Dynamic Coverage Optimization for 5G Ultra-dense Cellular Networks Based on Their User Densities

This paper has proposed a user-density-based coverage optimization technique for ultra-dense cellular networks. Antenna tilting is a promising coverage optimization technique to be used in 5G networks, that significantly improve the signal to interference plus noise ratio (SINR) by choosing the appropriate angle of tilt. In this paper, the cellular coverage has been optimized for scattered user densities/user hotspots using an adaptive antenna tilting mechanism that steers the beams towards the temporal hot spot in the coverage area. The proposed method has the competence to improve the desired SINR level and coverage area for a group of users rather than a single user. In this work, a reinforcement learning (RL) algorithm has been implemented to optimize the tilt angle. The performance of the proposed technique has been evaluated in the simulation platform considering a three-sectored multicellular mobile network where the groups of user clusters are distributed randomly. The result confirms the improvement in RSS and SINR values in the group of users having high density with maximum user satisfaction.


Introduction
Due to exponential growth in the number of connected devices supporting internet of things (IoT) applications and emergence of data hungry applications, the mobile network operator (MNO) faces significant challenges in deploying the next generation cellular network. The upcoming networks are more complex with an increase in its deployment expenditures i.e., operational expenditure (OpEX) and capital expenditure (CapEX). The quality of service (QoS) of such networks depends on the interference parameters; the cell edge users are greatly affected by interference due to their weak received power and low signal to interference plus noise ratio (SINR) level. To meet the several demand-specific requirements in the next-gen cellular network, various researches have been carried out to optimize the antenna parameters to achieve better SINR and hence good QoS [1,2].
As the nature of the cellular network is quite dynamic due to various factors (e.g., dynamic characteristics of the propagation medium, user distribution), it is desired to improve the scalability of the mobile network with increased throughput. It is very much expensive and time-consuming for the MNO to optimize the network performance manually. The 3rd generation partnership project (3GPP) has introduced the concept of self-organizing networks (SON) to make the network autonomous for the future cellular networks [3]. For the next-generation cellular networks, the coverage and capacity optimization (CCO) are defined as a part of SON use cases, where some self-executing network parameters are discovered to impart enough capacity and coverage [3]. The corresponding coverage is defined as the probability that, the received SINR level is greater than the threshold value. Hence, a sufficient QoS is provided in terms of downlink received signal power over the total area. Moreover, the future (5G) cellular networks are envisioned to be high ultra-dense and more user-centric where the network will follow the individual user or a group of users (hotspot zone) aiming to provide a high data rate and better QoS [4,5]. Hence, focusing on the coverage as an essential parameter for the future cellular network, a hardware-based technique commonly known as the antenna tilting approach brings more attention to it, due to its low complexity and OpEX. Due to the deployment of this technique, a notable increase in the desired SINR level and average throughput within a cell can be perceived in addition to the interference to the other cells can be minimized [4,5]. In this approach, based on the network environment, propagation medium characteristics, and user distribution, one base station (BS) can adjust its tilt angle of the antennas to achieve a better trade-off between coverage and capacity.
In this paper, the cellular coverage has been optimized for scattered user densities/user hotspots using an adaptive antenna tilting mechanism that steers the beams towards the temporal hot spot in the coverage area. This paper is an adjunct to the method proposed in [6], by taking the advantage of the proposed algorithm and processes therein. The analysis is done based on the antenna tilt performance for scattered user densities in terms of the SINR and received signal strength (RSS) which are the salient metrics used to determine the coverage probability (CP) and average throughput of a mobile network. In this work, a reinforcement learning (RL) algorithm has been implemented to optimize the antenna tilt angle to improve the desired SINR level and coverage area towards the cell with maximum user density in urban areas by reducing the exorbitant interference within the network. The performance of the proposed technique has been evaluated in the simulation platform considering a three-sectored multicellular mobile network where the groups of user clusters are distributed randomly. The result confirms the improvement in received signal strength (RSS) and SINR values in the group of users having high density with maximum user satisfaction. The proposed technique makes the networks more independent and self-optimized. The rest part of the paper is organized as: Sect. 2 focuses on the state-of-art, in Sect. 3 the system model and the proposed algorithm with the principle of operation are presented, wherein in Sect. 4 the simulation results are analyzed and in Sect. 5 the conclusion with the future scopes is outlined.

State-of-Art
To extend coverage in cellular networks, the base station antenna tilting is a promising technique; with an optimal tilt configuration interference from neighbouring cells can be minimized [7]. Yilmaz et al. have proposed a remote electrical tilting mechanism to down-tilt the antenna that can be performed electrically from remote location [8]. The electrical down tilt mechanism outperformed the conventional mechanical-based system in interference-limited systems like downlink of the 3GPP Long Term Evolution (LTE).
The self-optimization of remote electrical tilt from the perspective of CCO has been explicitly discussed in [9]. The simulation shows a significant improvement in the system performance for suboptimal network planning in 3G networks. The major problem in implementing electrical tilting mechanism is finding the optimum tilt angle settings. An optimization task is formulated by Partov et al. for adaptation of antenna tilt angle based on maximizing user utility [10]. However, the authors reformulated the optimization problem as convex under certain conditions.
Many factors influence the decision that are difficult to analyse during network planning or network operation. The machine learning algorithms are found as key approach for self-optimization of the cellular network [11]. Razavi et al. have proposed a fuzzy RL-based technique to facilitate self-CCO for LTE networks by using BS antenna downtilt in [12]. The proposed solution is a completely distributed fashion without any additional signaling overhead between LTE BSs. Berger et al. have proposed a CCO algorithm where coverage and statistical throughput information are used to construct an objective function to estimate the number of covered and uncovered users of each cell. From the simulation results, a significant improvement in the coverage performance and overall throughput gains can be observed [13]. Dandanov et al. have proposed a machine learning approach for self-optimization of mobile networks where the antenna tilt angle optimization is done using RL algorithm result in around 30% of improvement in the sum data rate of the network [6]. For the RL-based antenna tilt dynamic self-optimization approach given in [6], the effect of several wireless propagation models on the downlink performance of a cellular network in an urban scenario has been investigated in [14]. To solve the coverage and interference problem that arises in LTE networks, a novel self-tuning algorithm to adjust the antenna tilt is proposed in [15]. In this method, the heuristic algorithm has been implemented by two fuzzy logic controllers. The simulation results show that the SINR value at the cell edge is improved by more than 1 dB causing an increase in the spectral efficiency. Using the existing method as in [6], Samal et al. have proposed a CCO method for suburban areas in [16]. With the application of the proposed algorithm, there is substantial improvement in the SINR level at the cell edge can be achieved. The simulation results also confirm user satisfaction at the celledge which nearly equals 80-100%.
Pedras et al. has proposed a quality of experience (QoE) based approach to optimize the down-tilt angle of the cellular network [17]. A computational-efficient centralized method is proposed in [18] where the self-planning of antenna tilts is driven by QoE approach. However, due to the inconsistent nature of the radio propagation of different sectors/cells in the cellular network and due to the random geographic distribution of the traffic across the sectors, the adaptation of antenna down-tilt can be executed in a distributed fashion for individual sectors. This facilitates the network to be more perceptive toward the dynamic environmental changes. The work presented here, emphasizes more on the adaptive and dynamic antenna tilt adjustment by incorporating RL-based methodology. The proposed model in this work is more predictive in terms of envisioning the optimal tilt angle of a BS for different distributed user hotspots.

System Model
In this work, a mobile network is modeled by considering a sectorized and multicellular network. Assuming each cell is hexagonal in structure and is equally divided into three sectors (the term 'sector' will be referred to as 'cell' in the further part of this paper as the sector and cell carry the same meanings) at an angle of 120 o . A single BS mounted with three antennas is placed at the center of the cell, where the user within each sector is served by a single antenna as shown in Fig. 1.
This work mainly focuses on downlink (DL) analysis where the RSS is calculated based on the radio signal strength transmitted ( P T ) by the BS with antenna gain denoted as G T(n,i) , the channel gain |h (n,i) | 2 and the distance (d) between the user/user equipment (UE). Here, it is assumed that the users are randomly distributed in such a manner that they form different user densities at different locations within a cell/sector. This proposed model is based on the same scenario and utilizes the algorithm as in [6].
Let's assume that the number of cells in the network is I and as the number of BS is equal to the number of cells, so the ith BS antenna, i = {1...I} . Due to the random distribution of the users, C is denoted as the number of user clusters/user densities that are formed within For cth user cluster-1 the received power ( P R ) by nth user, Here,P N = Additive White Gaussian noise (AWGN), G R(n,i) is the gain of the receiving antenna (Assumed to be equivalent to '1' due to omnidirectional) and P I(n) is the interfered signal power (other than i ), received at user n , where P I(n) = ∑ N l=1,l≠n P R(n,l) . The data rate received at the nth user due to the ith servicing BS can be representedby using Shannon-Hartley theorem asfollows [19], where, Δf is the bandwidth of the transmitted signal.The channel gain for each user present nearly at equidistance from the BS ( ith ) in cth user cluster, where,λ is considered as the wavelength of the transmitted radio signal, α is thepath-loss exponent, d 0 is afar field reference distance of the antenna, and d (n,i) is thedistance of user b from the BS antennai . {Assuming the Eq. (1) Similarly, as perEquation (2) &Eq. (3) the received power and SINR for all userclusters can be expressed as follows, Similarly, by using Eq. (4), the channel gain for each user present nearly at equidistance from the BS ( ith ) in different user clusters can be calculated as, Integrating the shadowing effect and by considering multicarrier transmission, the data rate received by the n th user can be expressed as follows [19], Here, ψ is considered as log-normal distributed random variable ( 10 ψ dB 10 ).

Radiation Pattern Modeling of BS Antenna
The antenna tilt mechanism for changing the radiation pattern is shown in Fig. 2.To obtain better CCO, it is necessary to adjust the tilt angle of the BS antenna precisely. In general,  [20]. So, the total tilt angle ( θ tilt ) of the BS antenna is, Considering a macro cell with sectorized cell site (tri-sector cell), the vertical radiation pattern { A v (θ)} and horizontal radiation pattern { A h (θ)} as in [20] can be integrated into a 3-D radiation pattern as, where, A m is considered as maximum horizontal attenuation, φ and θ are the azimuthal and elevation angle respectively. So, the total antenna attenuation {A φ, θ, θ tilt } at any point in space can be expressed as in Eq. (9) Where, if θ tilt > 0 o , the antenna (of the servicing BS) is down-tilted i.e., towards the earth surface and if θ tilt < 0 o , the antenna (of the servicing BS) is uptilted i.e., away from the earth surface (towards the horizon).
The elevation and azimuth patterns of a V65S-1XR configured tri-sector antenna at 1900 MHz and 2600 MHz frequencies are shown in Fig. 3 [21].
The 1-D antenna gain in random direction can be modeled as [7], For the peak antenna gain G 0 (dBi), corresponding azimuthal antenna gain G(φ) and elevation antenna gain G(θ) with an overall sidelobefloor SLL 0 (dB).
As stated in [20], the 3-D antenna gain can be expressed as, where, G T,max | |dBi is considered as the maximum antenna gain.The Eq. (11) can be approximated as,

Application of RL Algorithm
To optimize the coverage dynamically and adaptively, it is necessary to operate the BS antenna angle tilt in a self-organized manner. RL-based solution is integrated to facilitate this behavior of the antenna. Another parameter is known as the weighting factor ( n , 0 < n < 1 ) which determines the priority of the user hotspot for receiving a high data rate. The application of the RL based algorithm to find the optimum antenna angel tiltfor maximum user cluster can be clearly perceived by following the steps as specified below;

Simulation Results and Analysis
In this section, the performance of the proposed antenna tilt optimization technique is evaluated for downlink of a sectorized multicellular network in simulative platform. In the simulation, a 5000 m X 5000 m terrain is considered that is covered by three cells having perfect hexagonal cellular geometry and each cell is served with three 120 0 sectoral antennas. The frequency reuse factor is considered as 1 to ensure maximum system capacity. Here, the pathloss characteristics is considered from [20] that implements a simplified path-loss model with path-loss exponent (α) as 4.5 for urban terrain. Multi-carrier modulation scheme is considered with number of subcarriers ( N SC ) as 5000 and carrier spacing (Δf ) of 10 MHz. The carrier frequency considered for the simulation (f c ) is 1 GHz. Details of the simulation parameters related to antenna is presented in Table 1.
The simulation is carried out by considering five different user hotspots (as represented in form of '*', the number of users in the densities are taken randomly), which are • towards x-axis and then an increment of 2 • (for each state) will be obtained in an anti-clockwise direction. After exploring all the possible states by using the proposed algorithm, the obtained received power map from the serving BS and the optimal tilt angel for maximum user satisfaction is shown in Fig. 4.
From the optimized received power map as shown in Fig. 4a, it can be observed that as cell 2 is not serving any user hotspot, so to avoid interference to other cells the optimal angle of the antenna is down-tilted. Furthermore, in this scenario,the P T for that cell can be reduced or put off for making the network more energy-efficient (EE). Again, sector 3 is serving three different user densities (10, 20, and 60), out of which only one user density (60) is present nearby the BS. So, to provide maximum numbers of user satisfaction the optimal angle for sector 3 is down-tilted ignoring the two other small number of user hotspots i.e., 10 and 20 which are present at a long distance from the BS. This also makes the network more EE, because the lesser user densities are far away from the serving BS and require maximum P T (by making the angle up-tilt) to achieve the desired SINR level or good QoS. Sector 3 serving two user densities (80, 90), out of which the maximum user density i.e., 90 is present at a longer distance compared to the other user density. So, to satisfy all the users in maximum user hotspot the optimized angle is up-tilted. From Fig. 4b (the captured debug value); it signifies that all the user hotspots (i.e., 100%) in simulated network are getting sufficient SINR levels. Figure 4c(the captured debug value) clarifies that, as the sector 2 is not serving any user, so to avoid unnecessary interference by BS transmitting power the antenna is down-tilted. Figure 5 shows the SINR level of each user densities with respect to the number iterations, where it can be observed that the SINR level varies after each iteration due to random variation in propagation channel properties. The SINR level is a function of the distance of the user clusters from their serving BS. From this figure, it can be clearly seen that due to the optimized tilt angle, the achieved SINR value for the user cluster 80 and user cluster 90 is maximum in every iteration. Figure 6 shows the RSS by each user densities, which signifies that the user densities (i.e., 80, 90) receive maximum signal strength. In fact, the RSS is a dependent parameter of the distance. From Figs. 5 and 6, it can culminate that due to the remarkable improvement in the SINR value and RSS for the user clusters with the highest density the good QoS and high data rate can also be attained.

Conclusion
In this work, a user density-based BS antenna tilt approach for the coverage and capacity optimization for ultra-dense cellular networks is presented. The proposed method utilizes an existing RL based algorithm for its analysis. This work mainly focuses on the maximum number of user satisfaction. From the simulation results, it can be observed that due to this proposed method the coverage area of the cellular network can be optimized in such a way that a group of users with maximum density will get ahigh SINR value and better RSS ( ≈ 100%usersatisfaciton ). This proposed solution also improves the EE of the network by reducing or making the transmitting power off for the cell which doesn't serve any user/user density. This facilitates green communication by reducing the CO 2 emission. For future work, this work can be jointly implemented with dynamic adjustment of the BS antenna height to find better CCO in a 5G cellular network.
Funding Not applicable.

Data availability Not applicable.
Code availability Not applicable.

Conflict of interest Not applicable.
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.