5G Network Deployment Planning Using Metaheuristic Approaches

Abstract

The present research focuses on optimizing 5G base station deployment and visualization, addressing the escalating demands for high data rates and low latency. The study compares the effectiveness of Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Simulated Annealing (SA), and Grey Wolf Optimizer (GWO) in both Urban Macro (UMa) and Remote Macro (RMa) deployment scenarios that overcome the limitations of the current method of 5G deployment, which involves adopting Non-Standalone (NSA) architecture. Emphasizing population density, the optimization process eliminates redundant base stations for enhanced efficiency. Results indicate that PSO and GA strike the optimal balance between coverage and capacity, offering valuable insights for efficient network planning. The study includes a comparison of 28 GHz and 3.6 GHz carrier frequencies for UMa, highlighting their respective efficiencies. Additionally, the research proposes a 2.6 GHz carrier frequency for Remote Macro Antenna (RMa) deployment, enhancing 5G Multi-Tier Radio Access Network (RAN) planning and providing practical solutions for achieving infrastructure reduction and improved network performance in a specific geographical context.

1. Introduction

In the realm of wireless communication, 5G’s high speed, ultra-low latency, and extensive device connectivity redefine the possibilities. Multi-gigabit data rates enable lag-free streaming, virtual reality experiences, and real-time gaming. Ultra-low latency facilitates groundbreaking applications in healthcare, transportation, and manufacturing, enabling remote surgeries, self-driving cars, and smart factories [1]. The scalability of 5G networks accommodates diverse devices, fostering a connected ecosystem with smart homes, cities, and infrastructure. The Groupe Speciale Mobile Association (GSMA) Mobile Economy series predicts over 5 billion 5G connections by 2030, underscoring the widespread adoption and transformative potential of this technology [2,3].

Strategic planning in 5G network development is essential, particularly in optimizing base station placements. This not only ensures efficient performance and maximized coverage but also contributes to increased network capacity. The intelligent use of modern techniques, such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), considers factors like population density, user dispersion, topography, existing infrastructure, and future expansion plans. These approaches aim to enhance signal intensity, reduce attenuation, and improve overall network connection, thereby enabling seamless, high-speed connectivity in the 5G era.

Furthermore, the deployment of 5G networks, utilizing millimeter wave frequencies, introduces novel topologies and ambitious goals for download speeds. With a focus on supporting up to 1 million devices per square kilometer, 5G emphasizes the enhancement of mobile broadband services, particularly for indoor users who constitute a significant portion of network users. Innovations like massive Multiple-Input Multiple-Output (MIMO), distributed antenna systems, spatial modulation, mm-Wave tech, and small cells with heterogeneous network deployments are explored to meet these objectives [4].

The deployment of 5G networks at mm-Wave frequencies faces unique challenges, primarily due to the physical characteristics of mmWave signals. These frequencies, while capable of delivering high data rates, have shorter ranges and are more susceptible to attenuation from obstacles like buildings and trees. This necessitates a denser network of base stations to ensure comprehensive coverage, especially in urban environments. The deployment strategy must, therefore, consider the urban landscape, identifying optimal locations for base stations that maximize coverage and signal quality while minimizing interference and cost.

In Nepal, the optimal method for 5G deployment currently involves adopting a Non-Standalone (NSA) architecture, which overlays a 5G radio access network (RAN) on existing 4G infrastructure. This approach minimizes upfront costs by utilizing already deployed infrastructure while enabling a faster rollout of 5G services [5,6]. However, NSA relies on the existing 4G core network, which may limit the full realization of 5G’s standalone capabilities until a later phase. To address this challenge, our study emphasizes the significance of metaheuristic algorithms in optimizing the placement of new 5G base stations, while also optimizing the use of spectrum resources, thereby enhancing the efficiency and effectiveness of 5G deployments in Nepal’s diverse geographic and economic landscape.

The focus of this study is on the optimization of 5G base station deployment at mmWave frequencies. The aim is to determine an optimal placement strategy that maximizes coverage, minimizes interference, and ensures efficient utilization of available spectrum resources. The primary emphasis of this research lies in maximizing data rates and coverage while minimizing the deployment cost of multi-tier remote radio units.

The main contributions of this paper are summarized as follows:

• Conduct a detailed comparative study of various optimization algorithms, including GA, PSO, Simulated Annealing (SA), and Grey Wolf Optimizer (GWO), in the context of 5G Radio Access Network (RAN) planning.
• Optimize the positioning of base stations and evaluate the performance and feasibility of the proposed metaheuristic algorithms.

The rest of the paper is organized as follows: Section 2 reviews related works, focusing on the use of metaheuristic algorithms for optimizing 5G network deployments. Section 3 details the methodology, including system architecture and the application of various metaheuristic algorithms. Section 4 describes the experimental setup, including simulation tools and parameters. Results and analysis are presented in Section 5, comparing the effectiveness of each algorithm. Section 6 discusses the implications of these findings, and Section 7 concludes with a summary and future research directions.

2. Related Works

The research highlights the strategic use of 5G millimeter-wave frequencies, particularly focusing on indoor mmWave small cell network planning between 26 GHz and 38 GHz, employing Close-In (CI) and CI model with a frequency-weighted path loss exponent (CIF) path loss model along with 3D ray-tracing methods for optimal path loss modeling [7,8,9,10,11]. These studies emphasize the need for precise path loss models that can predict network coverage effectively across various urban environments, terrains, and generational technology shifts [12,13].

Enhancements in 5G coverage have been demonstrated through the use of simulation tools like the Vienna 5G simulator, which address bandwidth limitations and improve network efficiency by optimizing base station placements [14,15,16,17]. The integration of machine learning techniques, particularly Random Forest algorithms, has proven effective in both line-of-sight and non-line-of-sight scenarios, optimizing coverage predictions and network planning [18,19].

Robust optimization techniques using multi-layer architectures enhance network resilience and adaptability, incorporating advanced strategies such as game theory for addressing ultra-dense network deployments in both 5G and future 6G scenarios [20,21]. The rapid advancements facilitated by machine learning algorithms have been speeding up the planning processes, achieving up to 15 times faster results compared to traditional methods [22].

Furthermore, the paradigm shift towards data-driven network management through network slicing and deep reinforcement learning has opened new avenues for managing network capacities and life-cycle complexities [23,24,25]. Despite substantial advancements, there remains a noticeable gap in integrating these sophisticated technologies into a cohesive framework that effectively manages both urban and rural deployment challenges. Our research fills this gap by utilizing a novel integration of metaheuristic algorithms with advanced path loss models to enhance both the accuracy and efficiency of network deployment strategies [5,26,27,28,29].

The paper further explores novel optimization techniques like PSO for base station deployment, which have shown to improve coverage significantly and reduce user outages, thereby enhancing the overall network performance [30,31,32,33,34]. These innovative approaches underscore the necessity for continuous research and adaptation in the face of evolving technological and operational demands, paving the way for more resilient and efficient 5G networks [35,36,37,38,39].

Moreover, enhanced strategies such as Smart Simulated Annealing (SSA) and Swarm Intelligence (SI) are optimizing the placement of base stations for better coverage and capacity, while beam-forming technologies in 5G networks are streamlining the number of required base stations, providing increased network capacity and efficiency [40,41,42,43,44]. These advanced methodologies are crucial for supporting the high demands of modern mobile networks, offering reliable solutions that cater to both current and future connectivity needs [45,46,47,48,49,50].

While refs. [29,38,40] evaluate frequencies and offer a broader perspective on the impact, they do not focus heavily on specific recommendation and deployment scenarios. Our study highlights a specific urban area (Thapathali, Kathmandu), providing a case study that showcases the application and benefits of metaheuristic algorithms in a real-world setting. It also provides a more detailed comparison of metaheuristic algorithm within the context of urban deployment.

3. Methodology

3.1. System Architecture and Block Diagram

The overall block diagram including system architecture is presented in Figure 1. The study begins by defining the coverage area, user density, and quality of service requirements for the proposed network. It collects geographical and user-related data, including terrain maps, population distribution statistics, and details of existing infrastructure. A network model is built to depict the coverage area accurately, considering obstacles, propagation models, and signal strength requirements. Network model is a detailed framework that combines system modeling and path loss calculations with metaheuristic algorithms (PSO, GWO, GA, and SA) to optimize base station locations and configurations. Four metaheuristic algorithms—PSO, GWO, GA, and SA—are utilized. PSO and GWO are population-based, SA uses a probabilistic approach, and the GA employs evolutionary principles.

System modeling includes path loss models for link budget design, covering calculations for uplink and downlink transmission. After calculating the Radius from the Link budget analysis, constraints are established, focusing on maximizing data rates and coverage while minimizing the cost of deploying multi-tier remote radio units. These constraints are integrated into the chosen algorithm to optimize network design. Performance evaluation involves metrics like coverage, signal quality, and data transfer rates. Comparative analysis of the algorithms determines the most efficient solution in terms of convergence speed and solution quality. The algorithm best meeting the study’s objectives and requirements is selected as the optimal solution.

3.2. Modeling 5G Network Parameters

The 3GPP 38.901 model [51] was chosen for its strong reputation and widespread acceptance in the telecom community, rendering it ideal for our 5G network deployment planning. Key factors included its compatibility with 5G frequency bands and its ability to handle urban propagation environments. The model was integrated into our simulation framework, with parameters fine-tuned for our area’s unique characteristics. Rigorous validation against real-world data was conducted to ensure accurate predictions. This approach, combined with optimization using metaheuristic algorithms, facilitated effective 5G deployment planning in complex urban landscapes. The symbols and their usual meanings for formula representation are presented in

Table 1. Symbols and their descriptions.

Notations	Descriptions
UMa	Urban macro cells
RMa	Rural macro cells
RRU	Remote radio units
S	The total surface of served area
$N_i^{Cov}$	The number of RRUs, having type i, needed to cover the area of interest
$N_i^{Cap}$	The number of RRUs, having type i, needed to satisfy the capacity constraints
$N_u^i$	Maximum number of users that can be served by one RRU
$N_i^{Ini}$	Estimated initial number of RRUs
MAPL	Maximum allowed path loss for cell
$P{L}_{UMa}$	Path loss for UMa cells
$P{L}_{RMa}$	Path loss for RMa cells
$S_{Cell}^i$	Surface of cells with type i
$C_s^i$	Channel Capacity Of Cell i per sector antenna type i
$N_s$	Number of Sector of antenna
BW	Bandwidth
$T_{DR}$(DL)	Target data rate in Downlink
SE	Spectral Efficiency
EIRP	Effective Isotropic Radiated Power

. From ref. [52], equations of the path loss model for No Line of Sight (NLOS) and Line of Sight (LOS) for Remote Radio Units (RRUs) can be written as follows:

Calculation of path loss for UMa on NLOS $P{L}_{UMa-NLOS}\left[dB\right]=max(P{L}_{UMa-LOS},P{L}_{UMa-NLOS}^{\prime }$)

$$for10\mathrm{m}\le d_{2D}\le 5\mathrm{km}$$

(1)

where

$$\begin{array}{c}\hfill P{L}_{UMa-NLOS}^{\prime }=13.54+39.08{\log }_{10}\left(d_{3D}\right)+20{log}_{10}\left(f_c\right)-0.6(h_{UE}-1.5)\end{array}$$

(2)

Calculation of path loss for UMa on LOS

$$P{L}_{UMa-LOS}=\left\{\begin{array}{cc}P{L}_1\hfill & 10\mathrm{m}\le d_{2D}\le d_{BP}^{\prime }\hfill \\ P{L}_2\hfill & d_{BP}^{\prime }\le d_{2D}\le 5\mathrm{km}\hfill \end{array}\right.$$

(3)

Note: Break point distance d’BP = 4 h’BS h’UT fc/c, where fc is the center frequency in Hz; c = 3.0 × 108 m/s is the propagation velocity in free space; and h’BS and h’UT are the effective antenna heights at the BS and the UT, respectively. The effective antenna heights h’BS and h’UT are computed as follows: h’BS = hBS − hE and h’UT = hUT − hE, where hBS and hUT are the actual antenna heights, and hE is the effective environment height. More details can be found in [52].

$$P{L}_1=28.0+22{log}_{10}\left(d_{3D}\right)+20{log}_{10}\left(f_c\right)$$

(4)

$$P{L}_2=28.0+40{log}_{10}\left(d_{3D}\right)+20{log}_{10}\left(f_c\right)-9{log}_{10}({\left(d_{BP}^{\prime }\right)}^2+)h_{BS}-h_{UT}{)}^2)$$

(5)

Calculation of path loss for RMA on LOS

$$P{L}_{RMa-LOS}=\left\{\begin{array}{cc}P{L}_1\hfill & 10\mathrm{m}\le d_{2D}\le d_{BP}\hfill \\ P{L}_2\hfill & d_{BP}\le d_{2D}\le 10\mathrm{km}\hfill \end{array}\right.$$

(6)

Note: Break point distance dBP = 2$\pi$ hBS hUT fc/c, where fc is the center frequency in Hz; c = 3.0 × 108 m/s is the propagation velocity in free space; and hBS and hUT are the antenna heights at the BS and the UT, respectively. More details can be found in [52].

$$\begin{array}{cc}\hfill P{L}_1=& 20{log}_{10}\left(40\pi d_{3D}{f}_{c}l3\right)+min(0.03{\mathrm{h}}^{1.72},10){log}_{10}\left(d_{3D}\right)\hfill \\ \hfill & -min(0.044{\mathrm{h}}^{1.72},14.77)+0.002{log}_{10}\left(h\right)d_{3D}\hfill \end{array}$$

(7)

$$P{L}_2=P{L}_1\left(d_{BP}\right)+40{log}_{10}(d_{3D}/d_{BP})$$

(8)

Calculation of path loss for RMa on NLOS

$$P{L}_{RMa-NLOS}=max(P{L}_{R\mathrm{m}a-LOS},P{L}_{RMa-NLOS}^{\prime })\mathrm{for}10\mathrm{m}\le d_{2D}\le 5\mathrm{km}$$

(9)

where

$$\begin{array}{cc}\hfill P{L}_{RMA-NLOS}^{\prime }=& 161.04-7.1{log}_{10}\left(W\right)+7.5{log}_{10}\left(h\right)-(24.37-3.7{(h/h_{BS})}^2)\hfill \\ \hfill & {log}_{10}\left(h_{BS}\right)+(43.32-3.1{log}_{10}\left(h_{BS}\right))({log}_{10}(d_{3D}-3)+20\hfill \\ \hfill & {log}_{10}\left(f_c\right)-(3.2{\left({log}_{10}\left(11.75h_{UT}\right)\right)}^2-4.97)\hfill \end{array}$$

(10)

where $d_{BP}^{\prime }$ denotes the break-point distance; $f_c$ is the center frequency in GHz; $h_{BS}$ and $h_{UT}$ are the antenna heights and user height, respectively; $d_{2D}$ is the 2D distance between the UE and RRU in meters; and $d_{3D}$ is the 3D distance in meters calculated as

$$d_{3D}=\sqrt{d_{2D}^2+{(h_{RRU}-h_{UE})}^2}$$

Note: Channel modeling involves two equations, ${\mathrm{PL}}_1$ and ${\mathrm{PL}}_2$, which vary based on the distance between the transmitter and receiver. There exists a critical distance known as the break point distance. Equation (1) is used for distances within this limit, while a different equation is applied for distances beyond it, reflecting changes in channel characteristics. The break point distance and subsequent loss characteristics are defined according to specifications outlined in the 3GPP document for each model type. More details can be found in [52], while RRU specifications details are presented in

Table 2. RRU Specifications [38].

RRU Specifications
Specifications	Midband	Midband	mmWave
Frequency Range	2.6 GHz	3.6 GHz	28 GHz
Band	n7	n78	n257
Duplex Mode	FDD	TDD	TDD
Channel Bandwidth	5, 10, 15, …50	10, 15, …100	20, 100, 200, 400
Selected Bandwidth	20 MHz	40 MHz	100 Mhz

.

3.3. Dimensioning

3.3.1. Coverage Dimensioning

Following the selection of the path loss model, the next crucial step is conducting thorough coverage and capacity calculations for the base stations. These calculations ensure sufficient coverage and capacity to meet user demand. Coverage calculation begins with a detailed link budget analysis, accounting for signal propagation and attenuation factors such as body loss, penetration loss, foliage loss, and more. The Maximum Allowable Path Loss (MAPL) is determined using the link budget equation:

$$MAPL=EIRP+AG\left(ReceivedPower\right(PRx)+BL+PL+FL+RIL+SLF+IM)$$

(11)

where BL = Body loss, PL = Penetration loss, FL = Foliage loss, RIL = Rain/ice loss, SLF = Slow fading, RS = Receiver sensitivity, IM = Interference margin, and AG = Antenna gain.

The MAPL represents the maximum permissible power level for successful communication. Using the MAPL and the 3GPP 38.901 path loss model, the coverage radius of the base station is determined.

Assessing the actual area helps calculate the total number of base stations required for complete coverage. Considering real-world conditions ensures effective network design tailored to specific coverage requirements [38].

Eventually, the number of RRUs needed to cover the area of interest is expressed as follows:

$$N_{Cov}^i=⌈{\displaystyle \frac{S}{S_{Cell}^i}}⌉$$

where the symbol $\lceil .\rceil$ denotes the ceiling function and $S_{cell}^i$ is the surface of the cell with radius $RB{S}^i$. For instance, $S_{cell}^i=\pi \ast {\left(RB{S}^i\right)}^2$ for a circular cell and $S_{cell}^i={\displaystyle \frac{3\sqrt{3}}{2}}{\left(RB{S}^i\right)}^2$ for a hexagonal cell.

3.3.2. Capacity Dimensioning

Capacity estimation is critical alongside coverage calculations to assess the network’s ability to meet user demand. It involves setting a fixed downlink data rate for users, ensuring a minimum guaranteed rate per user. From this rate, individual base station capacities are determined, indicating the number of users they can support concurrently while maintaining the specified data rate. Considering factors like population density and user distribution, the total capacity required is calculated. This estimation offers insights into the network’s ability to handle anticipated traffic effectively. Integrating coverage and capacity calculations helps identify the optimal number and placement of base stations for a robust 5G network deployment [38].

Let i = UMa or RMa,

$$\begin{array}{cccc}\hfill {N_u}^i& =⌈\frac{{N_s}^i·{C_s}^i}{T_{DR}\left(DL\right)}⌉\hfill & \hfill \mathrm{where},[{C_s}^i& =B{W}^i·S{E}^i]\hfill \end{array}$$

(12)

Now,

$$\begin{array}{c}\hfill {N_i}^{cap}=\frac{D_{p}A}{{N_u}^i}\end{array}$$

(13)

Here,

$D_{p}A$ = a function that gives the desired average data rates of all the users in that area (its output is in terms of number of users).

Finally, the estimated number of initial number of RRus ${N_i}^{Ini}$ for each cell type i = {UMa, RMa } is

$$\begin{array}{c}\hfill {N_i}^{Ini}=max({N_i}^{cov},{N_i}^{cap})\end{array}$$

(14)

3.4. Metaheuristic Algorithms

PSO emulates bird flocking behavior to optimize 5G base station deployment, treating station locations as particles and iteratively adjusting them based on local and global best solutions to minimize deployment cost while maximizing coverage and resource utilization [53]. The GA addresses conflicting objectives in deployment by balancing factors like coverage, cost, and user satisfaction, representing base station configurations as chromosomes and evolving them through genetic operations for decision-makers to choose from [54]. SA gradually refines base station configurations in a large search space, striking a balance between exploration and exploitation to achieve efficient deployment by escaping local optima [55]. The GWO algorithm, inspired by wolf social hierarchy, iteratively refines base station layouts through hunting behavior, balancing coverage, and cost factors efficiently via alpha dominance and collaborative hunting strategies, facilitating effective 5G deployment planning [56].

Steps in Metaheuristic Optimization

Figure 2 illustrates the common steps involved in the metaheuristic optimization process, combining core concepts from the PSO, GA, SA, and GWO algorithms [53,54,55,56].

3.5. Visualizing Geographic Data for 5G Network Deployment

The process of visualizing geographic data is fundamental in the strategic planning and deployment of 5G networks. This involves several key steps:

Coordinate Conversion: The transformation of X and Y coordinates into latitude and longitude is critical for accurately mapping the positions of 5G base stations. This conversion ensures that the geographic data align correctly with real-world locations, facilitating precise base station placement for optimal network coverage and capacity.

Data Frame Creation and Storage: After coordinate conversion, the next step is to create and store these data in a structured format. A data frame is generated to hold the latitude and longitude information, typically saved as a CSV file. This file becomes a crucial repository of geographic data, supporting various analytical processes that aid in decision-making related to network planning and deployment.

Interactive Visualization Using Folium: Folium is employed to create interactive maps that visually represent the spatial distribution of base stations. These maps are not just tools for displaying location data but are instrumental in analyzing spatial relationships and network coverage areas. By integrating features like circles and markers, Folium maps provide a clear and intuitive view of how base stations are spread across a region, helping network engineers and planners assess and optimize the deployment strategy.

These visualization techniques play an essential role in modern telecommunications infrastructure, particularly for the efficient deployment of 5G networks. By enabling precise geographic data handling and visualization, these tools contribute significantly to enhancing network connectivity and performance, ensuring that base stations are strategically placed to meet coverage demands effectively.

4. Experimental Setup

4.1. Link Budget Calculation and Simulation Setup

The experimental setup used a Jupyter notebook for simulation and optimization, leveraging results from a separate link budget calculator. The calculator required inputs such as system parameters (bandwidth, frequency, distance, average building height, and street width), design parameters (transmit power, antenna gains, transmitter and receiver heights, cable losses, noise figure, and SINR target), and loss parameters (rain loss, penetration loss, vegetation loss, body loss, snow loss, slow fading, and interference margin). The results, including propagation losses and cell edge distances for urban and rural models, were used in the Jupyter notebook. The notebook utilized Python libraries: NumPy for numerical computations, Matplotlib for data visualization, Pandas for data manipulation, SciPy for advanced scientific computing, and Folium for creating interactive maps.

4.2. Experimental Area

The experiment was conducted within a defined geographical area characterized by latitude and longitude coordinates. Specifically, the study area spanned from latitude (min) = 27.6866 to latitude (max) = 27.6937 and longitude (min) = 85.3129 to longitude (max) = 85.3207. This constrained location served as the experimental setting for the investigation, as shown in Figure 3a,b. The results of the experiment are presented in Tables 6–9, where key parameters such as latitude, longitude, and population density in the year 2018 are recorded [57]. This geographical focus allows for a targeted analysis within the specified coordinates, providing valuable insights into population dynamics and distribution trends in the designated area.

4.3. Algorithm Configurations and Assumptions Made in Simulation

The necessary configuration requirements of listed algorithms in our experiment are provided in

Table 3. Configuration of different optimization algorithms.

Algorithm	Configuration
PSO	Initial number of base stations is 115 and 147. Control parameters: Cognitive coefficients (0.5, 0.09), social coefficients (0.01, 0.0007), and inertial weight (0.9, 0.99). Fitness is evaluated by measuring the distance to the nearest neighbor (BS) and the capacity by the ratio of covered area to points within the radius. Termination at 1000 iterations.
GA	Begins with mutation rate of 0.2, 3000 generations, and a selection size of 250. These parameters guide the behavior across iterations.
GWO	Uses random values for alpha, beta, and delta components within a range [0, 1] corresponding to the number of particles and dimensions in the optimization space.
SA	Starts with an initial temperature of 1000, a cooling rate of 0.95, and a scaling factor of 0.01. Maximum iterations set at 2000.

, while the assumptions for simulation of network deployment planning are provided in

Table 4. Assumptions made in the simulation for network deployment Planning.

Assumption	Description
Bandwidth Utilization	Each population within the study area is assumed to utilize an identical amount of bandwidth or data rate, ensuring a uniform baseline for network resource allocation.
Device Capability	It is assumed that all individuals, regardless of age group, possess devices capable of running cellular services, standardizing the potential user base for the network.
Signal Loss	Equal signal loss is assumed across the entire specified area, providing a simplified framework for evaluating network performance without the influence of varying geographical or environmental factors.
Interference	Interference is not explicitly considered in the simulation, simplifying the modeling process by excluding the effects of interference on network performance.
Other Factors	Other potential factors not explicitly modeled or anticipated in the simulation may also affect the results. These could include unforeseen technological changes, regulatory updates, or unexpected user behavior patterns.

.

5. Results and Analysis

Table 5. Results from 5G link budget calculator using different carrier frequencies.

Using 5G Link Budget Calculator
Carrier freq	28 GHz	3.6 GHz
Bandwidth	100 MHz	40 MHz
DL(m)	220	795
UL(m)	75	170
Capacity	416	166
Coverage Area (sq. km)	0.0176	0.0907

presents results from a 5G link budget calculator, comparing the impact of different carrier frequencies (28 GHz and 3.6 GHz) on network parameters such as bandwidth, downlink and uplink distances, capacity, and coverage area. It illustrates how higher frequencies offer greater bandwidth and capacity but have a shorter range, aiding in strategic network planning and deployment.

5.1. Result for Different Carrier Frequencies Using PSO

Table 6. Result for different 5G Multi-Tier RAN planning scenario using PSO.

Using PSO for Optimization
	Carrier Freq (Table 5)	28 GHz	3.6 GHz
Initial Arrangement	Initial BS No.	115	147
	Initial Cov	0.8993	1
	Initial Cap	0.868	0.927
Before Elimination of base station	Final BS	115	147
	Final Cov	0.9895	1
	Final Cap	0.9756	0.7916
After Elimination of base station	Final BS	63	110
	Final Cov	0.9618	1
	Final Cap	0.9166	0.7395

presents the outcomes of employing PSO for diverse 5G Multi-Tier RAN planning scenarios. The study focuses on parameters within the 5G Link Budget Calculator, specifically targeting carrier frequencies of 28 GHz for downlink (DL) and 3.6 GHz for uplink (UL), along with corresponding bandwidths and distances. The careful consideration of these variables, such as 100 MHz for DL and 40 MHz for UL, and distances ranging from 220 m to 75 m for 28 GHz and 795 m to 170 m for 3.6 GHz, provides a comprehensive foundation for analysis.

The initial configuration involves 115 base stations (BSs) for 28 GHz and 147 for 3.6 GHz. Before the elimination of specific BSs, the final configuration maintains the BS count while achieving substantial improvements in coverage and capacity. After post-optimization, the final BS count remains at 115 for 28 GHz and 147 for 3.6 GHz, with coverage reaching 0.9895 for 28 GHz and staying at 1 for 3.6 GHz. Capacities witness significant enhancements, reaching 0.9756 for 28 GHz and 0.7916 for 3.6 GHz. However, after eliminating certain BSs, the final configuration indicates a reduction in the number of BSs to 63 for 28 GHz and 110 for 3.6 GHz. While the coverage for 28 GHz slightly decreases to 0.9618, the coverage for 3.6 GHz remains unchanged at 1. Capacities experience a decrease, reaching 0.9166 for 28 GHz and 0.7395 for 3.6 GHz. These findings demonstrate the effectiveness of PSO in optimizing 5G RAN planning, achieving a balance between BS count, coverage, and capacity for different frequency bands.

The initial arrangement of base stations (BS) for both the 28 GHz and 3.6 GHz frequency bands is depicted in Figure 4a. For the 28 GHz band, the figure illustrates the distribution of 115 initial BSs across the planning scenario. Similarly, for the 3.6 GHz band, the initial arrangement showcases 147 BSs strategically placed according to the specified parameters.

Figure 5a captures the arrangement of BSs before any elimination in the 28 GHz frequency band. Additionally, Figure 5b showcases a Voronoi Plot corresponding to the 28 GHz frequency band. Voronoi diagrams are instrumental in illustrating the partitioning of space based on the proximity to different BS locations. The distinctive polygons in the plot delineate regions of influence for individual BSs, providing a clear visualization of how the network space is divided among the remaining BSs after the PSO optimization and prior to any elimination.

Figure 6a showcases the configuration of BSs after the elimination process in the 28 GHz frequency band. Following the PSO and the subsequent removal of specific BSs, this visual representation provides insight into the revised network layout. The reduced number of BSs, now totaling 63, is strategically arranged, reflecting the optimized configuration after the elimination process. In Figure 6b, the corresponding Voronoi Plot for the 28 GHz frequency band is presented. The Voronoi diagram illustrates the adjusted coverage areas for each remaining BSs after the elimination, depicting the spatial partitioning of the network.

The reduction in the number of base stations, coupled with substantial improvements in coverage and capacity, underscores the efficacy of PSO in optimizing the 5G Multi-Tier RAN planning scenario. In addition to the existing optimization scenarios, an enhancement is proposed by introducing a 2.6 GHz carrier frequency for RMa deployment atop the existing UMa arrangement. This extension aims to achieve 100% coverage and further improve the performance of the 5G multi-tier RAN planning, which is shown in Figure 7a,b. A similar process is carried out to address the dead zones shown in Figure 8, Figure 9 and Figure 10.

5.2. Results for different Carrier Frequencies Using GA

In

Table 7. Results for different 5G Multi-Tier RAN planning scenarios using GA.

Using GA for Optimization
	Carrier Freq	28 GHz	3.6 GHz
Initial Arrangement	Initial BS No.	115	147
	Initial Cov	0.8993	1
	Initial Cap	0.868	0.927
Before Elimination of base station	Final BS	115	147
	Final Cov	0.9618	1
	Final Cap	0.9305	0.9166
After Elimination of base station	Final BS	60	131
	Final Cov	0.9618	1
	Final Cap	0.8784	0.8819

, before BS elimination, the final setup showcases improved coverage values of 0.9618 for 28 GHz and 1 for 3.6 GHz, along with increased capacities of 0.9305 for 28 GHz and 0.9166 for 3.6 GHz. The elimination of specific BS, despite reducing the count to 60 for 28 GHz and 131 for 3.6 GHz, maintains coverage at 0.9618 for 28 GHz and 1 for 3.6 GHz. Capacities, while slightly decreasing to 0.8784 for 28 GHz and 0.8819 for 3.6 GHz, remain robust. This pattern underscores the effectiveness of the Genetic Algorithm in refining 5G network deployment. The optimization process demonstrates the algorithm’s capability to streamline the number of base stations, showcasing an efficient balance between reduced infrastructure and improved coverage and capacity.

Figure 8 presents two significant aspects of the GA optimization in the 5G RAN planning scenario at 28 GHz: Figure 8a shows the base stations after elimination and Figure 8b the Voronoi Plot. Figure 8a displays the base stations after elimination; the impact of the GA optimization process is evident, showcasing a reduced number of base stations compared to the initial configuration. Despite the reduction to 60 base stations, the coverage for 28 GHz remains high at 0.9618.

5.3. Results for Different Carrier Frequencies Using GWO

Table 8. Results for different 5G Multi-Tier RAN planning scenarios using GWO.

Using GWO for Optimization
	Carrier Freq	28 GHz	3.6 GHz
	Initial BS No.	115	147
Initial Arrangement	Initial Cov	0.8993	1
	Initial Cap	0.868	0.927
Before Elimination of base station	Final BS	115	147
	Final Cov	0.9444	1
	Final Cap	0.9375	0.8194
After Elimination of base station	Final BS	65	115
	Final Cov	0.9201	0.993
	Final Cap	0.8576	0.7777

provides a comprehensive overview of the results obtained from different 5G Multi-Tier RAN planning scenarios using the Grey Wolf Optimizer. The GWO optimization process begins with an initial configuration featuring 115 base stations (BSs) for the 28 GHz spectrum and 147 for the 3.6 GHz spectrum. Initial coverage values show promise, with 0.8993 for 28 GHz and a perfect 1 for 3.6 GHz, while initial capacity values exhibit efficiency with 0.868 for 28 GHz and 0.927 for 3.6 GHz. Following the GWO optimization, the final configuration maintains the BS count but noteworthy enhancements are observed. With 115 BSs for 28 GHz and 147 for 3.6 GHz, coverage sees a significant boost to 0.9444 for 28 GHz while remaining at 1 for 3.6 GHz. Capacities witness an improvement as well, reaching 0.9375 for 28 GHz and 0.8194 for 3.6 GHz. After the elimination of specific BSs, the final configuration showcases a reduction in the number of BSs to 65 for 28 GHz and 115 for 3.6 GHz. Although the coverage for 28 GHz slightly decreases to 0.9201, the coverage for 3.6 GHz maintains a high level at 0.993. Capacities experience a decrease, reaching 0.8576 for 28 GHz and 0.7777 for 3.6 GHz. The GWO optimization process proves effective in refining the deployment of 5G networks, demonstrating a strategic reduction in the number of base stations alongside some trade-offs in coverage and capacity. Also, the stochastic nature of the GWO and other metaheuristic algorithms can indeed lead to varied results, which can influence the observed performance in reducing the number of deployed base stations.

Figure 9a showcases the results of the GWO on 5G network deployment at 28 GHz post-base station elimination. Figure 9a displays the reduced base station count (65) achieved through GWO optimization. Figure 9b features a Voronoi Plot illustrating the coverage areas of the remaining base stations, emphasizing the optimized spatial distribution achieved by GWO. This visual representation highlights the algorithm’s effectiveness in refining the 28 GHz spectrum deployment for improved coverage and capacity.

5.4. Result for Different Carrier Frequencies Using SA

From

Table 9. Results for different 5G Multi-Tier RAN planning scenarios using SA.

Using SA for Optimization
	Carrier Freq	28 GHz	3.6 GHz
Initial Arrangement	Initial BS No.	115	147
	Initial Cov	0.8993	1
	Initial Cap	0.868	0.927
Before Elimination of base station	Final BS	115	147
	Final Cov	0.9131	1
	Final Cap	0.87152	0.9201
After Elimination of base station	Final BS	55	90
	Final Cov	0.833	1
	Final Cap	0.7187	0.6006

after SA optimization, the number of BSs remained constant, but notable improvements in coverage and capacity were observed. The final configuration maintained 115 BSs for 28 GHz and 147 for 3.6 GHz with increased coverage values of 0.9131 for 28 GHz and sustained 1 for 3.6 GHz. Capacities also improved, reaching 0.87152 for 28 GHz and 0.9201 for 3.6 GHz. Following the selective elimination of specific BSs, the final configuration showcased a reduction in the number of BSs to 55 for 28 GHz and 90 for 3.6 GHz. Despite a decrease in coverage for 28 GHz to 0.833, the coverage for 3.6 GHz remained at 1. Capacities saw a decline, reaching 0.7187 for 28 GHz and 0.6006 for 3.6 GHz. This reduction in the BS count, coupled with trade-offs in coverage and capacity, resulted an enhancement in the existing optimization scenarios by proposing the incorporation of a 2.6 GHz carrier frequency for RMa deployment alongside the existing UMa arrangement. This extension aims to achieve 100% coverage.

Figure 10a illustrates the outcome of the SA optimization at 28 GHz after the elimination process. The configuration shows a reduced number of base stations (55 for 28 GHz) and likely highlights the refined deployment. Additionally, a Voronoi Plot—a graphical representation dividing space based on proximity to specific points—might be included to visualize the coverage areas and distribution of base stations in the 5G network.

5.5. Final Results for Different Carrier Frequencies Using Metaheuristic Algorithms

From

Table 10. Results for different 5G Multi-Tier RAN planning scenarios using Metaheuristic Algorithms.

	Carrier Freq [Table 5]	28 GHz	3.6 GHz
Using PSO for Optimization	Final BS	63	110
	Final Cov	0.9618	1
	Final Cap	0.9166	0.7395
Using GA for Optimization	Final BS	60	131
	Final Cov	0.9618	1
	Final Cap	0.8784	0.8819
Using GWO for Optimization	Final BS	65	115
	Final Cov	0.9201	0.993
	Final Cap	0.8576	0.7777
Using SA for Optimization	Final BS	55	90
	Final Cov	0.833	1
	Final Cap	0.7187	0.606

, it is evident that the PSO algorithm demonstrates superior performance across various metrics. PSO achieves a balance between the number of base stations, coverage, and network capacity for both carrier frequencies. It outperforms GA, GWO, and SA in providing optimal solutions. Therefore, based on the conducted analysis, PSO and GA emerge as the most effective metaheuristic algorithm for the optimization of 5G Multi-Tier RAN planning in this scenario.

6. Discussion

The study systematically evaluated the performance of PSO alongside GA, GWO, and SA in the context of 5G Multi-Tier RAN planning. The meticulous consideration of parameters such as carrier frequencies, bandwidths, and distances provided a robust foundation for analysis. PSO exhibited superior performance across various metrics, achieving a commendable balance between the number of base stations, coverage, and network capacity for both 28 GHz and 3.6 GHz carrier frequencies. The algorithm outshone its counterparts, GA, GWO, and SA, emerging as the most effective metaheuristic approach for optimizing 5G Multi-Tier RAN planning in the given scenario.

Furthermore, the results highlighted the efficiency of GA in refining 5G network deployment by striking an optimal balance between reduced infrastructure and improved coverage and capacity. GWO showcased notable enhancements in coverage and capacity post-optimization, maintaining a strategic reduction in the number of base stations. Simulated Annealing demonstrated improvements in coverage and capacity despite a reduction in the number of base stations, emphasizing its effectiveness in achieving a refined deployment. The overall comparative analysis underscored the strengths of PSO and GA as the most effective algorithms for optimizing 5G Multi-Tier RAN planning, considering their ability to navigate the trade-offs between infrastructure reduction and network performance improvement.

7. Conclusions

This study presented a comprehensive approach to 5G network deployment planning in the specific geographical area of Thapathali, Nepal. The initial phase involved meticulous definition of coverage area, user density, and quality of service requirements, followed by the collection of geographical and user data, construction of a detailed network model, and integration of the 3GPP 38.901 model into the simulation framework. The choice of the 3GPP 38.901 model was justified based on its compatibility with 5G frequency bands and its proven ability to handle urban propagation environments. The model, integrated into the simulation, underwent rigorous validation against real-world data, ensuring accurate predictions. The utilization of metaheuristic algorithms, such as PSO, GA, GWO, and SA, facilitated effective 5G deployment planning in complex urban landscapes. The comparative analysis of these algorithms revealed PSO and GA as the most effective metaheuristic approaches, striking a commendable balance between the number of base stations, coverage, and network capacity for both 28 GHz and 3.6 GHz carrier frequencies. PSO and GA demonstrated efficiency in refining 5G network deployment by optimizing the balance between reduced infrastructure and improved coverage and capacity. GWO showcased enhancements in coverage and capacity post-optimization, strategically reducing the number of base stations. Simulated Annealing demonstrated improvements despite a reduction in base stations, emphasizing its effectiveness in achieving a refined deployment. In addition to these findings, the study proposed an enhancement by introducing a 2.6 GHz carrier frequency for RMa deployment, aiming to achieve 100% coverage and further improve the performance of the 5G Multi-Tier RAN planning. Overall, the study provided valuable insights into optimizing 5G Multi-Tier RAN planning, offering practical solutions for achieving a balance between infrastructure reduction and network performance improvement in a specific geographical context.