Sponsored by:
5G Network Analytics & Automation Operator Survey
This report presents the results of the Heavy Reading 5G Network Analytics and Automation Operator Survey conducted in May 2023. Heavy Reading’s first survey focused on this topic provides insight into network operator views on 5G network automation and analytics.
5G RAN, cloud native technology, and disaggregation introduce agility, scale, and flexibility to the network, but they also present complex challenges for mobile operators. Advanced analytics and automation systems are fundamental to providing real-time service management, generating operational insights, and supporting lifecycle changes. The 5G RAN and 5G core analytics and automation ecosystem are diverse and evolving rapidly.
5G service operation has demanding requirements. To gain real-time actionable insights, operators require the ability to analyze and process numerous network data feeds efficiently across the entire network. Analytic and automation solutions driven by artificial intelligence (AI) and machine learning (ML) will simplify and enhance operational efficiency and support demanding performance requirements. Furthermore, continuous software release cycles in 5G will rely on automation and analytic processes to support the cadence of cloud native network functions and to compute infrastructure updates.
This report aims to help the industry better understand the status of network analytics and automation and provide insights into operators’ strategies.
5G SA DEPLOYMENT AND SERVICES
5G SA timeframes and service selection have a significant impact on operator analytics and automation choices within their network. Today, most 5G public wide-area networks are NSA, with the complexity of new cloud native technology, the capabilities of 5G RANs, and the availability of compatible devices often slowing the transition to 5G SA. However, the frequency of SA launches is increasing. Figure 6 aims to establish a timeline for 5G SA public wide-area network deployment.
A combined 58% of respondents confirm 5G SA public wide-area network support within 12 months, based on “already supported” (28%) and “within 6–12 months” (30%). About a fifth (21%) say they will have live 5G SA “within two years,” and the remaining smaller groupings split between “within three years” (10%) and “no plans/don’t know” (10%). US respondents are more optimistic (when studying region variation), with a combined 70% determining live public wide-area 5G SA networks are “already supported” or will be available “within 12 months” compared to the RoW with 48% for this combined total.
This is more optimistic than the present worldwide outlook, and Heavy Reading interprets these timelines to include some of the recent and upcoming 5G SA soft launches to restricted numbers of users and within smaller defined geographic areas before full rollout. Yet, 5G SA public wide-area network launches are starting to occur more frequently.
5G SA heralds the arrival of new advanced mobile services that offer operators new opportunities to recover the costs of their significant spectrum and network investments. Figure 7 shows the weighted scores for respondents ranking which services are most attractive for 5G revenue growth in their organizations.
“Mobile broadband subscriber growth/retention/higher prices” is deemed the most attractive 5G service for revenue growth, ahead of private 5G and network slicing, in second and third places, respectively. Connected devices rank fourth and performance SLAs for enterprise services fifth. Enhanced mobile broadband (eMBB) is already available with 5G NSA, and the category is also likely to represent some fixed wireless access (FWA) services that are not explicitly separated. Most 5G revenue is also likely to be from eMBB, with many operators still to deploy 5G SA and its services—such as network slicing and private 5G. This also explains the lower ranking of “performance SLAs for enterprise customers,” with enterprise likely to be a valuable market for network slicing and private 5G. Operators seem to prioritize full 5G SA capable services, explaining the lower ranking of connected devices, a sizeable area of 4G.
Smaller operators have different opinions than larger operators on the revenue growth question. Mobile operators with less than 9 million subscribers ranked private 5G first, mobile broadband subscriber growth/retention/higher process second, and network slicing third (with connected devices and performance SLA scoring lower). These findings suggest smaller operators may feel they are already exploiting MBB services and see little scope for revenue growth with SA.
AUTOMATION PRIORITIES AND STRATEGIES
Operators have many options to consider as they automate their 5G networks. This section looks to understand their priorities for new automation technologies.
Figure 8 asks respondents about organizational priorities for automating various performance, test, SLA, customer experience, and fault management aspects.
Roughly a third of respondents (36%) believe “network performance management” is the highest priority for automation, followed by nearly one-quarter (24%) who prioritize “automated network and service testing.” The spread of votes indicates several automation strategies under consideration, with a combined three-quarters (75%) of respondents prioritizing automation of network management areas as follows: “network performance management” (36%), “QoS/SLA management” (16%), “customer experience management” (14%) and “fault and event management” (9%). This result may confirm a reliance on network automation for customer retention or to guarantee a superior network performance for future readiness of 5G service types, including low latency, network slicing, and edge applications.
The enthusiasm for “automated network and service testing” validates the growing maturity and understanding of automated software deployment cycles (CI/CD) and their role within service agility. Only 9% of respondents prioritized fault management, suggesting operators are still exploring options. Another explanation may be that most operators have passive probing systems within their networks, already analyzing near-real-time data. Although active monitoring is better equipped to work in disaggregated, cloud native environments on dynamic workloads, operators might believe these areas have lower importance than the others, or they are still exploring their options.
Automation technologies such as orchestration, CI/CD, and containerization will be critical to support network scalability and agility as operators plan their networks. Figure 9 shows orchestration (ranks first) and CI/CD (second) as the most critical automation technologies. The validation of orchestration as the most critical is likely to imply the need for operators to manage the new services across hybrid networks. CI/CD is transformational for software delivery processes and updates across virtual machines (VMs) or cloud native deployments and operations, and it is encouraging to see respondents recognize the need for this technology.
Containerization ranks third, possibly suggesting operators have VM environments to manage or there is more vendor management of early network function containerized deployments. However, US operators have greater enthusiasm for the technology, ranking containerization (first) ahead of CI/CD (second) and orchestration (third). This marked change may suggest the US market is currently more mature and confident with cloud native technology.
Finally, “dynamic resource allocation” ranks fourth and “closed-loop control” much lower at fifth, implying operators are waiting for technology maturity and organizational readiness.
Network API exposure is an area of growing importance, allowing operators to work in innovative ways to develop and enable services. Functions such as the NEF create the opportunity to securely expose network capabilities to internal services and ecosystem partners for application development. These mechanisms also provide opportunities to automate service aspects (e.g., new service provisioning by trusted applications into the 5G core network).
Figure 10 asks operators to assess the importance of network API exposure to automation. A combined 83% indicate it is “extremely important” (31%) or “important” (52%) to their network automation strategy. Importance within the US region is even greater, with a combined 100% of respondents determining it to be “extremely important” (44%) or “important” (56%). This unequivocally demonstrates that network APIs will have a significant role in automation strategies. It is prudent, however, to accept timescales; widespread practical implementations may still take a few years.
Automation is becoming essential across the network in such areas as network planning, service delivery, optimization, fault resolution, etc. Many operators already have some automation technology—for example, technologies such as the self-organizing network (SON) in the RAN domain. However, strategies need to evolve and make further use of the advanced capabilities AI/ML can offer. Figure 11 asks operators, “What is your network automation strategy for using machine learning and artificial intelligence?”
The survey indicates that AI/ML-driven network automation strategy is most important for RAN optimization, with respondents ranking it first ahead of other network processes. The RAN domain has a strong business case for the prioritized adoption of AI/ML due to its scale, complexity, and potential for optimization and efficiency benefits. In addition, new automation technologies, such as the RAN intelligent controller (RIC), have significant AI/ML elements supporting RAN automation. “End-to-end service delivery over multi-generational, multi-vendor networks” and “customized service delivery insights and recommendations” rank reasonably close in second and third place, respectively. These findings are also reinforced later in the survey (Figure 17), with operators confirming the priority of hybrid and service monitoring insights for their assurance solution.
The lowest scoring by a considerable margin is “traditional closed-loop control using existing service assurance tools for discrete network segments.” These lower rankings may imply that operators are less likely to apply AI/ML technology to traditional/proprietary tools and processes, possibly due to the engineering effort required. Also, findings might show that end-to-end service-centric strategies are more valuable to operators. However, the continued trend of fewer closed-loop control votes suggests this is very challenging.
A FOCUS ON RAN AUTOMATION
RAN automation is a priority in 5G networks. By replacing manual tasks with automated operations driven by AI and ML functionality, operators can save costs, reduce errors, and become more agile. Operational success and accuracy directly correlate to the ML models and datasets used. This section explores operators’ views on ML processes and datasets, plus how to determine the effectiveness of RAN automation.
ML is a maturing technology. Accurate and efficient AI and ML-driven automated solutions depend on learning processes. Figure 12 explores the challenges of ML learning processes.
Expert knowledge is the most challenging aspect of applying ML to the RAN. The lead challenge is the availability of “ML application experts” with “RAN domain experts” close behind. “Model generalization (adaption to new unseen data)” follows in third place, confirming the conclusion that as RAN AI/ML technology matures, so will the sophistication of self-correcting and re-training models techniques, with ML application and RAN domain experts likely to oversee processes for some time. “Implementing ML in the RAN” (fourth), “model hypothesis selection” (fifth), and “data normalization” rank lowest despite still being challenging areas, reflecting the growing maturity in this area.
- Drive Test Datasets (often specific in time and location)
- Crowdsource Data (some variability among handset/measurement apps)
- Call Trace Records (CTR) (sometimes containing vendor-specific extensions)
- Call / Billing Records
- Network Equipment Cell-based Counter KPIs
- RAN Session Records
- Synthetic Data Records (regularly built from mathematical models)
Figure 13 asks operators to select their primary data source for RAN planning, optimization, and assurance processes. A lead group of respondents, representing about a third of respondents (36%), confirm they use “network equipment cell-based counter KPIs” as their primary data source for RAN planning, optimization, and assurance processes. This is an expected result given the selection of network equipment performance indicators available and the history of use through many years and mobile generations. A spread of close groupings for “drive test logs” (18%), “crowdsource data” (15%), and “CTR” (14%) follow, illustrating how operators have several data sources in use. Finally, “RAN session records” (9%), “call/billing records” (6%), and “synthetic data” (1%) were the least popular with respondents.
US respondents rated drive testing much lower, only achieving 10%, which is surprising given its widespread use. Overall, results suggest granular-level key performance indicators (KPIs) are the most valuable information, a view also confirmed by 60% of larger mobile operators with more than 50 million subscribers.
Figure 14 asks respondents to select the top two most valuable success measurements for determining the effectiveness of RAN automation. Operators recognize the need to measure the success of RAN automation, possibly to support future business cases or to justify RAN automation spending. Survey respondents selected 1.7 responses each, which is high given that they could have selected the “all of the above” option instead.
The survey confirms that “using KPIs to measure network performance” (51%) is the most valuable measure of success. “Quality of service metrics” and “reducing manual intervention for network problems by continuously adding closed-loop” follow, both with 36%, before “using availability, latency and throughput to measure SLA adherence” (28%) and then “all of the above” (22%).
Methods clearly demonstrating effectiveness via direct measurement or labor savings are preferred over aspects such as “using availability, latency, and throughput to measure SLA adherence,” which might be less directly correlatable to RAN automation alone.
ASSURANCE AND VISIBILITY
The transition to 5G multi-vendor, disaggregated, and cloud native technology has challenged established monitoring, visibility, and assurance methodologies. Traditional monitoring techniques have often been vendor proprietary and domain-specific in nature (e.g., RAN, core, transport, data centers, etc.). This section gathers insights on current solutions and the challenges foreseen for a holistic 5G SA solution.
Figure 15 asks respondents, “What is the biggest assurance challenge your organization anticipates with 5G?”. “End-to-end visibility of the network/service” leads, according to 39% of operators. This reflects the complexity and the difficulties operators face in gaining insights with solutions now depending on the sophisticated correlation of information across multiple layers: cloud infrastructure layer, orchestration/containerized environments, and 5G network domains.
A fifth (21%) agree that “managing/testing a multi-vendor network environment” will be challenging, possibly due to the growing operational understanding of these new technologies and tools for providing network security, traffic management, observability, etc. The increased cadence of software updates expected for 5G cloud native technology operation is also a marked change from previous generations. “Gaining visibility into the public cloud” (15%) and “identifying issues in the dynamic, edge environment” (14%) represent almost a third of the votes between them, perhaps indicating the lack of current operator availability and immaturity. Finally, “avoiding SLA violations tied to network slicing, private 5G, etc.” (8%) rates lowest in this question, possibly under-represented due to the higher standings of network/service visibility and test environments and again due to minimal current availability by operators.
Figure 16 looks to quantify the number of different vendor solutions which service providers have in use to gain full visibility of services and the entire network.
93% of operators confirm they use more than one solution. The majority of operators (55%) use “2–5” tools, with 31% using “5–10” and 7% “more than 10” tools. Larger organizations have invested more across multiple vendor solutions, and almost a third (31%) of organizations earning more than $5bn in revenue had “5–10” tools and 14% “more than 10.”
Maintaining a large number of tools across multiple domains, vendors, and technologies is inefficient. While future networks may still utilize more than one tool, many operators will undoubtedly look to consolidate as they move toward new network visibility solutions. However, operators will need solutions able to deliver real-time, reliable, and consistent visibility across multiple network layers.
ASSURANCE STRATEGIES AND PRIORITIES
5G SA architecture and technology changes require different assurance priorities than previous generations requiring more dynamic and layered monitoring systems to support increased levels of automation. Many operators will also have hybrid or brownfield network environments, adding additional requirements and complexity, which this section explores.
Figure 17 examines the top priorities for an organization selecting a 5G assurance solution. A wide range of options for 5G assurance solutions are under consideration, and on average, operators chose 2.9 responses each, indicating multiple requirements and complexity−and also that the market is still maturing. Support for hybrid 4G/5G (55%), cloud native (50%), service SLA/KPI monitoring (43%) slimly lead. Hybrid 5G/4G support within an assurance system is unsurprising given the number of mobile operators still utilizing previous mobile network generations.
“Automation and CI/CD approaches” (34%), as well as “consolidation of assurance systems” (33%) and “AI/ML” (28%), are roughly equal. “Public cloud support” (24%) indicates current immaturity and deployment strategies within this space. The lowest scoring, active test support (20%), also similar to Figure 8, is unexpected given the real-time nature of 5G and its emphasis on ultra-reliable, low latency services. One speculation may be unfamiliarity with the term “active testing” and what it entails. However, 5G (and many of the other choices in this question) are likely to necessitate a more proactive approach to assuring network performance.
It is interesting to note that for organizations with more than $1bn revenue, “automation and CI/CD approaches” (44%) are much more critical, placing third behind “cloud native support” (53%) and “hybrid 5G/4G” (49%), possibly conveying the greater scales and complexity within larger organizations.
Figure 18 illustrates the responses cast against strategic options for assembling a 5G end-to-end assurance and analytics solution. The survey shows a good spread of views, confirming the widespread choices and features available for end-to-end assurance and analytic solutions. 5G network encompasses some very complex and diverse technology domains. This may explain why “single vendor solution” (21%) was the least popular choice, with operators believing a single vendor solution may not give the breadth across all service and network assurance and analytics. The opinion is further confirmed as respondents chose “best-of-breed solution formed from multi-vendor components” (50%) as the top choice. “Public cloud pre-integrated” (44%) solution is second, confirming that the appetite for cloud native support suggests operators are looking to solve some of the biggest assurance challenges around visibility into the public cloud, addressed in Figure 15.
As expected in a multi-vendor environment, open source scores highly with “a solution incorporating open source software” (41%). The survey also shows interest across “per domain solutions” (38%), indicating operators consider a wide variety of decisions that are under review to assemble their ultimate solutions. “Under evaluation” (31%) follows the other options relatively closely, indicating the decision process is still underway.
Disaggregation of the RAN has encouraged vendor diversity and introduced new service creation models. RAN innovation within the RIC (for example) relies on multiple ecosystem partners to co-develop and test new solutions for traffic steering, resource, energy optimization, and so on. These new forms of collaboration will drive changes to working practices, testing, and validation.
Figure 19 examines how operators will work with equipment vendors and software providers to test new RAN network designs, infrastructure, devices, and apps.
PoC testing will remain the dominant working practice for validating new RAN designs, infrastructure, devices, and apps. Respondents confirm traditional testing practices such as “PoC testing” (69%) and field trials (47%) to be their preferred working practices. Voting indicates that “open innovation platforms” (33%), collaborative environments (21%), and joint development (20%) are viewed as maturing and is a possible reflection of the low number of these environments currently in use across the ecosystem. Innovation platforms, joint development, and shared environments are gaining traction, with organizations such as the O-RAN Alliance and the Telecom Infra Project leading this approach to collaboration and testing.
ANALYTICS AND TROUBLESHOOTING
This section examines service providers’ views on analytics and troubleshooting along with the technologies and network functions supporting it.
Figure 20 asks operators where they expect their operational teams to use AI/ML technology most. Unsurprisingly, operators believe “5G SA RAN optimization” (28%) will utilize AI/ML most heavily. The responses reflect the optimism on AI/ML use within RAN optimization shown in Figure 11.
“Anomaly detection, e.g., DDoS, within the mobile core” (26%) follows, confirming the importance of AI/ML around pattern identification tasks involving large datasets and the time savings achievable over manual/human methods.
The other choices are close: “IMS core troubleshooting” (16%), “network capacity planning” (12%), “network load prediction models” (10%), with “SA packet core troubleshooting” (7%) last. IMS troubleshooting scores were slightly higher than expected, possibly due to well-documented VoLTE (voice over LTE) and VoNR (voice over New Radio) issues across the industry. Load prediction and capacity planning form part of the wider RAN optimization, which was the highest, so these areas’ importance is still significant. Although some of the main AI/ML use cases seem to be determined, operators appear to be deciding on other areas, which may also explain the lower ranking for SA packet core troubleshooting.
The NWDAF (again, network analytics data function), defined originally within 3GPP Release 15, has had relatively low adoption amongst service providers. The NWDAF architecture has considerably changed and matured with Release 17 proposals incorporating further analytics, data collection, and ML training functions.
Figure 21 asks respondents if the NWDAF Release 17 is important to 5G SA and also asks them to rate certain features. The majority of respondents (73%) believe aspects of NWDAF Release 17 are important to their organizations. Respondents chose (on average) 1.9 answers, indicating that most find the NWDAF features within the question useful as the last two options: “conceptually important” and “not important,” are mutually exclusive.
Respondents confirmed the most important features with “predictive network analysis” (47%) and “network and traffic security protection” (47%) chosen first. Although the NWDAF does not implicitly have a security function, its ability to act as both a consumer and producer of network services allows real-time optimization and security risk mitigation. “Slice-level assurance” (34%) and “closed-loop automation” (31%) features came in lower. With many network operators yet to deploy 5G SA and some expecting to offer static slicing options before full automation, this question acknowledges the current immaturity.
ASSURANCE AND SECURITY
The mobile network landscape has changed, with increased certain threats and vulnerabilities as new 5G architectures introduce disaggregated functions with increased software update cadences, new interfaces and protocols, and exponentially more devices. These changes will present unique challenges for monitoring and securing networks across the RAN, edge, transport, and core. Figure 22 addresses and ranks the best approaches and security practices to ensure secure and resilient RAN operations.
“Monitoring of RAN anomalies characterizing potential security breaches” is ranked as the most important approach to mitigate risks and vulnerabilities in RAN operation. 5G security encryption, ciphering, and integrity have all seen major changes to improve security weaknesses from previous generations and vulnerabilities (e.g., 4G IMSI transmission in the clear). However, multi-generational device use such as IoT services may suggest the prominence of this security threat and the need to identify and act quickly to potential security breaches in networks supporting more stringent 5G reliability and SLAs.
“Following industry standards/guidelines, software/firmware updates, secure coding practices, and protocols/algorithms for security and resilience” is the next most preferred approach, ranking ahead of “regular risk assessments, vulnerability scans and implementing controls.” “Conducting regular security testing to identify security weaknesses in RAN solutions” and “staff training to identify security weaknesses in RAN solutions” is ranked the lowest.
Human error and social engineering are very often the cause of security breaches. So the lower rankings for security testing and staff training are surprising, suggesting that operators may be over-confident in their current security practices or in the belief that monitoring tools are better equipped to quickly identify some of the well-publicized threat vectors, such as viruses and worms, botnets, advanced persistent threat attacks, and so on. This is one area in which respondents may have misjudged risk. Staff training on security practices should continue to be a high priority.
Nevertheless, as RAN security continues to mature, the ongoing complexity of implementing consistent security policies, frameworks, and guidelines in hybrid multicloud RAN networks will continue.
Analytics and assurance solutions are important to security detection and DDoS attack prevention in 5G SA. Figure 23 investigates the importance of detecting and preventing security violations.
5G SA security detection and DDoS attack prevention via analytics and assurance solutions are of great importance to operators. “Malware compromised UEs” and “outbound traffic anomalies (e.g., DDoS attacks going through the 5G wireless network)” are considered the greatest areas of importance, with a combined 80% of respondents in each case selecting either “extremely important” or “important.” These results confirm the need to continually adapt and secure the end-to-end mobile network as threat landscapes evolve. Detecting “malware compromised UEs” leads the “extremely important” voting with respondents, possibly highlighting the need to identify individually compromised UEs rather than viewing the data volumetrically.
“Outbound traffic anomalies, e.g., DDoS attacks going through the 5G wireless network,” are also important. “Flow and state exhaustion attacks” and “masqueraded traffic” was also of great concern for service providers, with respondents rating these second highest identically as being “extremely important” (33%) and “important” (43%). Control and user plane separation within the 5G architecture limits previous generational threats from being able to disable both planes at once. However, early detection of these threats is still vital, and the results confirm the priority for operators to continually adapt to secure their end-to-end mobile networks as threat landscapes evolve. In addition, some operators are now planning for fallback or disaster recovery scenarios to stand up a 5G network hosted in the public cloud in the event of severe security compromises.
“User plane traffic engineering analytics” follows in third place with 30% “extremely important” and 41% “important,” being an area already supported with some tools in previous mobile generations. “Top talker analytics” and “spoofed IP addresses on UEs” are rated the lowest, possibly due to information being available from other network sources and these being well-established concepts.
BACKGROUND TO THIS STUDY
Heavy Reading’s 2023 5G Network Analytics and Automation Operator Survey was conducted in May 2023, and this analysis was written in June 2023. The online survey generated up to 86 responses from individuals working at communications service providers after non-qualified responses were deleted from the survey.
2020 Open RAN Operator Survey: Measuring Progress and Looking Ahead to Open RAN at Scale
Principal Analyst, Heavy Reading
