities. It covers the problem of the antenna misalignment in the backbone link for UABS, which reduced the antenna gain. Results show that for Hybrid beamforming (HBF) with SDMA, the achievable channel efficiency could be higher than 16 bit/s/Hz for a Signal to Noise Ratio (SNR) of 30 dB. Our previous work [5] compares the coverage of existing cellular networks in sub-6 GHz (1.8 GHz) to upcoming 5G mmWave (38 GHz) frequencies.
Through semi-deterministic simulations, we show that UAVs flying at an altitude just above rooftop level have perfect coverage. On the other hand, UAVs operating at more significant altitudes suffer from large interference levels, resulting in UAVs not having any coverage at high altitudes. To tackle this problem, we show that using more directive links like beamforming in the mmWave spectrum will essentially solve these issues since directive transmission reduces the interference towards other flying equipment. Previously in [7], we concluded that mmWave beamforming could also provide a solution to the existing handover problems at significant altitudes due to fragmented assignment behaviour in existing networks.

B. MASSIVE MIMO BEAMFORMING AND BEAM MANAGEMENT ON UABS
Several authors are studying the advantages of using beamforming for UABSs. These studies can be divided into two groups based on the considered link type: (i) in the backhaul (between the UABS and the core network) and (ii) in the access network (between the UABS and the ground users).
First, we discuss the studies related to backhaul connectivity. In [26], the authors propose an algorithm that performs 3D tracking of the pencil beams connected to UAV. The algorithm predicts the angular speed and determines the direction of future frame transmissions. The authors of [27] developed an optimised beam stabilisation algorithm suitable for a relatively slight misalignment of unstable beams. In [28], a beamforming model is used for UAVs serving as User Equipment (UE) to maximise the Up-Link (UL) throughput while minimising the interference between traditional terrestrial and the UAV users. The results show that the capacity can be up to five times higher when using beamforming than an omnidirectional antenna system. The authors of [15] discuss the benefits of a multi-beam UL in a UAV to reduce the interference of ground BS and maximise the UL sum rate by using the BH as a link to establish a cooperative environment between ground BS. In [29], the authors explore the problems that arise from pilot contamination in MaMIMO for aerial users and propose a decontamination algorithm to minimise the channel misalignment for the UAV and ground BS.
Second, we discuss the studies considering beamforming in the access network. The authors of [30] introduce Beam Division Multiple Access (BDMA) to provide multiple access to ground users from aerial BSs in mmWave frequencies. However, they have found that the number of beams is limited for ultra-dense networks, restraining the network's performance. Next, Halvarsson et al. in [19], [31] present the results of testbeds for beamforming management in the 3.5 GHz, demonstrating bandwidth efficiency of 8.75 bit/s/Hz for LoS links and 2.5 bit/s/Hz in NLoS links for the 3.5 GHz band.
Finally, recent works have focused on the advantages of combining the access and backhaul network under the same network. This technology is called In-band integrated Access and Backhaul (IAB) systems. It explores the advantages of multi hoping to extend the ranges and capacity of the backhaul to the cost of additional overheads, multi-link backhaul management, and complex interference management protocols. [16], [17], [32] Although several authors have researched the benefits of MuMIMO beamforming, in state of the art, there is no joint allocation mechanism for access and BH links using beamforming and beamsteering. In the following section, we propose a network architecture with an allocation mechanism to serve ground users with MuMIMO in UABS.

III. NETWORK ARCHITECTURE FOR UABS
The following section describes in detail the network architecture for UABS. First, the access and BH network are described, then the details of the antenna and channel models are depicted, and finally, the parameter of the drones are presented. Details of the notations of these models can be found in Table 1.

A. RADIO ACCESS AND BH NETWORKS ARCHITECTURE
We consider a UABS network to provide wireless communication to ground users with video traffic requirements under VOLUME 10, 2022 mobile conditions. The network architecture is shown in Fig. 1. It consists of the access network that allows ground users to be served by aerial Base Stations and the backhaul network to enable UABS to be connected to the core network through terrestrial Base Stations. This architecture is based on our previous work [18], [33], [34] with two main improvements. First, the system model considers now the dynamic allocation of ground users based on realistic movements (differing from the fixed random location). Secondly, we include a complete physical model for the Massive MIMO and Multi-user MIMO, including the 3D geometry of the UAV system that consists of the Uniform Planar Rectangular (UPR) array antennas.
The Radio Access Network (RAN) provides access to the mobile ground users (GU) through the UABSs, as shown by the orange links (horizontal patterns) in Fig. 1. On the other hand, the BH network (blue links (vertical patterns)) connects each UABS with the core network in Fig. 1). As shown in Fig. 1, the BH network consists of terrestrial base stations (TBS). The UABS typically connects to the TBS, from which it experiences the lowest path loss. We assume intra-cell handover and frequency management are internally done on the TBS, and details of the allocation procedures will be described in the methodology section.

B. MULTI-USER MIMO SYSTEM
In order to serve multi-user using MIMO technology, MuMIMO beamforming has been implemented. Fig 2 illustrates the MuMIMO hybrid beamforming model used to serve multiple beams simultaneously, based on the BDMA method in [30], [35], where diverse beams are steered based on a specific code for each beam set. The signal streams for each beam are Regularized Zero Force (RZF) precoded in the baseband and further transmitted to the RF chains. RZF precoding is selected to reduce the intra-beam interference based on the knowledge of the Channel State Information (CSI) [36]- [39]. Each RF chain is in charge of each beam, so the number of RF chains is determined by the maximum number of supported beams, in our case, 8 beams [40]. Then, the phase shifters are controlled by a code-book selector (CBS) in charge of steering to the desired location of each beam based on pilots and CSI values. Here, the hybrid beamforming provides digital precoding to minimise the intra-beam interference based on the RZF, while the analogue precoding, controlled by the CBS, is in charge of the steering process. The CBS controls the division of the total number of antennas of the antenna array into sub-arrays with an equal number of antennas to maintain equal maximum gain among the different beams; in this way, the same beamforming model works for 1, 2, 4 or 8 beams as controlled by the CBS [41], [42]. We assume that our CBS plans a jointly coordinated allocation method among the UABS to reduce interference by selecting the appropriate code-book for each beam in a coordinated fashion [44]. Finally, based on previous simulations obtained in our tool, we added an Interference MuMIMO Margin of 2 dB to our link budget to account for other interference values not accounted for [43].

C. UNIFORM PLANAR RECTANGULAR ARRAY ANTENNAS
The antennas of the TBS and on-board of the UABS are described here. Each TBS is provided with a 3-sector antenna to support 360-degree coverage. Each sector has a Uniform Planar Rectangular array (UPR). The UPR contains an M g * N g Cross-polarised panel. M g represents the number of panels on the columns (x), while N g is the number of panels on the rows (y). We will evaluate the performance of the 8 × 8, 16 × 16, and 32 × 8 array sizes for the BH link. As described in [44], [45], these antenna panels are uniformly spaced in both directions and dual-polarised. The 3D radiation pattern of each antenna is constructed based on the array's gain using spherical angles on the Cartesian coordinates. The antenna gain A g is determined by (1) [44], [45]: where AF is the antenna array factor in dB as a function of the elevation θ and the azimuth φ. EG in dB, is the element gain of the antenna and defined as follows: where A (θ, φ = 0 • ) and A (θ = 90 • , φ) are the vertical and horizontal cut of the radiation power patterns in dB, θ ∈ , θ 3dB is the aperture angle of the antenna and is set to 65 • , A min is set to -30 dB and EG max is the maximum directional gain of the antenna element.
The Array Factor AF in dB is the logarithmic function of the product of the array factor in the x-axis (AF x1 ) and the y-axis (AF 1y ). I m1 = I 1n = I 0 is the excitation amplitude of the elements, and for a UPR, they are constant. β x and β y are the progressive phases of the array. Denoted as d, is the distance between the radiation elements and is defined as d = λ/2 and k = 2π/λ. If the beams are steered, θ and φ will be pointed towards the direction of the desired steering determined by the code-book; otherwise, θ = θ and φ = φ.
Multibeam forming is achieved by dividing the UPR into sub-arrays containing an equal number of antenna elements. As shown in Fig 2, a 16 × 16 antenna array is divided into 8 beams with 8 sub-arrays of 4×8 elements and equations (1) to (9) will be applied for each beam.
The UABS will carry a similar UPR antenna system for the access and backhaul networks. While the backhaul antenna array will be pointed towards the horizon, the access antenna array will be pointed downwards. We also assume that the placement of these antennas does not affect the flying capacities of the drone. However, this system will be limited in size for practical reasons, i.e., 0.68 × 0.68 m for the 16 × 16 elements and 0.34×0.34 m for the 8×8 elements. Larger antenna sizes for the sub-6 GHz band will be impractical on the UAV. Using 64 and 256 antennas leads to channel hardening in reducing the impact of small-scale fading and spatial correlation. This is because the collinearity of the channel matrix for these antenna sizes disappears for inter-user distances of 20 m and 10 m, respectively [46], [47]. A coordinate conversion for the angles needs to be implemented as the access antennas will be pointed downwards.

D. MISALIGNMENT MANAGEMENT
Due to the nature of the moving environment of the UABS, some assumptions are made regarding the misalignment problems. First, the digital beamforming uses a resolution of TABLE 2. Link budget parameters for access and backhaul networks. [18], [33], [50], [51].
1 degree in each axe, meaning that the code-book used to steer the beam in the UPR is a matrix of 130 by 130 phased signals. This also means that if the link angle (denoted by θ and φ) is in the middle of two possible digitally steered beams (denoted by θ and φ ), the gain assigned to this link will be calculated according to equations (5) to (9). Second, we assume that pilots can beam track the locations of the movable equipment in each timestamp. Based on [7], [26], we assume a 3D beamtracking reference model that estimates the angular speeds of the beams through a dynamic pilot insertion which can maintain 95% beamforming gain while increasing only 0.1% pilot overhead. This model also includes a fusion sensor that aids in long-range and fast-moving systems [48].
In addition, we assume that the weather conditions of the environment do not affect the ability of the UAV to maintain a stable beam and that the maximum angular speed of the movable nodes is smaller than the maximum shifting delay of the digital beams, which is around 10 ms [49].

E. FREQUENCY AND CHANNEL MODELS
The following section describes the frequency and channel models used in the backhaul and access links. For this network, we propose a narrow bandwidth scenario. This means access and backhaul channels are limited, i.e. 5 MHz for access and 20 MHz for BH in sub-6 GHz bands.

1) BACKHAUL NETWORK
For the backhaul links, we evaluate two bands, the 3.5 GHz and the 60 GHz. First, we assess the 3.5 GHz band, which has the best compromise between coverage and capacity in lineof-sight (LoS) and non-line-of-sight (NLoS) environments, supporting the easy operation of the network. However, the limited offered spectrum, larger antennas, and the need for efficient interference management are significant disadvantages for this band [18], [52], [53]. Current 5G trials have shown up to 10.4 Gbps data rate peaks using 200 MHz bandwidth on this frequency, demonstrating this band's potential in fulfilling the backhaul requirements [54]. Second, we study the 60 GHz band for having massive available bandwidth  [18], [33], [50], [51].
that will overcome the high path losses of these frequencies. This band is also suitable for UABS applications because it has more than 9 GHz of unlicensed spectrum that could be easily used and deployed for diverse sports events or other infrequent situations. The channel model used for the backhaul link is the 3GPP TR 36.777 path loss model [55] defined for connectivity between TBS and aerial vehicles.

2) ACCESS NETWORK
Compared with the BH links, we only evaluate the 2.6 GHz band for the access link. We used a different frequency band for the access network to limit the interference between both networks. Like the 3.5 GHz band, the 2.6 GHz presents a good balance between coverage and capacity. This band also performs better in NLoS communications, which is very useful for the access network where ground users could be covered by buildings or inside of them. We assume a Time Division Duplexing (TDD) network for 5G technologies with 25% for DL and 75% for the UL in both links.
In addition, we selected Al-Hourani's Air2Ground path loss model for our access link [56], which was modelled based on ray tracing urban and suburban scenarios and tuned for our selected frequencies.
Shadowing fading has a substantial impact on the MIMO channel. We select the standard deviation for the shadowing fading based on the frequency and flying altitude of the UABS described in Table B [44] standards. These are based on Rayleigh distribution and have proven to provide a tangible impact on the fading channels for MIMO [9].
In addition, the LoS/NLoS calculation is realised by a Ray tracing where the obstruction is calculated based on the environment and the obstacles of the direct link between transmitter and receiver. We use the approach of [57] to realise this.
Finally, the radio link budget used for the proposed network is enclosed in Table 2, with detailed emphasis on the Receiver SNR for different Modulation and Coding Scheme (MCS) within diverse technologies (See Table 3) [50], [51].

F. UNMANNED AERIAL VEHICLES
Each UABS comprises BS equipment mounted in a long endurance Harris H4 HL hybrid quadcopter able to support  [18], [33], [58]. nearly 3 hours of flight time. Its payload capacity and a wingspan of 2m are sufficient to carry a large antenna array. Details of the drone parameters used in our architecture are found in Table 4. 1 Also, if drone endurance is insufficient, we consider some facility locations with storage capabilities near the TBS for recharging operations, as shown in Fig. 1. These facilities, close to the initial TBS, can store up to 25 carriers with the possibility of switching and recharging the batteries of drones as they are depleting to keep all of them in the air. The location of the facilities is further studied in this paper based on the scenario selected.

G. POWER CONSUMPTION MODELLING
The total network power consumption P(t) is the total power consumption of the wireless network, including the access and BH networks, at time t: where P A,n (t) and P BH ,n (t) are the access and BH power consumption of each UAV-BS n during time slot t. If UAV-BS n is not active, P (t) a,n and P (t) BH ,n are expressed as zero. The LTE power consumption models (in watt) for the access and BH networks for each UAV-BS n are based on [33] and [59]: where n sec is the number of sectors of the access network, P DSP is the Digital Signal Processor (DSP) power consumption in watts, the A g is the MIMO gain of the transmitter, the P trans is the power used by the transceiver, the P T is the actual radiated power by the antenna and η is the power amplifier efficiency.
To evaluate the transmitted power per beam, the total transmitted power of the antenna is divided by the used beams as described in (13): where P Beam is the power of the evaluated beam in Watts, P T is the transmitted power of the antenna in Watts, and B used is the number of used beams of that particular link.
Finally, the power needed by a UAV-BS for flying is considered constant; it does not vary with weather conditions or weight variances and allows to determine the service time of the specific UABS. i.e. the service time will be determined by the network power consumption. Once a UABS runs out of battery, a new UABS will fly to the position of the ageing UABS, and a seamless handover is applied.

IV. METHODOLOGY: NUMERICAL ANALYSIS AND SIMULATION MODELLING
The following section describes the Tour of Flanders scenario used for the numerical analysis and the java-based tool developed to perform the simulations.

A. SCENARIO DEFINITION
To make this work more tangible, we consider a real scenario of the Tour of Flanders, which is an annual cycling race on Flanders' cobblestone roads. It is one of the major cobbled classics in Flanders, Belgium. The cyclists cover a track of 267 km over both flat terrain and climb stretching from Antwerp to Oudenaarde. 2

1) ARCHITECTURE
We design our scenario to match this cycling race's communication requirements and constraints realistically. The communication link is composed of a backhaul and an access link, where the BH link is between a TBS and the UABS following the cyclists overhead at an average speed of 45 km/h. The access link exists between the UABS and ground users represented by the motorbikes following the cyclists and filming the race from the ground. There is one motorbike cameraman designated for the leader of each participating team, resulting in a total of 25 ground users. Each cameraperson uploads a video stream with an average bit rate of 10 Mbps. We evaluate the scenario every 30 seconds, defined as a timestamp, for 756 timestamps over the whole race duration. The location of the cameraman's bike is determined by the average cyclist speed in each location using the Simulation of Urban Mobility (SUMO) tool 3 and the route map. The UABSs are deployed over the bikes at 80 m height, as suggested in [18], at random locations. Further, a proximity avoidance algorithm restricts the minimum distance between drones to 25 m to prevent collisions.

2) SCENARIOS
Next, we define three scenarios to evaluate the beam steering performance in the UABS network (see Table 5). Scenario 1: ''Single-beam digital beamforming'' considers a single UABS connected to the TBS to explore the digital beamforming capabilities of the BH, such as beamsteering connectivity, frequency, and size of antennas. The No-Beamsteering scenario is based on a 3-sector antenna in 2 Ronde van Vlaanderen -Elite Men: http://www.rondevanvlaanderen. be/en/rvv/elite-men 3 Simulation of Urban Mobility (SUMO): https://www.eclipse.org/sumo/  Table 5.

3) GROUND INFRASTRUCTURE
The locations of the TBS are a critical aspect of the scenario's definition. The area considered covers 65 km from east to west and 55 km south to north, covering the whole race track.
In this area, we use two placement strategies. The first one considers a grid-based private BS allocation, proposing a grid with equally distributed 1, 4, 16, and 64 private locations, as shown in Fig. 3a. The second approach considers the list of 2151 existing Terrestrial BSs from the operators in the Flanders region. The distribution of these BS locations is presented in Fig. 3b. In both figures, the route is depicted with the green dots for each timestamp in the race. The blue, red, purple, and orange markers in Fig. 3a. describe the locations of the grid-based 1, 4, 16, and 64, respectively, while the yellow stars in Fig. 3b. depict the actual locations of the BSs in Flanders.
To understand the impact of the different location strategies, Table 6 lists important distance parameters for each strategy; the last column presents the distance experienced by a drone to a TBS in the farther case. The last column represents the furthest distance from a TBS to a drone following the race. Using the second strategy, the furthest TBS is at a maximum distance of 3.5 km. On the other hand, within the 64 BS grid approach, the maximum experienced distance is doubled compared with the existing BS list approach. Still, it is a feasible solution since the deployment will not depend on an established operator but rather on a private network operator.

B. SIMULATION TOOL
We propose a Java-based simulation tool that comprises network planning capabilities for UABS considering the access VOLUME 10, 2022  and BH links. It is an extension of the capacity-based, energyefficient simulation tool, which takes into consideration 3D realistic maps, semi-deterministic channel models and legal beam power limitations [18], [33], [60]. This version includes the creation of the MuMIMO digital beamforming calculation of the MIMO gain for 5G deployments. The UABSs are dynamically placed based on ground users' requirements and the network's restrictions within the proposed scenarios in the tool. The allocation procedures are done in a centralised way and are located in the facility entity. This assumption includes perfect knowledge of the location of UABSs, ground users, and its traffic requirements. Moreover, the mobility aspect of the tool is achieved in a time stamp fashion, which is defined as a period in time of each evaluation of the network. Fig. 4, presents the flow diagram algorithm to perform the analysis of the MuMIMO. It is divided into four parts: 1) Initialises the simulation algorithm.
2) Describes the Access and BH allocation process.
3) Defines the mobile parameters implemented. 4) Collects the statistics for post-processing analysis.

1) PART 1: SIMULATION INITIALISATION
The algorithm starts by defining the three-dimensional locations of the considered area, as shown in Fig. 3. This includes the locations of the buildings in the area to evaluate whether a link is LoS or NLoS. Next, all the ground users' locations and traffic requirements are defined. Next, the possible locations list of the UABSs is created following the altitude and safety distance parameters described in the previous section. This information is used to create data for more than 6 hours of locations and requirements for all the network elements.

2) PART 2: ALLOCATION PROCEDURES
This part of the algorithm will be repeated for each timestamp of the simulation and is responsible for the design of the access network (orange steps) and the backhaul network (blue steps). First, the algorithm will select a random user and calculate the path loss (PL) to each possible UABS location (green steps). Second, the algorithm will order the list of the possible UABSs according to this PL value by starting with the active UABSs and then the inactive ones. An active UABS is the one that has already a viable BH link and could support at least one ground user. On the other hand, an inactive UABS is the one that could have a possible BH nut no users have been assigned to it. The algorithm will now select the first UABS from this list (orange steps). Depending on whether the UABS is active, the tool will take the following actions: 1) If the UABS is not active, the most appropriate TBS needs to be determined (blue steps). The algorithm will select the closest TBS and identifies if it was already used. Again we have two options: I The TBS is already in use: The algorithm will determine if some backhaul beams are still available. If this is the case, the initialisation of a new BH beam can start as follows: i) The total transmitted power of the antenna is calculated. ii) The azimuth and elevation angles of the UABS are calculated. iii) The exact MIMO gain is calculated. iv) The Path Loss is calculated. If this path loss is below the maximum allowable path loss, the UABS can be connected to this TBS. If no beams are available or the BH is not supported, the next UABS of the list is selected. II The TBS is not in use: The algorithm initialises the TBS, and the initialisation of a new BH beam begins. 2) If the UABS is already active, we have to evaluate if there are available beams on the access network (orange steps): I If no beams are available, the next UABS of the list is selected. II If there are available beams, the UABS initialises a new beam in the access network following similar procedures to the one in the BH network. III Then, an evaluation of the capacity is performed. 3) After the access and BH beams are initialised, the UABS proceeds to jointly assess if the capacity of both links could be served with the actual parameters by the following actions: I Increase the antenna's input power (P Antenna ) until the link is served or the power reaches the maximum limit. II If the maximum limit is reached in some of the access or the BH antenna, the input power is returned to the previous values, and the next UABS is evaluated. III The SNR is used to determine the MCS. It is used with the requested bit rate to calculate the RB [18], [34] and add them to the aggregated BH link. IV If both links could support the requested capacity, this is added to both links, and the user is allocated to that UABS. 4) Finally, an optimisation process of the other links is done to maximise the network's capacity. The allocation algorithm evaluates if UABS with one single user could transfer this user to the next UABS in the PL list. This is done by increasing the transmitted power, providing better MCS that reduces the number of required resource blocks in each link.

3) PART 3: MOBILITY
This part of the algorithm updates the location of the ground users based on the selected scenario and the UABS accordingly, preparing for the next timestamp. In addition, the list of active UABSs is evaluated, looking for UABSs that need battery replacement/recharging.

4) PART 4: POST-PROCESSING
In this final part, the algorithm collects the results and metrics of all the timestamps, users, and UABSs and logs them in CSV files for future post-processing and analysis.

C. SIMULATION METHODOLOGY
To achieve confident results in our simulation, we investigate the stability of the simulations. It is reached when the average value of the variables in a specific simulation has a deviation of ± 0.5 % of the average value over 100 simulations. This result is obtained with 40 simulations per scenario. Hence, for each simulation configuration, 40 independent simulations are executed. For each scenario, a series of parameters are varied to evaluate the performance of each scenario. In scenario 1, we modified the beamsteering type (No beamsteering v.s beamsteering following the UAV), the number of Terrestrial BS, and the central frequency (3.5 GHz and 6 GHz). In scenarios 2 and 3, we modified the number of requested users and the maximum beams in the MuMIMO link. For the last scenarios, we evaluate the coverage and capacity of the network as explained in detail in Sec V-B and Sec V-C.
Finally, the whole bicycle race has 6 hours and 20 min (756 timestamps). Numerical results present the 95 th percentile (p 95 ) over these 756 timestamps if not stated otherwise.

V. SIMULATION RESULTS AND DISCUSSION
The following Section provides insights into the three proposed scenarios and discusses the main findings. A general overview of scenarios is given in Table5.  Fig. 5b depicts the BH network resource block (RB) usage (p 95 ), where the number of resource blocks is inverse proportional to the number of available TBS. The more available TBS, the closer it is to the UABS, as seen in Table6, resulting in lower PL values. As a result, using the operators' TBS list, the use of RBs is four times fewer RBs compared to only one private TBS. Also, when using the beamsteering technique, 25% fewer RBs are used compared to no beamsteering. For the 16 and 64 private TBS, a reduction of 65% and 45% in the RB usage can be seen.

3) 60 GHz EVALUATION
We evaluate the performance of the 60 GHz frequency compared with the 3.5 GHz., as shown in Fig. 6. The 60 GHz band performs poorly compared with the 3.5 GHz and cannot provide 100% connectivity as described by the p 95 , even when using a 16 × 16 antenna system and the operators' TBS list. The beamsteering technique outperforms the no beamsteering by more than three times (94% versus 25%), still not sufficient for 100% connectivity. This happens because the maximum antenna power in 3.5 GHz is 6 dB higher than 60 GHz. Moreover, the bandwidth is 50 times larger, raising the noise floor by 17 dB for 60 GHz, while the minimum SNR to achieve connection is 7 dB higher, leading to a difference of 30 dB for the minimum requirements for establishing the connection. In addition to the differences in the path loss due to the frequency, this leads to the poor performance of the 60 GHz band. To solve this problem, higher antenna power, improvements in the sensitivity of the MCSs, and better channelling for this band will improve the mmWave band's performance.

4) MULTI GROUND USERS' CONNECTIVITY
We include various ground users in the network to see the capacity in the single beam network with beamsteering and 16 × 16 antenna size. As explained in SectionIV-A, we increase the number of camerapeople motorbikes from 1 to 25. Results show that a single drone could support up to seven ground users in the best-case scenario (yellow line). This corresponds to the maximum capacity of a single BH link which adds up 72 Mbps (p 95 ) when using the best MCS, as shown in Fig. 7. For this case, the BH is the network's limitation since we assume no limit in the access network capacity. Also, the operators' TBS list usage doubles the served capacity for the multi-user scenario

5) SERVICE TIME
The average service time per carrier depends on the transmitted power of the UAV. For beam steering cases, the average service time is nearly 150 minutes (p 95 ) and will increment slightly depending on the availability of terrestrial infrastructure. For the cases of 16 TBS, 64 TBS and operator TBS list, the increment is 2%, 5% and 8% compared with the 1 and 4 TBS cases. However, three carriers are needed to support the whole event.

B. SCENARIO 2: MULTIBEAM BEAMFORMING IN THE BH NETWORK
In this second scenario, Multibeam beamforming in the BH network is evaluated. Here, we assume that the element gain of the MaMIMO antenna is the same in every sector. In contrast, the antenna transmission power is divided equally to each beam as described in (13).  We evaluate the performance of the maximum number of MIMO layers that expresses the maximum number of simultaneous beams over a MuMIMO antenna. In this case, we evaluate 1, 2, 4, and 8-layers; the latter was considered for future improvements of the UL, suitable for supporting 2 code words [61]. The following results are collected assuming the best performance case for the BH network, i.e., a 16 × 16 antenna system placed on the UABS to serve the ground users combined with the TBS allocation based on the operator BS list. Fig. 8a shows the provided user coverage (p 95 ) for the Multibeam in the BH network. The lines represent the absolute number of served ground users while the boxes the percentage of served ground users. Blue, green, yellow, and orange represent the evaluated 1, 2, 4, and 8-layers, respectively. When Multi-beamforming is applied, the number of served ground users increases from 14% to 39% for 25 simultaneous ground users, as shown in the blue and orange boxes of Fig. 8a. Full coverage can only be provided to up to five ground users when using eight beams. One might expect ground users to be covered eight times more when using eight beams than one, but this is unfortunately not the case. This is because the number of served ground users per UABS is reduced when the number of beams increases from 3.6 users/UABS (blue line in Fig. 8b) down to 2.6 users/UABS (orange line in Fig. 8b). This is due to the division of the beam power in the BH that reduces the link's supported power, resulting in a limitation of the BH capacity per UABS. Furthermore, the capacity of the access network is a limiting factor here since the maximum bitrate for our narrow band 5G access network is 42 Mbps per UABS, so only each UABS can serve 4 ground users requesting 10 Mbps.

1) COVERAGE IN SCENARIO 2
When using 1 and 2 beams, the number of UABSs that can be served simultaneously is the limiting factor: each beam can serve only 1 UABS. This is, however, no longer the case when increasing the number of supported beams to 4 and 8. In this case, the power per beam is the limiting factor: VOLUME 10, 2022 3 and 4 UABSs could simultaneously be supported using 4 and 8 beams.

2) CAPACITY IN SCENARIO 2
The capacity served by the network increases more than 2.5 times when using eight beams compared to only1 beam, providing an average of 97 Mbps (p 95 ) for a 20 MHz channel, achieving a channel efficiency of 4.85 bit/s/Hz. This served capacity doubles the ones found [28] where the performance of the interference is considered. The BH resource block (RB) usage behaves comparably to the capacity: the number of used RBs is three times higher for eight beams than1 beam. Also, the number of RB per UABS is reduced by 2/3rds, consistent with the number of users per UABS in Fig. 8b.

C. SCENARIO 3: MULTIBEAM BEAMFORMING IN ACCESS AND BH NETWORKS
The third scenario considers the multibeam in both the access and BH network.

1) COVERAGE IN SCENARIO 3
Here, the number of served users increases significantly with the usage of multiuser beamforming in the access network. Fig. 9 shows that the p 95 percentage of covered users (in boxes) rises from 32% to 69% when the number of access beams increases from 1 to 8 (blue and orange boxes) for 25 requested users. Compared to Scenario 2, only BH beamforming (see Sec V-B), more ground users can be covered, increasing the full coverage 100% from 5 to 10 users. Nonetheless, the behaviour of 4 access beams and access beams is very similar, with a small increment of less than 2%. This is because the power division for each beam reduces the maximum achievable power. This leads to the point where no more users could be served by the same UABS, maintaining the required SNR.
As a result, the number of served users per UABS is relatively constant, independent of the number of requested users. For 4 access beams, nearly 2.6 users/UABS are served, while almost 3.1 users/UABS are served for 8 access beams, a small increment of 20%.
The number of required UABSs to serve the ground users is reduced when the number of access beams increases. More users per UABS will be served when using more beams, reducing the need for more UABSs. In this case, only 5.5 drones (p 95 ) are needed for 8 access beams supporting 69% of the users, while 8 drones are necessary to support only 32% of the ground users with just one access beam. Here, the UABS requirements of 4 access beams and 8 access beams are similar, with a small increment of 10% for the 4 access beam case.

2) CAPACITY IN SCENARIO 3
The served capacity (p 95 ) of the network is presented in Fig. 9b. The requested capacity is described by the red dotted lines, whereas the straight lines represent the served capacity. For 4 and 8 access beams, the served capacity of 10 ground users is covered. Unfortunately, the network capacity served by the UABSs is not sufficient to serve all the 25 requested users. This is because, for many access beams, the power per beam is reduced, reducing the SNR, which limits the served bitrate per beam. Here, up to 172 Mbps will be served with 20 MHz of BW in the BH and 32.4 Mbps per UABS with 5 MHz in the Access. This means an efficiency of 8.6 bit/sec/Hz in the BH network and 6.5 bit/sec/Hz in the access network. These results are comparable with the HBF technique found in [24] where efficiency of 8.6 bit/sec/Hz is found for nearly 25 dB of SNR, which is relatively higher than the experience in our simulations. Similarly, in [62], the usage of low complexity Electronically Steerable Parasitic Array Radiators (ESPARs) provide similar performance despite that the maximum antenna gain is nearly 17dB lower than our UPR, accounting 8 bit/sec/Hz with an SNR of 25dB for the BH link. This is achieved based on the antenna adaptation algorithm provided for the ESPARs. In addition, the usage of the RB in the BH increases 2.7 times for 8 access beams compared with 1 beam, and a small increment in the bitrate per resource block is found in the 8 beam case, from 600 kbps/RB (1 beam) up to 672 kbps/RB (8 beams). This is because, for more traffic served in the access network, a more efficient allocation of the RB on the BH will be achieved, as explained in [34]. Besides, if fewer UABS are needed, the optimised network chooses UABS locations with better PL, allowing better MCS, and increasing the average bitrate per RB.

D. SUMMARY AND COMPARISON OF RESULTS
Finally, Table 7summarises the key results of the multiuser beamforming scenarios for a network using the BS list for the TBS locations, an antenna size 16 × 16, and 25 requested users. This Table collects the p 95 results from the coverage and capacity perspective.

1) COVERAGE COMPARISON
The percentage of served users (p 95 ) increases from 14% in Scenario 2 up to 69% in the best case of Scenario 3. However, this is not sufficient to serve all the requested users. The average number of served users per UABS is limited: around 3 users/UABS (3.2 for the best-case scenario), 40% of the maximum theoretical users per UABS (8 users). The required number of UABS behaves opposite for Scenario 2 compared with Scenario 3. For Scenario 2, the used UABS rises if the number of beams increases since BH beams are directly related to the number of UABS. While for Scenario 3, the more access beams available, the fewer UABS are needed to support the same number of requested ground users.

2) CAPACITY COMPARISON
The performance of the capacity is significantly better for Scenario 3. Here, the best-case supports up to 172.5 Mbps (p 95 ) compared with only 97.0 Mbps of Scenario 2. Similarly, the resource block usage is much higher in Scenario 3 with up to 256 RB and only 144 RB for Scenario 2. The RB per UABS is also 42% larger for scenario 2. The channel efficiency also improves in Scenario 3, having a maximum of 6.5 bit/s/Hz for the access network and 8.6 bit/s/Hz for the BH network, compared with the 4.6 bit/s/Hz and 4.9 bit/s/Hz for scenario 2.

VI. CONCLUSION AND FUTURE WORK
Unmanned Aerial Base Stations are a promising solution for fast deployable networks in crowded scenarios or when terrestrial connectivity is limited. As the drones' technologies provide longer flying times, coverage of UABS could be extended. This imposes a challenge to the BH link where the restriction of the propagation losses limits the link range. To surpass this, digital beamforming will increase the antenna gain and improve the access and BH link quality.
This work evaluates the digital beamforming technique's performance to support MuMIMO in the BH and the access networks. To this end, we implement a simulation tool that includes beamforming capabilities in a mobile scenario for an UABS-aided network in a realistic environment. Simulation results show that using the actual terrestrial BS locations improves the connectivity performance compared with a private network. If the beamsteering technique is used in the BH, the served capacity increases two times. Moreover, when MuMIMO is used, we found that the user coverage increases four times compared to a single beam MIMO network. For the proposed scenario, we achieved a channel efficiency of 6.5 bit/s/Hz for the access link and 8.5 bit/s/Hz for the BH link, comparable with results found in the literature.
Future work will cover a dynamic ground facility location that can extend the range of the BH link and will enable improved user coverage. It will include an upgrade on the sensibility of recent 5G New Radio (5GNR) transceivers to improve the requested SNR of the system for the 3.5 GHz, 26 GHz, and 60 GHz bands. Finally, we will complement our work by modelling and evaluating beam tracking techniques to maintain beam alignment in terms of the physical layer and drone movement. We will develop a channel model to include the impact of velocity on the capacity and performance of the system. GERMAN   In October 2015, he became a part-time Professor in medical applications of electromagnetic fields in and around the human body. His scientific work focused on modeling indoor and outdoor propagation through field measurements. VOLUME  His professional interests include electromagnetic field exposure assessment, propagation for wireless communication systems, antennas, and calibration. Furthermore, he specializes in wireless performance analysis and quality of experience.