Analysis of Mission Critical Services Radio Access Network Capacity Limitations Over 5G

The evolution of Critical Communications has been conditioned by the underlying networks ranging from previous narrowband almost voice-only Private Mobile Radios (PMR) to now multimedia capable broadband Long Term Evolution (LTE) / fifth generation (5G). Due to the nature of critical communications, the 3rd Generation Partnership Project (3GPP) has defined a series of requirements and recommendations to ensure that availability and performance of Mission Critical Services (MCX) are guaranteed specially in massive group communications. However, no study has been conducted to determine if required capacity and scalability conditions in crisis events can be actually satisfied by commercial or even hybrid LTE/5G networks. This paper presents an analysis of the capacity limitations of the MCX voice service over broadband networks in terms of number of concurrent streams that can be served per cell by comprehensively identifying the different transmission and protocols related bottlenecks. The aggregation of such constraints results in a theoretical model for the maximum capacity of LTE/5G, far more realistic than coarse grain estimations that consider overall cell peak rate. Furthermore, a comparison between MCX over LTE and 5G is provided, indicating the ratio of unusable or wasted radio resources to the total amount available. This way not only actual maximum achievable capacity but also a measure of the efficiency in terms of radio resource usage is provided, which is extremely important for the business model behind hybrid deployments. Both results will contribute towards better network dimensioning as well as possible finer tuning of network and service parameters.


I. INTRODUCTION
Critical communications have traditionally used specific narrowband radio technologies (typically called Private Mobile Radio (PMR) or Land Mobile Radio (LMR)) and have been focused on voice communications.By contrast, commercial mobile networks have long ago evolved towards broadband networks [1].

Despite multiple original concerns about adopting commercial technologies in Public Safety
The associate editor coordinating the review of this manuscript and approving it for publication was Stefan Schwarz .
regarding shared infrastructures and security aspects, this kind of technologies are expected to coexist in commercial networks, or at least in hybrid ones, combining private and public infrastructures.This is driven by the increase of requirements and needs in Public Protection and Disaster Relief (PPDR) communications, looking forward extending traditional voice-centric mission critical communications, towards new applications and devices.Final implementations will be highly influenced by the targeted ratio between isolation and sharing and associated cost and security trade-offs [2].
In order to integrate critical communications in broadband networks, the 3rd Generation Partnership Project (3GPP) defined Mission Critical Services (MCX) over the Long Term Evolution (LTE) wireless standard back in Release 13.These specifications were extended by adding new functionalities and requirements in later releases till Release 18 and beyond, including MCX over fifth generation (5G) new radio.3GPP defines three main MCX: Mission Critical Push-To-Talk (MCPTT) [3], Mission Critical Video (MCVideo) [4], and Mission Critical Data (MCData) [5].The relevance of these standards has increased through last releases, while more vendors are getting interested on this technology (from device manufacturers to service providers), participating actively in MCX Plugtests for interoperability organized by the European Telecommunications Standards Institute (ETSI).Governments have also understood that this technology is an enabler to improve public safety communications networks as FirstNet in the US, ESN in the UK, and SafeNet in South Korea.MCX has also been the base for other new promising technologies like Future Railway Mobile Communication System (FRMCS), extending mission critical features to the railway context.
One of the cornerstone requirements of the MCX standards is supporting massive group communication between multiple users (group members).Unlike traditional Voice Over LTE (VoLTE) and Voice Over 5G New Radio (VoNR), in MCPTT and MCVideo, each member has the ability to gain access to the permission to talk in an arbitrated manner and typically only one transmits simultaneously.
In order to satisfy the MCX requirements in terms of KPIs and priority specific mechanisms must be applied in hybrid networks where radio resources are shared among different kinds of users and traffic.Following the Policy and Charging Control (PCC) architecture, the 3GPP has defined a network initiated dynamic reservation of dedicated mobile network resources to guarantee the transmission quality of mission critical media streams, using newly defined Quality of Service Class Identifier (QCIs) and 5G Quality of Service Identifier (5QI) values [6], [7].These QCI/5QI are associated to a Packet Delay Budget and a Packet Error Rate, providing to MCX streams higher priority over other traffic in network overload scenarios and specific priority for the Allocation and Retention Priority handling mechanisms.In terms of performance, for MCPTT QCI/5QI value of 65, a Guaranteed Bit Rate (GBR) bearer is defined, expecting maximum end-to-end delays of 75 ms with an error rate lower than 1%, needed in order to fulfill the demanding Key Performance Indicators (KPIs) required for MCPTT, like the 300 ms mouth-to-ear delay.These time and availability requirements are essential in Public Safety operations, where each User Equipment (UE) minimal disconnection or latency could reduce reaction time in disaster or emergency situations.
Additionally, the need for guaranteed connectivity for MCX further resulted on the definition of different mechanisms to ensure access to the network, including usage of enhanced Access Class Barring (eACB), Quality of Service (QoS), Priority and Preemption (QPP) and even Isolated Operation for Public Safety (IOPS) for disaster situations where even the backhaul is lost.Multicast and Broadcast Services (MBS) were also introduced to improve the performance and capacity of Radio Access Network (RAN), when same traffic is shared by multiple devices through same physical channels.Although those mechanisms could guarantee MCX users required connectivity in hybrid commercial-mission critical radio networks hybrid scenarios, the actual capability of the RAN even for mission critical only users (i.e. for full mission critical mobile network operators) remains a challenge and focuses our paper, especially when some of the evaluations of these mechanisms, like multicast, are currently still under study along different 3GPP Work Items (WIs) [8] and they are not widely deployed over the commercial networks of main front-runners.MCX over 5G multicast is also being currently addressed by the 3GPP Technical Specification Group Service (TSG) and System Aspects (SA) Working Group (WG) 6 under the WIs FS_MC5MBS and MCOver5MBS [9].
In parallel to these developments, the 3GPP has already identified the increasing relevance of AI/ML in mobile broadband network management.In Release 18 there exist multiple WIs devoted to how to make AI/ML available not only for internal 5G Network Functions (NFs) but also for third party services.Some of the WIs are related to RAN features, like FS_NR_AIML_air by WG RAN1, and NR_AIML_NGRAN by WG RAN3 [10].Other ones are related to 5G core and services functionalities, like FS_AIML_MT, and AIML_ML by WG SA1 [11].These are ongoing efforts to deliver promising features and functionalities throughout the development of Releases 18 and 19, but still focus on the AIML architecture and its deployment, training, and evaluation frameworks.The research presented in this paper will help future AI/ML based management techniques by providing them with more accurate system modeling capabilities and a better understanding of the bottlenecks and tuning parameters across the whole MCX service chain.
Therefore, our work focuses on the analysis of the maximum capacity of MCPTT when deployed over both LTE and 5G networks.The main objectives are to analyze the capacity in RAN of MCX, in particular over 5G, making a comparison with LTE, and to check to which extent the 3GPP requirements can be satisfied in conventional setups, i.e., without specific tuning or using multicast (eMBMS/5MBS).The study covers the voice service (i.e., MCPTT), assuming it is the natural evolution of legacy critical communications and the service to prioritize (MCVideo requires audio and video, consuming more resources; MCData traffic is heterogeneous, and calculations vary based on the type of data).In this way, a baseline scenario for future research in optimization and development of standards and functionalities for the next generation of wireless networks can be defined, identifying the main bottlenecks that would need to be solved and the major thresholds to be surpassed.
To this end, the rest of this paper is organized as follows.Section II, provides an overview of related work regarding LTE and 5G downlink capacity limitations specially when supporting voice services over broadband networks.Then, in Section III, maximum MCPTT capacity estimation mechanisms are presented.Next, Section IV presents MCPTT concurrent call capacity calculations results for LTE and 5G networks assuming similar radio configurations, also providing a comparison between the results for both environments.Finally, Section V summarizes the main conclusions and future research lines.

II. RELATED WORK
MCX involves all multimedia services including voice, video, and data.For the voice service (MCPTT) some characteristics are shared with VoLTE/VoNR: the codecs defined for MCPTT (AMR-WB) and the use of (S)RTP for the audio streams are remarkably similar to those in VoLTE/VoNR, establishing a common scenario with similar media characteristics, allowing a proper comparison and the generalization of same calculations for multiple services.
In fact, traditional voice services over broadband networks have been analyzed in several studies but focused on time restriction and peak data rate based KPIs.VoLTE analysis based on latency and link speed is described for example in [12].Similarly, VoNR analysis is covered in [13], based on latency, link speed and coverage.However, both calculations show maximum rates and determine the amount of data for each service that could be provided if every endpoint is available and there are no RAN capacity bottlenecks.Unfortunately, they do not address the impact of multiple (eventually thousands) concurrent users being served exactly at the same time, which is one of the aims of this paper.
An estimation of VoLTE capacity per cell has been calculated and considered in [14] and [15], where only data Physical Resources Blocks (PRBs) based analysis is developed, without considering possible limitations due to control channels.5G capacity calculations have been indirectly introduced in [16], presenting control channel resources and some simplified VoNR capacity calculations while comparing results with VoLTE.A more detailed study is provided in [17], where Blocking Probability of downlink control channel is obtained.Control mechanisms impact on VoNR is presented in [18], including some results from real-world tested scenarios.Nevertheless, none of above-mentioned studies provide detailed maximum capacity calculations, including both data and control channels.
MCPTT KPIs and time limitations analysis has been explicitly studied for LTE in [1] and [19], and for 5G in [2] and [20].MCPTT scalability and limitation related to the amount of concurrent supported UEs are presented for LTE in [21].In the case of 5G technology, this study has not been developed yet and it is required to provide an extended analysis.
In summary, although sharing some similarities, VoLTE/VoNR and cell capacity related existing work does not provide a detailed analysis of MCX capacity over 5G, so it is necessary to elaborate additional calculations to define its performance and requirements, determining if the 3GPP capacity requirements and recommendations are satisfied.In this way, maximum capacity limitation needs to be presented in a more realistic scenario, evaluating the impact of both 3GPP control and data mechanisms and protocols, determining the main sources of bottleneck.

III. MCPTT CAPACITY ANALYSIS
The considered MCPTT voice service capacity refers to the maximum number of simultaneous clients in active calls (i.e., involved in a group call) that can be served in the same LTE/5G cell depending on specific radio and service conditions.In order to do so, MCPTT application voice traffic is first characterized.Then, broadband network Mission Critical quality mechanisms and requirements are introduced.Finally, increasingly more accurate capacity calculation approaches are outlined.

A. MCPTT MEDIA CHARACTERIZATION
In terms of the characteristics of the media streams, the 3GPP selected the Adaptive Multi-Rate Wideband (AMR-WB) audio codec for MCPTT voice [22].AMR-WB operates with nine different modes, each one corresponding to a different bit rate.The typically chosen mode for MCPTT is 12.65 (corresponding to a rate of 12.65 kbit/s), providing enough speech quality in a clean environment.Thus, 12.65 mode with an audio frame sampling time of 20 milliseconds (ms) is selected as reference for the following calculations in this paper.After including subsequent protocol overhead related to Real-time Transport Protocol (RTP), User Datagram Protocol (UDP) and Internet Protocol (IP), the resulting required MCPTT bit rate over the network layer is 29.2 kbps for IP version 4 (IPv4), and 37.15 kbps for IP version 6 (IPv6).
To extend the packet size calculation overhead for remaining link and physical layers needs to be added.For both LTE and 5G, there are common protocols to be applied for link layer: Packet Data Convergence Protocol (PDCP) and Radio Link Control (RLC).In the case of 5G an specific sub layer in data link layer is further included, named Service Data Adaptation Protocol (SDAP).Resulting packet size, over physical layer, to be scheduled and mapped into radio resources is shown in Figure 1, where total size in bits depends on the used AMR-WB codec, which defines the number bits of each audio speech frame (X), and the corresponding IP version used.

B. MCPTT CAPACITY CALCULATIONS
The majority of studies use network-layer-measured packet size and represent related capacity usage in length (bits), but radio resources are mapped and divided into frequency-time domain elements.Different capacity limitation calculation approaches are here introduced.Estimation strategies are described starting from the more coarse-grained approaches to the most accurate ones, considering mobile network scheduling and radio resources limitations.The final target is considering the radio resources usage on physical layer, so that for each Modulation and Coding Scheme (MCS) and broadband technology, the number of Resource Blocks (RBs) consumed per MCPTT packet will be calculated.
Prior to detailing each calculation, the following relevant assumptions are detailed: 1) All single cell's radio resources are either available or reserved for MCX, simplifying calculations while isolating the mission critical services to analyze their specific limitations.Thus, it is assumed that only mission critical group call users corresponding UEs are active in the broadband network.Note that this scenario is actually plausible in hybrid networks where prioritization and preemption capabilities would enable MCX users to allocate all available resources in extreme congestion situations.2) Regarding MCPTT traffic characterization, the inherent group communication (one to many) nature results on most of the UEs only receiving media, so that downlink channel is the main bottleneck in terms of raw network capacity, when compared with VoLTE/VoNR where the uplink would typically be the constrained one due to its reduced capacity and voice symmetric nature.Control information is also mainly transmitted over the downlink.Therefore, in the following sections and calculus only downlink channels capacity is considered.Another commonly used technique is the outer loop link adaptation (OLLA), which corrects the inaccuracies and delays associated with CQI reporting [23].For the purpose of this paper, it has been assumed that these mechanisms allow to map each CQI to a single MCS value, based on modulation, spectral efficiency, and code rate values defined for each MCS index and each possible CQI.Normally, the farther away from the eNodeB/gNodeB, the lower the MCS and the higher the number of PRBs that are required to transmit the same information with a higher ratio of bits used for error control mechanisms.RB units are defined as 12 subcarriers in frequency domain per each time symbol, and PRB units are defined as 1 RB per 14 symbols.Even a UE placed close to the station could have a low MCS if there is high attenuation or interference in the radio signals.The MCS values are obtained from tables in section 7.1.7.1-1 in [24] for LTE and for 5G in section 5.1.3.1 in [25], assuming UEs only support 64QAM as maximum modulation to achieve the greatest possible interoperability.3) (S)RTP is used (disabling/enabling security) to analyze the impact of overhead due to encryption and security parameters.Robust Header Compression (ROHC) is also studied, reducing IP, UDP and RTP headers related overhead.AMR-WB 12.65 mode is used as audio codec.IPv4 and IPv6 are included, to check their overhead impact.Applying these protocol assumptions separately, the following scenarios have been defined: • Case A (Smallest Packets -best case scenario-): RTP, ROHC, IPv4.
First calculation approach, in Section III-B1), corresponds to the over-simplified scenario and considers the peak eNodeB/gNodeB bit rate in the specific defined cell configuration.Second calculation, in Section III-B2), considers the resource usage related to data transmission.Third calculation approach, in Section III-B3), only considers limitation related to resources usage linked to control of data transmission.Fourth approach, in Section III-B4), the more complex and realistic one, considers every other calculation to analyze the more conservative scenario.

1) MAXIMUM THROUGHPUT (TH)
This approach is the most commonly used in commercial discussions, where maximum achievable peak rate in an LTE or 5G cell (ranging typically from 100s megabits to 10s gigabits per seconds) seem to be enough to easily support several thousands of MCPTT audio streams.This analysis is an oversimplified theoretical approach, considering that every resource is available to transmit audio data, without requiring upper levels of control for radio and network protocols.
Therefore, the maximum number of concurrent calls could be calculated dividing the maximum available throughput, or cell peak rate, between the required throughput by each MCPTT audio stream.Cell peak bit rate is calculated applying the 3GPP specifications and equations of Section 4.1.2in [26].The corresponding pseudo-code sequence is shown in Algorithm 1 where the result is MCPTT users .
Previously, MCPTT traffic characterization has concluded that each active mission critical user requires a specific static bit rate for each AMR-WB mode.Therefore, MCX capacity in this theoretical approach depends only in the base station configuration and the MCS value assigned to the mission critical UE.

2) PHYSICAL DOWNLINK SHARED CHANNEL (PDSCH) LIMITATION
PDSCH capacity limitation based calculation approach considers how the MCPTT voice data is structured inside the shared downlink data channel.To do so, it computes the number of PRBs dedicated to data transmission that are required by a mission critical UE to be able to receive a MCPTT audio stream correctly.The number of PRBs required to receive audio data is defined by the assigned MCS value to the UE, based on minimum Transport Block Size (TBS) within which audio data fits.Meanwhile, the base station total available PRB number for each TTI, only depends on the specific cell configuration, defined by the 3GPP specification.PRBs, MCS, and TBS values can be obtained from applied configuration, using tables and formulas provided by 3GPP's Technical Specifications.In LTE, MCS values are mapped into TBS indices using Table 7.1.7.1-1 in [24].These indices make it possible to directly obtain the number of required PRBs from Table 7.1.7.2.1-1 in [24], when packet size in bits fits within TBS indicated by the table.In 5G, this operation is more complicated, being necessary to apply a combination of equations and tables defined in Section 5.1.3.2 in [25].The specific sequence is provided in pseudo-code shown in Algorithm 2 where the obtained value is MCPTT users .

Algorithm 1 Maximum Throughput
Based on Section III-A and Section III-B assumptions, it is possible to identify the number of PRBs that are required per MCPTT audio packet and the total amount of available resources, according to the assigned MCS value.For each MCS and specific case (A, B and C), the minimum number of required PRBs per MCPTT packet in both LTE and 5G, is presented in Table 1.
Thus, the maximum number of concurrent UEs being served can be calculated using the number of PRBs required to transmit a MCPTT voice packet and the total number of available PRBs in the cell.The evaluation period has been defined to be equal to the MCPTT audio frame sampling time (20 or 40 ms).When computing the number of voice frames that can be scheduled per TTI, it is assumed that, due to RLC fragmentation, a voice frame can be allocated in PRBs corresponding to different radio sub-frames, allowing a more efficient scheduling mechanism, reducing the number of wasted PRBs [30].
This estimation analyses only those limitations due to raw data transmission capacity in the PDSCH channel for different MCS (or combinations of those).However, in order each UE to index and decode its corresponding information within the shared PDSCH channel additional control related signaling is needed, possibly resulting in additional constrains.Such control mechanisms (part of the so called PDCCH channel) are introduced in the following section.

3) PHYSICAL DOWNLINK CONTROL CHANNEL (PDCCH) LIMITATION
PDSCHs are by definition shared channels, meaning they convey data packets for different receiving UEs in the downlink.Some mapping is needed in order to identify which set of PRBs among all in the radio subframe correspond to a data packet for a particular UE.In order UEs to be able to locate and decode properly the PDSCH structure, mapping Downlink Control Indicators (DCIs) information is transmitted over PDCCH.In both LTE and 5G, DCI information is mandatory to enable each UE indexing or detecting the location of the information transmitted over PDSCH and decoding its data.
Therefore, the capacity limitations due to control mechanisms can be defined as the maximum number of indexable targeted UEs in the PDSCH by the control information in the PDCCH.More specifically those indexable users related constraint is the maximum number of UEs whose data position in the PDSCH radio resources can be addressed by a valid DCI in the PDCCH.Depending on the Aggregation Level (AL), each DCI is allocated in one or multiple Control Channel Elements (CCEs).Each CCE is composed by multiple Resource Element Groups (REGs), in turn made up of groupings of REs, which number vary for LTE and 5G.To simplify the calculations, a relationship between MCS and AL has been determined as a function of whether the radio conditions are optimal (22)(23)(24)(25)(26)(27), average (15)(16)(17)(18)(19)(20)(21), bad (7)(8)(9)(10)(11)(12)(13)(14) or very bad (0-6).The corresponding MCS-AL mapping is defined in [18], as 2 for average conditions, 4 for bad, and 8 for very bad.To extend the analysis, an additional value of AL equal to 1 has been added for the optimal case, when users are really close to base stations and there is no interference or attenuation.To obtain the maximum number of indexable users, it is assumed that each DCI is mapped into one individual CCE (i.e., AL is set to its lowest value).This assumption is taken because higher AL values increase the resources consumed, therefore reducing capacity.
It should be mentioned that PDCCH indexable UEs capacity has been made more flexible in 5G, providing an adaptive way of resource allocation for control information on each TTI, configurable per UE.Each UE uses Search Spaces (SS) to locate control information, where PDCCH candidates are sought to be blind decoded.SS are formed by contiguous sets of CCEs, which number is established by AL.The CCE AL is given by the PDCCH format and determines the number of PDCCH candidates in a search space.In other words, PDCCH candidates point to specific SS location, with a size of AL number of CCEs, in the Control Resource Set (CORESET).In conclusion, CORESETs are a set of physical resources, configured in time-frequency domain, which are used to transmit the control information, where PDCCH and DCI data are looked for decoding for each monitoring occasion [31].In the case of LTE, PDCCH occupies the full bandwidth, not being as flexible as 5G, where CORESETs allow dividing each channel bandwidth into multiple parts, differentiating control information per UEs individually or per groups.
PDCCH can be formed by one to three OFDM symbols per TTI.In other words, by one to three RE per TTI, in time dimension, multiplied by the number of subcarriers used in frequency dimension.Note that, in order to reduce the likelihood of the PDCCH becoming the capacity bottleneck, it has been assumed to use as many resources for control region as possible (i.e., three OFDM symbols PDCCH configuration).
Calculations are based on the total resources of the control region for each evaluation period, divided by the number of RBs required for each UE PDCCH information.For this purpose, each UE consumption is defined by the number of CCEs required for PDCCH information.The number of REs composing each CCE is obtained establishing an AL value equal to one, for LTE and 5G separately, trying to minimize the consumption of resources per UE data.PDCCH CCE limitation calculations are provided in [18].A pseudo-code implementation is provided in Algorithm 3, where CCE AL is the number of CCE per AL, REG CCE is the number of REG in each CCE, RE REG is the number of RE for each REG, and the calculated value is MCPTT users .
In this approach, PDSCH raw capacity is not contemplated, and the only limitation is the PDCCH related one.Numerical values for the number of REs and bundles of these frequency-time domain resources vary according to the technology used.Parameters for both broadband standards are summarized in Table 2, where PDCCH calculation results per time span of 20 ms are also included.
These maximum values are the baseline of the analysis, considering the optimal conditions and using the highest MCS value for all the UEs, an improbable situation in most cases.
It is also necessary to consider an additional limitation provided by the 3GPP for 5G technology, to reduce UE decoding complexity and provide power saving features for devices.This limitation is defined in Table 10.1-2 in [33], establishing the maximum number of Blind Decodes (BDs) to a value of 44, when using numerology equal to zero.This means that a UE could monitor a maximum of 44 PDCCH candidates per slot and serving cell.Assuming the lowest AL value, the maximum number of CCEs that could be decoded would be 44 too, being possible to be less than the maximum allowed by radio resources.Both values are included in this analysis for 5G.The maximum capacity enabling BDs restriction is represented by ''3GPP(maxBDs)'' in figures of Section IV-B.

4) COMBINED LIMITATION (COMBINED)
This final approach takes into consideration the reduction of useful resources caused by PDSCH limitations and the overhead originated by control mechanisms and PDSCH scheduling, indicated by PDCCH information.As a result, it represents the expected real capacity limitations in currently deployed services over broadband networks.
Depending on the radio conditions and UE location, the MCS assigned will define which approach is limiting, varying for LTE and 5G.In any case, PDSCH or PDCCH limitations will overcome traditional extended maximum throughput theoretical approach, providing a model that is closer to reality.

IV. RESULTS
In the current section, the simulation results of MCX audio service capacity over broadband networks for each considered scenario (A, B and C) are presented.First, MCX capacity over LTE is evaluated, setting the starting point of the analysis for newer generation networks, being the most extended and implemented broadband network until mid-2023.Then, aiming to quantify the mobile broadband technology evolution and its potential improvement for MCX capacity, 5G capacity performance is presented, under the same radio configuration, analyzing also the 5G base functionalities provided by the 3GPP, like numerology.Finally, a comparison for both technologies is provided, analyzing broadband networks evolution's impact in MCX solutions.
MCPTT over LTE capacity study will demonstrate that there exist capacity limitations, due to the fact that the 3GPP targets cannot be met.In addition, the comparison of the results of the different calculation methods aims to demystify throughput based quick calculations and provide a clearer view of broadband networks actual limitations.Eventually, the assessment of 5G vs. LTE comparison intends to show if MCX capacity requirements are finally satisfied or otherwise determine that there is still a limitation despite 5G new specifications and features.
In order to provide a clear perspective about required MCPTT capacity values, the 3GPP provides a guide for scalability in MCPTT, including numeric values which are used as a reference in further sections [3].There are multiple statements but, for this paper purposes, the selected ones that will establish the base requirements are:

1) ''The MCPTT Service might support an MCPTT Group
Call where all the Participants are located in one cell of the MCPTT system.'' 2) ''The MCPTT Service might support a range of 36 to 150 simultaneous MCPTT Group Calls in every cell of the MCPTT system per regional regulatory requirement.''3) ''The MCPTT Service might support a minimum of 2000 MCPTT Users within an MCPTT Group or a combination of different MCPTT Groups, in every cell of the MCPTT system.''This last statement constitutes the threshold for determining that the service meets the scalability requirements.This threshold is used as baseline in all MCPTT figures and calculations.

A. MCPTT CAPACITY OVER LTE
The total number of concurrent MCPTT calls in an evaluation period of 20 ms for LTE technology, considering PDSCH and PDCCH limitations over maximum theoretical throughput, is calculated for every possible MCS and presented in Figure 2.Only when considering the most optimistic scenario, with maximum throughput ideal approach and the best radio conditions assuring the highest MCS value, the number of concurrent users is higher than the 3GPP threshold recommendation of 2000.The most realistic scenario does not allow supporting enough users to satisfy the targets and some additional tuning or enhancements would still be necessary to be applied to get closer to the desired threshold.
From the results, it is possible to identify ranges where capacity limitations are clustered around same values.These MCS ranges behavior for the number of users is: 0-8 (increases linearly gradually), 9-15 (stable around 400), 16-17 (stable about 550), 18-26 (remains around 800) and 27-28 (maximum calculated for PDCCH limitation approach, 1633).This behaves according to the values in the MCS tables provided by the 3GPP [24], where Modulation Order changes approximately in the same range values.
The main source of bottleneck for case A is the PDCCH, due to the small size packets and inability of the PDCCH channel to map all such assignments.In cases B and C, the source of bottleneck is PDSCH, not having enough resources to transmit bursts of larger packets.This analysis matches statistics gather in a real-world massive event depicted in Figure 3.As shown during the busy hours of the event involving a significant concentration of first in the same cell, a vast majority of the downlink transfer requests are dropped due to unavailability of signaling resources in the PDCCH leading to congestion.

B. MCPTT CAPACITY OVER 5G
Same calculations are done for 5G, and results are presented in Figure 4.The maximum theoretical throughput approach results are higher than LTE as expected.On the other hand, the realistic analysis that combines PDSCH and PDCCH limitations shows that results have worsened, being able to support slightly more than 1000 users only for the highest MCS values.
Results show again a clustering of number of users in certain ranges of MCS values: 0-6 (hardly changes, around 120), 7-15 (stable around 260), 16-25 (stable about 520), 26-27 (remains at maximum calculated for PDCCH limitation approach, 1060).This behaves according to the values in the MCS tables provided by 3GPP [25], where Modulation Order changes approximately in the same range values, affected also by code rate, determining different spectral efficiency values.
Another observation from results is that generally the PDCCH constraint is prevalent, for all three cases.This is explained by the fact that MCPTT packets have small size related to the total amount of resources available per evaluation period.Therefore, indexing many UEs receiving small packets is more restrictive than indexing few UEs with large packets.
While the per-MCS value allows identifying the main bottleneck, in order to simulate the average statistical behavior of users in a mixed scenario, a simple Monte Carlo method has been applied, generating the MCS values based on a uniform distribution for the corresponding ranges of the different radio signal conditions (optimal, good, bad, and very bad).One additional scenario has been also analyzed, when the uniform distribution is performed over the complete set of MCS values.The results of these simulations for 50 iterations of Monte Carlo experiment are provided in Figure 5, distinguishing between the smallest packet size (Case A) and the largest packet size (Case C) scenarios.As the number of supported users corresponds to an average number of users in bad radio conditions, it is observed a considerable decrease in the overall MCPTT capacity over 5G.
In the case of 5G, standards include native features to provide more flexibility and dynamic configuration, like above-mentioned CORESETs, for control information, or numerology, for frequency-time resource configuration.This is one of the promising features to support massive devices with low latency requirements in 5G.For the same frequency range and channel bandwidth, it is possible to configure additional parameters referred to the physical waveform characteristics.Numerology value allows varying the Subcarrier Spacing (SCS) (e.g., permitting operating over higher frequency bands if increased), which automatically implies modifying, inversely proportional to the SCS, the OFDM symbol duration time.In LTE this SCS is fixed to 15 kHz.The comparison between LTE and 5G applying lower numerology values is presented in Figure 6.
Results show that, although counter-intuitive, the capacity is actually degraded, and this is caused by the increase in guard bands required for higher SCS values.In the case of numerology value of 0, the number of RBs per symbol is 106 for 20MHz bandwidth.If SCS is doubled, i.e., using numerology value of 1, the number of available RBs per symbol is 51.For numerology value of 2, RBs per symbol number is 24.
Therefore, despite doubling the number of slots per TTI, in global calculations, usable RBs are lost in order to dedicate more resources to the guard band.This demonstrates that 5G numerology feature does not help to provide more UE capacity for MCX and even increases the detected limitation.
In the case of 5G, the 3GPP MCPTT requirements and recommendations are also not met and additional enhancements (such as multicast or configured vs. dynamic scheduling) should be considered.In fact, using inherent features provided as the solution for massive low latency devices support is even counterproductive.

C. MCPTT RADIO USAGE EFFICIENCY OVER LTE AND 5G
Another interesting approach is to analyze the amount of available radio resources that are unusable because of the above-mentioned capacity limitation.For each 20 ms period defined in previous analysis, the total amount of radio resources is defined as the maximum number of REs in PDSCH that could be used to transmit user data in the optimal scenario.Based on the limiting factor, unusable resources are calculated differently.When PDSCH limits the capacity, unusable resources are the REs not used when every single user data has been attempted to be allocated and no more packets can be fitted in PDSCH remaining resources.If PDCCH is the limiting capacity, unusable resources are obtained subtracting the amount of REs required for every single user data transmitted and indexed by PDCCH control information, to the total amount of available REs.These results are calculated for LTE and 5G, obtaining the results in percentage units.Results are shown in Figure 7, where the increase of unusable radio resources for 5G technology can be appreciated.
These results highlight the need to apply some enhancements to increase the efficiency of radio resources  usage, while increasing the capacity of broadband networks, allowing communication between more UEs.
To demonstrate the room for improvement, RTP bundling has been studied in the three defined cases.The procedure to calculate the number of bits required for each MCPTT packet using RTP bundling is shown in Figure 8.
The number of additional users that could be served in same radio conditions enabling RTP bundling of 2 samples (or 40 ms audio frames) are shown separately for LTE and for 5G in Figure 9.Using this technique, it is not required to add full overhead for single audio frames, using same headers for bundled groups of audio frames, allowing an increase of capacity.This method produces major improvements in PDCCH limited cases, due to the reduction of the burst of small packets, distributing data through PDSCH requiring less PDCCH information.

D. MCPTT OVER 5G SYNTHETIC MOBILE WIRELESS NETWORK TRACES
Due to the difficulties in implementing a full real scenario where the analyzed metrics and parameters would be isolated and out of unwanted interference (scheduling, access control and other mechanisms that would introduce noise) with such a high number of MCX users, the calculations have been applied on a synthetic generated dataset of 5G mobile radio networks using mobility patterns [34].This dataset provides a list of MCS values assigned for each UE for each timestamp, over a period of 6 hours.For the purposes of this paper, it has been assumed that each user of the dataset is an MCX user, and that each UE is connected to the same cell, so that the capacity of MCPTT over 5G RAN can be analyzed.The results of this experiment are provided in Figure 10, providing the percentage of unusable radio resources and the total amount of concurrent users per timestamp, for Cases A (best case) and C (worst case), with and without using RTP bundling (case of 40 ms).
As can be observed, the average maximum number of concurrent users is well below the recommended threshold.The results are quite consistent with the results of the Monte Carlo simulation for random MCS values over the complete set (0-27).Although the number of concurrent users is similar for both cases, the percentage of unusable resources is much lower in the Case C.This is because, for the same number of concurrent users when PDCCH is limiting the performance, larger packets will use more available resources of PDSCH.This is a further indicator that combining audio frames and packets would help using the radio resources in a more efficient way, because of the increase in the ratio between the speech transmitted data and the total size of control overhead per user.That is demonstrated applying RTP bundling, in this case for 40 ms speech frames bundles.In this way, it has been obtained an average increase of 300 concurrent users in calls, while reducing the wastage of radio resources for a 15%, in both cases and in the same service and radio conditions.

V. CONCLUSION
This paper presented an extended analysis of capacity limitations in Radio Access Networks over 5G, comparing results against LTE.Results show that MCX is still constrained over Broadband networks for massive group calls.This limitation is present due to control protocols mechanisms used to index data in the downlink, regardless the broadband technology used.However, the impact of control mechanisms limitations is greater in 5G, where more resources are required to transmit control information.
Initially, it was expected to achieve a better performance and capacity in 5G, due to its new features (higher throughput and flexibility of control information).However, the calculations and simulations show that they do not provide any significant improvement for the same bandwidth and baseline scenario.Moreover, they penalize performance to an even greater extent under the same radio conditions.Furthermore, if we think about actual private and hybrid deployments for MCX, we will probably obtain a higher raw capacity for 5G RAN, using additional frequency bands, carrier aggregation, mm-WAVE and so on.However, this would also increase the total amount of wasted resources, energy consumption and radiation level, requiring more equipment and its maintenance, but remaining the same inefficiency problems.
At this point, there is an undeniable need to incorporate additional enhancements (multicast radio transmission (eMBMS/5MBS), header compression, SPS -dynamic scheduling-or AL optimization) to satisfy the 3GPP requirements and recommendations, ensuring communication for every user in massive emergency scenarios.Additionally, media enhancements include RTP speech frame bundling and dynamic codec selection.
Another theoretical proposal is to allow dynamically configuring the number of maximum BDs defined by the 3GPP using additional Radio Resource Control (RRC) mechanisms.Increasing this value, would reduce the limitation due to the maximum number of PDCCH candidates.Reducing it, would reduce the power consumption when the devices does not require to decode many PDCCH candidates.
These studies are defined as the future lines of research of author's group, looking forward optimizing MCX over broadband networks.This optimization seeks to prove dynamic solutions adaptable to each scenario characteristics and functionalities, reducing exposed capacity limitation, introducing Artificial Intelligence (AI) technologies and Machine Learning (ML) algorithms.

FIGURE 1 .
FIGURE 1.Size in bits of MCPTT audio frame adding overhead headers from application layer to physical layer.

FIGURE 2 .
FIGURE 2. Maximum number of concurrent MCX UEs supported in group calls per MCS and 20 ms in LTE.

FIGURE 4 .
FIGURE 4. Maximum number of concurrent MCX UEs supported in group calls per MCS and 20 ms in 5G.

FIGURE 5 .
FIGURE 5. Maximum number of concurrent MCX UE supported in group calls per MCS in 5G for 50 iterations in Monte Carlo Simulation.

FIGURE 6 .
FIGURE 6. Maximum number of concurrent MCX UE supported in group calls per MCS in LTE compared with 5G numerology values in FR1 for Combined Approach.

FIGURE 7 .
FIGURE 7. Percentage of Unusable Radio Resources when maximum number of concurrent MCX UE supported in group calls per MCS is reached in LTE and 5G for Combined Approach.

FIGURE 8 .
FIGURE 8. Size in bits of MCPTT audio frame adding overhead headers from application layer to physical layer using RTP bundling.

FIGURE 10 .
FIGURE 10.Percentage of Unusable Radio Resources and maximum number of concurrent MCX UE supported in group calls per MCS in 5G using synthetic dataset for mobile wireless networks, with and without RTP bundling application.

TABLE 1 .
Minimum number of PRBs required for each MCPTT packet for audio frames and evaluation period of 20 ms in LTE and 5G for cases A, B and C.

TABLE 2 .
PDCCH resource elements numerical parameters for LTE and 5G per time span of 20 ms.