A Comparative Benchmark of Scale-Up and Scale-Out MIMO Architectures for 5G and Prospective 6G Networks

Abstract

The evolution toward prospective sixth-generation (6G) wireless networks is expected to significantly increase user density, bandwidth demand, and architectural complexity, reinforcing the need for scalable multiple-input multiple-output (MIMO) deployments. In this context, two fundamentally different design strategies have emerged: scaling up centralized antenna arrays and scaling out distributed cooperative infrastructures. This paper presents a system-level comparative benchmark of scale-up and scale-out MIMO architectures under identical operating conditions of three representative downlink deployments: centralized Massive MIMO, centralized XL-Massive MIMO, and distributed Cell-Free MIMO. All architectures are assessed under identical urban channel conditions, transmit power, bandwidth, and traffic assumptions, considering sub-6 GHz (3.5 GHz) and millimeter-wave (28 GHz) frequency bands as proxies for 5G and prospective 6G operation. A unified Monte Carlo simulation framework is employed to jointly evaluate aggregate throughput, spectral efficiency, coverage performance, interference behavior, and energy efficiency over a wide range of user densities and service radii. The results highlight the distinct architectural trade-offs between centralized and distributed deployments: XL-Massive MIMO maximizes aggregate throughput and spatial reuse in dense hotspot scenarios, whereas Cell-Free MIMO provides superior coverage uniformity and improved energy efficiency in wide-area deployments. By isolating the impact of architectural scaling under consistent assumptions, the presented benchmark offers quantitative guidance for 6G network design and deployment planning.

1. Introduction

The continuous evolution of mobile communication systems toward higher data rates, lower latency, and massive device connectivity has reinforced the central role of multiple-input multiple-output (MIMO) technologies in modern wireless networks. Since their large-scale adoption in fifth-generation (5G) systems, MIMO architectures have enabled substantial gains in spectral efficiency and capacity by exploiting spatial multiplexing and advanced beamforming techniques [1,2]. As network densification and traffic demands increase, these spatial techniques are expected to remain a cornerstone in the transition toward prospective sixth-generation (6G) networks [3].

While classical Massive MIMO deployments rely on centralized base stations equipped with tens of co-located antennas, further scaling of this approach has led to the concept of extra-large, or XL-Massive MIMO, in which hundreds of antenna elements are integrated within a single site. This scale-up strategy increases the available spatial degrees of freedom, enabling higher multiplexing gains and improved interference suppression, particularly in dense user scenarios [4]. However, enlarging centralized arrays also introduces challenges related to signal processing complexity, pilot contamination, and hardware scalability.

In parallel, an alternative scale-out strategy has gained significant attention in recent years. Cell-Free MIMO architectures abandon the notion of fixed cell boundaries by distributing multiple cooperative access points across the coverage area, jointly serving users through coordinated transmission and reception [5]. By exploiting macro-diversity and reducing cell-edge effects, Cell-Free MIMO has demonstrated the potential to deliver more uniform service quality and improved energy efficiency compared to conventional cellular layouts [6,7]. Nevertheless, this distributed paradigm shifts complexity toward fronthaul capacity, synchronization, and scalable user-centric processing.

Despite the extensive body of literature on centralized and distributed MIMO systems, quantitative comparisons between scale-up and scale-out architectures are often conducted under heterogeneous assumptions. Many studies focus on isolated performance metrics, such as spectral efficiency or throughput, or rely on differing channel models, power budgets, or bandwidth configurations, making direct comparisons difficult. As a result, there remains a lack of fair, like-for-like benchmarks that isolate the architectural trade-offs between centralized and Cell-Free MIMO deployments under identical operating conditions.

Scope and Contributions

This paper addresses this gap by presenting a comparative benchmark of scale-up and scale-out MIMO architectures for 5G and prospective 6G networks. Three representative downlink systems are evaluated: centralized Massive MIMO, centralized XL-Massive MIMO, and distributed Cell-Free MIMO. All architectures are assessed using a unified simulation framework with the same urban channel model, transmit power, bandwidth, and traffic assumptions, and are evaluated in both sub-6 GHz and millimeter-wave frequency bands. By jointly analyzing throughput, spectral efficiency, coverage, interference, and energy efficiency across a wide range of user densities and service radii, this work aims to clarify the fundamental performance trade-offs between centralized and distributed MIMO design strategies and to provide practical insights for future wireless network deployments.

The main contributions of this work are summarized as follows:

• A fair and reproducible benchmarking framework that enables a systematic comparison of scale-up and scale-out MIMO architectures under identical channel, power, bandwidth, and traffic assumptions, including centralized Massive MIMO, centralized XL-Massive MIMO, and distributed Cell-Free MIMO, conducted under identical channel, power, bandwidth, and traffic assumptions.
• A quantitative evaluation of the fundamental trade-offs between centralized and distributed MIMO design strategies for 5G and prospective 6G networks, clarifying the impact of architectural scaling on system performance.
• A multi-metric performance assessment jointly analyzing throughput, spectral efficiency, coverage, interference, and energy efficiency over a wide range of user densities and service radii.
• Practical design insights identifying the deployment scenarios in which scale-up or scale-out MIMO architectures provide the most favorable performance trade-offs, offering guidance for future dense wireless network planning.

2. Related Work

As wireless systems evolve beyond 5G toward prospective 6G networks, the literature increasingly emphasizes that future architectures must simultaneously support much higher user densities, broader bandwidths, stricter energy constraints, and more heterogeneous service requirements. Recent works on 6G architecture evolution have highlighted the need for more flexible, modular, and efficient end-to-end designs capable of integrating communication, sensing, intelligence, and sustainability requirements into a unified framework [8,9,10,11,12]. In parallel, surveys on emerging 6G enablers have shown that next-generation networks will not rely on a single technology, but rather on the combination of advanced antenna systems, distributed processing, reconfigurable environments, and intelligent resource orchestration.

Among the emerging antenna-domain innovations for 6G, fluid and movable antenna paradigms have recently attracted considerable attention. Fluid antenna multiple access (FAMA) has been proposed as a new multiple-access framework that exploits receiver-side spatial fluidity to mitigate inter-user interference and enable massive access over the same time–frequency resources [13]. More broadly, fluid antenna systems (FAS) and related next-generation reconfigurable antenna architectures introduce new spatial and morphological degrees of freedom that may improve coverage, reliability, spectral efficiency, and energy efficiency while reducing some of the scaling limitations of conventional large-array systems [14]. Similarly, movable antenna (MA) architectures and six-dimensional movable antenna (6DMA) systems have been proposed as promising 6G enablers capable of exploiting spatial channel variations through antenna position and orientation control, thus opening new possibilities for adaptive communications and sensing [15,16]. These developments confirm that antenna reconfigurability is becoming an important complement to traditional MIMO scaling strategies in future 6G networks.

Within this broader 6G context, Cell-Free Massive MIMO has emerged as one of the most promising distributed architectures. By deploying multiple cooperative access points (AP) over the service area and jointly serving users without predefined cell boundaries, Cell-Free MIMO exploits macro-diversity to mitigate inter-cell interference, improve coverage uniformity, and increase user fairness [5]. Early comparative studies already showed its potential advantage over small-cell and conventional cellular deployments [5,6,7]. More recent works reinforce the role of Cell-Free MIMO as a key architectural candidate for 6G, particularly in scenarios requiring ubiquitous coverage, high spectral efficiency, and flexible user-centric operation [17,18,19]. In particular, the literature has examined how Cell-Free systems can evolve from purely distributed communication platforms into scalable 6G infrastructures integrated with user-centric clustering, radio stripes, and hybrid centralized/distributed processing models [17,19].

The signal processing dimension of Cell-Free MIMO has also attracted increasing attention. A comprehensive recent survey reviews estimation, combining, detection, and precoding methods for Cell-Free massive MIMO, showing that channel estimation, scalable signal processing, and fronthaul-aware design remain central challenges for practical deployments [20]. At the same time, the integration of Cell-Free architectures with other 6G enablers has opened new research directions. For example, Cell-Free MIMO combined with Reconfigurable Intelligent Surface (RIS) and reconfigurable metasurfaces has been identified as a promising approach to jointly enhance spectral, energy, and coverage efficiency in future 6G networks [21,22]. These works further confirm that distributed architectures are likely to play an important role in the 6G ecosystem, although their practical implementation depends strongly on coordination complexity, fronthaul scalability, and robust signal processing.

In parallel to scale-out strategies, centralized scale-up approaches have led to the emergence of extra-large or XL-MIMO systems, where antenna arrays are extended to hundreds of elements and very large apertures at a single site. This scaling improves beamforming resolution, multiplexing capability, and spatial selectivity, especially in dense hotspot scenarios, but also introduces new propagation and processing effects that are not fully captured by conventional massive MIMO models. In particular, recent works have shown that near-field propagation, spherical wavefronts, visibility regions, and channel non-stationarity become increasingly relevant as array dimensions grow [23,24,25,26,27]. These effects fundamentally alter channel modeling, CSI acquisition, feedback, and distributed processing requirements in XL-MIMO systems.

Several recent contributions have addressed these XL-MIMO-specific issues from complementary perspectives. Near-field CSI feedback has been investigated using deep learning techniques to compress and recover large CSI matrices in spherical-wave propagation conditions [24]. Variational Bayesian and sparse Bayesian estimation methods have also been proposed to exploit near-field channel sparsity and improve channel estimation accuracy in subarray-based and hybrid-field XL-MIMO systems [25,28]. Beyond estimation, measurement-based studies have revealed important subarray-wise non-stationarity and distributed-processing implications for practical mid-band XL-MIMO deployments [26]. Moreover, the convergence between XL-MIMO and Cell-Free concepts is beginning to receive attention, as illustrated by recent work on low-complexity distributed combining for near-field Cell-Free XL-MIMO systems [29]. Altogether, this literature confirms that XL-MIMO is not merely a larger version of conventional massive MIMO, but a qualitatively different regime that raises new questions on channel modeling, CSI management, and processing scalability.

Massive MIMO itself remains the foundational reference architecture against which many of these newer paradigms are compared. Since its formal introduction, Massive MIMO has been widely recognized as a cornerstone technology for high-capacity wireless communication systems [1,2]. Early seminal works demonstrated that equipping base stations with a large number of co-located antennas enables substantial gains in spectral and energy efficiency through spatial multiplexing and favorable propagation conditions [1,2,4]. More recent studies have extended this perspective to 6G-oriented operating regimes, showing that realizing the full potential of Massive MIMO at mmWave and THz frequencies requires rethinking antenna architectures, estimation algorithms, signal processing complexity, and deployment topologies [30]. Thus, while Massive MIMO remains a fundamental benchmark for future wireless systems, the literature increasingly acknowledges that its performance, scalability, and implementation constraints must be revisited in light of 6G requirements.

A recurring challenge across centralized, distributed, and extra-large MIMO architectures is the acquisition and processing of channel state information (CSI). In conventional large-scale MIMO, CSI limitations such as pilot contamination, estimation errors, and feedback overhead have long been recognized as major performance bottlenecks [4]. These issues become even more critical in XL-MIMO due to near-field channel complexity and larger CSI dimensionality [24,25,27,28], and in Cell-Free MIMO due to the need to estimate, share, and coordinate CSI across geographically distributed access points [18,20]. As a result, CSI scalability is increasingly viewed as one of the most important cross-cutting challenges in the evolution from conventional Massive MIMO toward distributed and extra-large architectures.

In this context, alternative approaches have been investigated to mitigate the scalability challenges associated with explicit channel estimation. In particular, non-coherent massive MIMO schemes have been proposed to avoid explicit CSI acquisition by relying on energy-based or differential modulation and detection techniques [31]. These approaches aim to reduce pilot overhead and improve scalability when serving a very large number of users. However, despite their theoretical potential, non-coherent massive MIMO architectures face several practical challenges, including reduced spectral efficiency, increased detection complexity, and limited maturity for real-world deployments [32,33]. These approaches are particularly relevant in large-scale, distributed MIMO architectures, where CSI acquisition and distribution can become a major scalability bottleneck.

Despite the extensive literature on Massive MIMO, XL-MIMO, Cell-Free MIMO, and their integration with emerging 6G technologies, most existing studies analyze these architectures independently or under heterogeneous assumptions. Comparisons are often limited to selected performance metrics or rely on different channel models, bandwidths, power budgets, and deployment conditions, making it difficult to isolate the true impact of architectural scaling. In contrast, the present work provides a fair and reproducible benchmarking framework for systematically comparing scale-up and scale-out MIMO architectures under identical operating conditions, thereby enabling a clearer assessment of their performance trade-offs in 5G and prospective 6G scenarios.

3. Methodology and Research Objective

This work aims to address the following research question: How do scale-up and scale-out MIMO architectures compare under identical operating conditions in terms of throughput, spectral efficiency, coverage, interference, and energy efficiency?

To answer this question, a system-level benchmarking methodology is adopted. In this context, benchmarking refers to a systematic and controlled evaluation framework in which multiple MIMO architectures are assessed under identical operating conditions. This includes the use of a unified simulation framework, consistent channel models, identical transmit power and bandwidth, and the same user distribution across all evaluated scenarios.

This approach enables a fair and reproducible comparison that isolates the impact of architectural design choices—namely, centralized (scale-up) versus distributed (scale-out) deployments—on key performance metrics. Unlike heterogeneous simulation setups, the proposed benchmarking framework ensures that all observed performance differences arise exclusively from architectural characteristics rather than from parameter inconsistencies.

The evaluation is conducted through a parametric sweep over two key dimensions: the number of simultaneously served users K, representing network load, and the service radius R, representing coverage conditions. These parameters are varied over a wide range to emulate realistic deployment scenarios, from dense urban environments to extended coverage configurations.

For each $(K,R)$ operating point, a Monte Carlo simulation approach, as explained in Section 4, is employed to capture the stochastic nature of wireless channels and user distributions. Multiple independent realizations are generated, and performance metrics are averaged to ensure statistical robustness.

The overall simulation workflow is illustrated in Figure 1. As shown in the figure, the process begins by defining the simulation scenarios by selecting $(K,R)$ pairs. Then, user locations are randomly generated following a uniform spatial distribution within the service area. The wireless channel is modeled as a combination of large-scale path loss (COST-231 urban model) and small-scale Rayleigh fading.

Next, beamforming is applied using linear or hybrid precoding techniques (e.g., zero-forcing), enabling spatial multiplexing across users. Based on the resulting signal-to-interference-plus-noise ratio (SINR), key performance metrics are computed, including throughput, spectral efficiency, coverage probability, interference levels, and energy efficiency. These metrics are then averaged over multiple Monte Carlo realizations to obtain statistically stable results.

Overall, this methodology enables a comprehensive and reproducible evaluation of MIMO architectures across a wide range of operating conditions, providing insight into the fundamental trade-offs between centralized and distributed deployment strategies.

4. System Model and Simulation Framework

This section describes the system models, simulation assumptions and performance metrics used to conduct the comparative benchmark of scale-up and scale-out MIMO architectures. A unified simulation framework is adopted to ensure a fair, like-for-like evaluation of all considered systems under identical operating conditions.

4.1. Evaluated MIMO Architectures

Three representative downlink MIMO architectures are considered in this study, covering both centralized scale-up and distributed scale-out design strategies. The considered configurations allow evaluating two complementary dimensions: (i) centralized versus distributed architectures under the same total number of antennas, and (ii) the impact of scaling the number of antennas in centralized deployments.

All systems operate under the same total transmit power budget and serve a common set of users distributed within a circular coverage area. This configuration is selected as a representative baseline for distributed deployments rather than an optimized design, allowing a controlled and fair comparison with centralized architectures under identical power constraints.

4.1.1. Massive MIMO (Centralized)

The baseline centralized architecture corresponds to a conventional Massive MIMO system, where a single base station is equipped with a uniform planar array (UPA) comprising $M=64$ co-located antenna elements arranged in an $8\times 8$ configuration. The base station simultaneously serves multiple single-antenna users using linear zero-forcing precoding techniques. This configuration reflects typical large-scale MIMO deployments in current 5G networks and serves as a reference for performance comparison.

4.1.2. XL-Massive MIMO (Centralized)

To represent the scale-up strategy, an XL-Massive MIMO architecture is considered, where the centralized base station is equipped with $M=256$ antenna elements arranged in a $16\times 16$ UPA. By significantly increasing the array aperture and the number of spatial degrees of freedom, this architecture enhances beamforming resolution and spatial multiplexing capability. However, it also entails greater signal-processing complexity and more stringent hardware requirements than conventional Massive MIMO.

4.1.3. Cell-Free MIMO (Distributed)

The scale-out strategy is implemented using a Cell-Free MIMO architecture comprising $N_{\mathrm{AP}}=4$ cooperative access points, each equipped with a $4\times 4$ UPA ($M_{\mathrm{AP}}=16$ antennas). The access points are geographically distributed over the service area and jointly serve all users through coordinated transmission, without predefined cell boundaries. The total number of antenna elements equals that of the Massive MIMO baseline ($M_{\mathrm{tot}}=64$), enabling a direct comparison between centralized and distributed deployments with the same antenna count. The number of access points is selected to meet a fixed total transmit power constraint, ensuring a fair comparison with centralized Massive MIMO. This configuration is selected as a representative baseline for distributed deployments rather than an optimized design, allowing a controlled and fair comparison with centralized architectures under identical power constraints. Since the total transmit power is equally distributed among the APs, the number of APs and the number of antennas per AP are chosen to maintain an equivalent total radiated power across all architectures. This design choice ensures that performance differences are primarily driven by architectural characteristics rather than power imbalances.

4.2. Channel Model and Propagation Assumptions

An urban macro-cell propagation environment is assumed for all simulations. Large-scale path loss follows a COST-231-based urban model, while small-scale fading is modeled as Rayleigh fading. Although millimeter-wave channels such as in 28 GHz often exhibit sparse and directional propagation, a Rayleigh fading model is adopted for consistency and to isolate architectural effects. More realistic channel models may influence absolute performance and will be considered in future work. User locations are generated according to a homogeneous spatial distribution within a circular area of radius R, centered at the serving infrastructure. Unless otherwise stated, perfect channel state information (CSI) at the transmitter is assumed to isolate architectural effects from estimation errors. While perfect CSI is assumed to isolate architectural effects, practical systems face channel estimation errors, pilot contamination, feedback delays, and hardware impairments. These non-idealities affect centralized and distributed architectures differently. In centralized Massive and XL-MIMO systems, antenna co-location enables more consistent channel estimation and centralized processing. However, imperfect CSI reduces precoding accuracy, lowers beamforming gain, and limits spatial multiplexing performance, especially in high-load scenarios where pilot reuse is unavoidable. Conversely, Cell-Free MIMO architectures are more vulnerable to CSI inaccuracies due to their distributed setup. Channel estimation occurs at multiple geographically separated access points, and the resulting CSI must be shared or coordinated across the network. This adds challenges such as fronthaul signaling overhead, synchronization mismatches, and inconsistencies in channel knowledge between access points. Consequently, coherent joint transmission can be compromised, decreasing interference suppression and macro-diversity benefits. Additionally, the overhead for CSI acquisition increases with both the number of users and antennas, becoming a key bottleneck in large-scale deployments. In response, alternative methods such as non-coherent massive MIMO, based on energy detection or differential modulation, have been proposed to reduce CSI acquisition overhead. Although these approaches generally trade some spectral efficiency, they provide better scalability and robustness, especially in scenarios with many users or rapidly changing channels.

4.3. Simulation Parameters and Scenarios

Two representative frequency bands are considered to emulate current and future wireless systems: a sub-6 GHz band at $f_c=3.5$ GHz with a bandwidth of $B=80$ MHz, representative of 5G deployments, and a millimeter-wave band at $f_c=28$ GHz with a bandwidth of $B=100$ MHz, used as a proxy for prospective 6G operation. The total transmit power is fixed to $P_{\mathrm{tx}}=40$ W for all architectures. The simulation framework was implemented in MATLAB R2025b.

The number of simultaneously served users, K, is varied over a wide range to emulate different network load conditions, while the service radius, R, is swept to capture the impact of coverage size on system performance. For each operating point, multiple Monte Carlo realizations are executed, and performance results are averaged to ensure statistical reliability.

To ensure transparency and reproducibility, the key simulation parameters used throughout the benchmark are summarized in

Table 1. Main simulation parameters used in the comparative benchmark.

Parameter	Value
Carrier frequency ($f_c$)	3.5 GHz (sub-6), 28 GHz (mmWave)
Bandwidth (B)	80 MHz (3.5 GHz), 100 MHz (28 GHz)
Total transmit power ($P_{tx}$)	40 W
Channel model	COST-231 urban macro + Rayleigh fading
User distribution	Uniform in circular area
Service radius (R)	100–1000 m (swept)
Number of users (K)	Variable (load-dependent)
Massive MIMO antennas	64 (8 × 8 UPA)
XL-Massive MIMO antennas	256 (16 × 16 UPA)
Cell-Free APs	4 APs × 16 antennas each
Channel state information	Perfect CSI (ideal case)
Monte Carlo realizations	Multiple realizations per operating point

. These values are kept identical across all evaluated architectures in order to isolate the impact of architectural scaling.

For each operating point defined by the number of users K and service radius R, a Monte Carlo simulation framework is employed. Specifically, 30 independent realizations are generated per (K, R) pair to ensure statistical reliability. In each realization, user locations are randomly generated following a uniform spatial distribution within a circular area of radius R, using polar coordinates with uniformly distributed angles and radial distances proportional to the square root of a uniform random variable.

The wireless channel is modeled by combining large-scale and small-scale effects. Large-scale path loss follows a COST-231 urban macro model, while small-scale fading is generated using a Rayleigh multipath channel with multiple delay taps and Doppler effects.

For each realization, the channel matrix is constructed, and hybrid beamforming is applied using zero-forcing precoding. The resulting SINR is used to compute key performance metrics, including throughput, spectral efficiency, coverage, interference, and energy efficiency.

Finally, all performance metrics are averaged over the Monte Carlo realizations to obtain statistically stable results.

4.4. Performance Metrics

To provide a comprehensive assessment of system behavior, five key performance metrics are jointly evaluated. Aggregate throughput is computed as the sum of the achievable user data rates. Spectral efficiency is obtained by normalizing throughput by the system bandwidth. Coverage is defined as the fraction of users whose received signal reference power exceeds a predefined threshold. Interference is measured as the aggregate undesired received power originating from transmissions intended for other users. Finally, energy efficiency is defined as the ratio between aggregate throughput and total transmit power.

• Aggregate Throughput T (bit/s): sum of instantaneous user rates, averaged over Monte Carlo.
• Spectral efficiency: ${\eta }_{\mathrm{SE}}$ (bps/Hz): ${\eta }_{\mathrm{SE}}=T/B$.
• Coverage: $C\in [0,1]$: fraction of users with $\mathrm{RSRP}\ge -95$ dBm.
• Interference I (dBm): total undesired received power induced by beams serving other UEs.
• Energy efficiency ${\eta }_{\mathrm{EE}}$ (Mbit/s/W): ${\eta }_{\mathrm{EE}}=T/P_{\mathrm{tx}}$.

5. Results and Discussion

This section presents and discusses the performance results obtained from the comparative benchmark of scale-up and scale-out MIMO architectures. Results are organized by performance metric to highlight the trade-offs between centralized and distributed deployments as multiplexing and coverage demands increase.

5.1. Throughput Performance

Figure 2 shows the aggregate downlink throughput achieved by the three evaluated MIMO architectures at 3.5 GHz as a function of the number of served users K and the service radius R. As expected, throughput decreases with increasing K for all architectures due to power sharing and residual inter-user interference under linear precoding.

XL-Massive MIMO consistently achieves the highest throughput across the entire $(K,R)$ domain, benefiting from its larger number of spatial degrees of freedom. The increased array aperture improves beamforming selectivity and spatial multiplexing capability, particularly in dense scenarios. Cell-Free MIMO exhibits a smoother throughput decay as K increases, since the distributed access points reduce path-loss disparities and mitigate unfavorable user geometries. Classical Massive MIMO remains competitive at moderate loads but is progressively outperformed as network scale increases.

Figure 3 reports the corresponding throughput results at 28 GHz. Although higher path loss is observed at millimeter-wave frequencies, the availability of larger bandwidth and narrower beams enables substantially higher aggregate throughput. The relative performance trends remain consistent with the sub-6 GHz case, with XL-Massive MIMO achieving the largest absolute throughput and Cell-Free MIMO offering increased robustness to coverage expansion.

5.2. Spectral Efficiency

Figure 4 presents the spectral efficiency at 28 GHz for the three architectures. Spectral efficiency follows trends similar to throughput, but highlights the improved spatial reuse achieved by architectures with higher spatial diversity.

XL-Massive MIMO benefits from its enlarged antenna array to sustain higher spectral efficiency under increasing multiplexing pressure. Cell-Free MIMO achieves comparable or higher spectral efficiency in large service areas, where macro-diversity compensates for reduced beamforming gain at individual access points. These results indicate that distributed deployments can efficiently exploit spatial resources in wide-area scenarios.

5.3. Energy Efficiency

Energy efficiency is a key metric for future wireless systems and is illustrated in Figure 5 for a representative load of $K=100$ users. While XL-Massive MIMO maximizes throughput, its higher processing and beamforming complexity reduce its energy efficiency advantage in wide-area scenarios.

In contrast, Cell-Free MIMO consistently achieves the highest energy efficiency as the service radius increases. By shortening the average propagation distance and distributing transmission power across multiple access points, Cell-Free deployments reduce path-loss and improve power utilization. Classical Massive MIMO exhibits lower energy efficiency due to its centralized nature and less favorable user geometry at larger radii.

5.4. Coverage Performance

Coverage performance is evaluated as the fraction of users whose received signal level exceeds the predefined threshold. Figure 6 shows that Cell-Free MIMO provides the most uniform coverage across the service area, maintaining high coverage probability even at large radii.

This behavior is a direct consequence of macro-diversity, where users are simultaneously served by multiple access points. Centralized architectures exhibit more pronounced coverage degradation with distance, particularly in millimeter-wave operation. These results highlight the suitability of scale-out architectures for coverage-limited deployments.

5.5. Impact of the Number of Access Points

To further investigate the scalability of distributed architectures, an extended Cell-Free MIMO configuration with an increased number of access points is evaluated. In particular, the number of APs is increased from 4 to 16 while maintaining the same total transmit power, thereby redistributing the available power across a larger number of geographically dispersed nodes. This configuration enables a deeper analysis of the fundamental trade-off between macro-diversity gains and reduced per-AP transmit power. Increasing the number of APs enhances spatial densification and reduces the average user–AP distance, but at the cost of lower individual transmission power and increased coordination requirements.

Figure 7 and Figure 8 show the aggregate throughput at 3.5 GHz and 28 GHz, respectively. The results indicate that increasing the number of APs leads to a more stable, spatially uniform throughput, particularly as the service radius increases. While peak throughput values may not increase significantly and, in some cases, slightly decrease due to reduced per-node power, the distributed configuration exhibits improved robustness against distance-dependent degradation, especially in wide-area scenarios.

A similar trend is observed in the spectral efficiency results (Figure 9), where the 16-AP deployment maintains more consistent performance across the entire coverage area, reducing sensitivity to unfavorable user geometries. This behavior highlights the ability of highly distributed architectures to better exploit spatial diversity, even under strict power constraints.

Finally, the energy efficiency comparison in Figure 10 reveals that the 16-AP deployment outperforms both centralized architectures and the 4-AP Cell-Free configuration in wide-area scenarios. By shortening propagation distances and improving power distribution, the system achieves a more favorable throughput-to-power ratio despite the reduced per-AP transmit power.

Overall, these results demonstrate that the number of access points is a critical scalability parameter in Cell-Free MIMO systems. Increasing AP density improves robustness and energy efficiency, particularly in large service areas. However, these benefits come at the cost of increased system complexity, including higher fronthaul capacity requirements, tighter synchronization constraints, and more demanding coordination among APs. Therefore, the optimal number of APs depends on the targeted deployment scenario and requires a balance between performance gains and implementation complexity.

6. Design Insights

Based on the comparative results presented in Section 3, this section summarizes the main design insights derived from the benchmark of scale-up and scale-out MIMO architectures. Rather than focusing on individual performance curves, the following observations condense the results into practical guidelines for the deployment of centralized and distributed MIMO systems in 5G and prospective 6G networks.

Scale-Up Architectures Favor High-Density Hotspots:

Centralized scale-up architectures, particularly XL-Massive MIMO, are best suited for high-density hotspot scenarios where maximizing aggregate throughput is the primary objective. The large number of co-located antenna elements provides increased spatial degrees of freedom, enabling aggressive spatial multiplexing and robust interference suppression. These properties make XL-Massive MIMO particularly attractive for dense urban environments with high user density and relatively limited coverage.

Scale-Out Architectures Provide Superior Coverage and Energy Efficiency:

Distributed scale-out architectures, exemplified by Cell-Free MIMO, offer clear advantages in terms of coverage uniformity and energy efficiency. By exploiting macro-diversity and reducing the average propagation distance between transmitters and users, Cell-Free MIMO mitigates cell-edge effects and improves power utilization. These characteristics make scale-out deployments particularly suitable for wide-area scenarios, coverage-limited environments, and energy-constrained networks. However, distributed deployments also introduce additional system-level challenges, including fronthaul capacity requirements, synchronization overhead, and increased coordination complexity among access points. In addition, the energy consumption associated with fronthaul signaling and network synchronization is not explicitly modeled in this work and may partially offset the energy efficiency gains observed at the transmission level in practical implementations. This additional energy consumption depends on fronthaul technology, signaling overhead, and synchronization requirements, which may vary significantly across deployment scenarios.

Throughput Maximization Does Not Imply Energy Efficiency Optimization:

The results reveal that the architecture achieving the highest aggregate throughput is not necessarily the most energy-efficient. While XL-Massive MIMO maximizes throughput, its centralized processing and beamforming complexity reduce its relative energy efficiency as coverage expands. In contrast, Cell-Free MIMO achieves a more favorable balance between throughput and power consumption, highlighting the importance of jointly considering multiple performance metrics when selecting a MIMO architecture.

Classical Massive MIMO Remains a Viable Baseline but Faces Scalability Limits:

Conventional Massive MIMO remains a competitive baseline for moderate user densities and service radii, offering a favorable trade-off between performance and implementation complexity. However, its relative performance degrades as network scale increases, particularly in terms of coverage and energy efficiency. These limitations indicate that classical Massive MIMO may struggle to meet the more stringent requirements anticipated for future large-scale 6G deployments without architectural evolution.

Architecture Selection Depends on Deployment Objectives:

Overall, the benchmark demonstrates that no single MIMO architecture is universally optimal. Instead, the choice between scale-up and scale-out strategies should be guided by deployment objectives, including user density, coverage requirements and energy constraints. Centralized architectures are preferable for capacity-driven hotspots, whereas distributed architectures are better suited for coverage-driven and energy-efficient network designs.

To facilitate the interpretation of the benchmark results and support practical design decisions,

Table 2. Qualitative comparison of scale-up and scale-out MIMO architectures based on the benchmark results.

Criterion	Massive MIMO	XL-Massive MIMO	Cell-Free MIMO
Architecture type	Centralized	Centralized (Scale-Up)	Distributed (Scale-Out)
Aggregate throughput	Medium	High	High
Spectral efficiency	Medium	High	High
Coverage uniformity	Medium	High	Very High
Energy efficiency	Medium	Medium–High	High
Scalability with user density	Medium	High	Medium–High
Implementation complexity	Medium	High	High
Preferred deployment scenario	Urban macro	Dense hotspots	Wide-area/energy-limited

provides a qualitative summary of the main performance trends observed across the evaluated MIMO architectures. Rather than reporting absolute numerical values, the table condenses the comparative behavior of centralized and distributed deployments in terms of throughput, spectral efficiency, coverage, energy efficiency and scalability. This high-level comparison highlights how scale-up and scale-out strategies differ in their suitability for specific deployment objectives, offering an intuitive synthesis of the quantitative results presented in Section 3.

7. Conclusions and Future Work

This paper presented a comparative benchmark of scale-up and scale-out MIMO architectures for 5G and prospective 6G networks under identical operating conditions. Centralized Massive MIMO, centralized XL-Massive MIMO, and distributed Cell-Free MIMO deployments were assessed under identical channel, power, bandwidth, and traffic assumptions across sub-6 GHz and millimeter-wave frequency bands. By jointly analyzing throughput, spectral efficiency, coverage, interference, and energy efficiency, the study provided a comprehensive view of the performance trade-offs associated with architectural scaling.

The results show that scale-up strategies, particularly XL-Massive MIMO, consistently achieve the highest aggregate throughput, with gains of up to 40–50% compared to conventional Massive MIMO in high-density scenarios. These architectures also maintain stronger interference suppression as the number of users increases.

In contrast, scale-out strategies based on Cell-Free MIMO provide more uniform coverage and improved energy efficiency, particularly at larger service radii. Coverage probability improvements of approximately 10–20% are observed compared to centralized deployments, while energy efficiency gains can reach up to 20–30% in wide-area scenarios.

Classical Massive MIMO remains a competitive baseline under moderate user densities and smaller coverage areas; however, its relative performance degrades as network scale and density increase, particularly in terms of interference management and edge-user performance. These results confirm that no single MIMO architecture is universally optimal: scale-up approaches are better suited for throughput-oriented, high-density scenarios, while scale-out approaches are preferable for coverage-limited and energy-constrained deployments.

It is important to note that the presented benchmark assumes ideal CSI and does not explicitly model fronthaul latency or synchronization overhead in distributed deployments. In particular, the energy cost associated with fronthaul operation and synchronization in Cell-Free architectures may partially offset the energy efficiency gains observed at the transmission level and should be considered in future system-level evaluations.

Future work may extend the current benchmark by incorporating more realistic system constraints, including imperfect CSI, pilot contamination, fronthaul capacity limitations, synchronization overhead, and distributed CSI acquisition. In particular, imperfect CSI is expected to affect centralized and distributed architectures differently: centralized systems may experience reduced beamforming accuracy due to estimation errors, whereas distributed Cell-Free architectures may additionally suffer from inconsistencies in channel knowledge across access points and increased coordination overhead. While these impairments will impact absolute performance, the comparative trends observed between centralized and distributed architectures are expected to remain qualitatively consistent. Furthermore, future studies may investigate how these effects influence the trade-offs between scale-up and scale-out strategies, particularly in terms of scalability, robustness, and energy efficiency. Additional extensions include adopting more realistic channel models, incorporating near-field propagation effects, and exploring advanced resource management techniques. Alternative paradigms, such as non-coherent massive MIMO, may also be considered scalable solutions to mitigate CSI acquisition overhead in dense, dynamic network scenarios. Additional studies may also explore the interaction of scale-up and scale-out architectures with emerging technologies such as reconfigurable intelligent surfaces, hybrid analog–digital beamforming, and user-centric clustering strategies. Finally, experimental validation and large-scale system-level simulations would further support the translation of these insights into practical 6G network designs.