The Evolution and Future of SON in 5G NR
SON (Self-Organizing Networks) has been there since 3GPP release 8. Will it see the light of day in 5G networks, let’s explore!
Introduction
The concept of Self-Organizing Networks (SON) was introduced by Next Generation Mobile Networks (NGMN) and it became standard in 3GPP starting from Release 8. While SON primarily encompassed self-configuration, self-optimization, and self-healing, the advent of 5G NR saw its evolution starting in Release 16. The evolution of SON in 5G NR necessitates adaptation to the new architectures and protocols. Notably, machine learning technologies have been extensively integrated into SON solutions for effective network management.
As networks progress, traditional SON requires augmentation with advanced algorithms making it more dynamic and extending its coverage from the access network to core, and transport. Many SON functions are applicable in Network Slicing as well and RAN Intelligent Controller (RIC) shows significant overlap with traditional SON functionalities. Moreover, energy efficiency stands as a critical concern for Network Equipment Providers (NEPs) aiming to optimize costs pertaining to Capital Expenditure (CAPEX) and Operating Expenditure (OPEX).
SON Overview — 4G
Initially, the primary aim of SON was to automate the Operation and Maintenance (O&M) tasks within cellular networks, presenting an opportunity for operators to enhance network efficiency and performance.
The architecture of SON offers options such as centralized, distributed, or a hybrid solution.
It is widely acknowledged as a pivotal technology driving improvements in spectral efficiency, streamlining management, and reducing operational costs within the next generation Radio Access Network (RAN).
The primary objectives of SON can be broadly categorized into three key points:
1. To bring intelligence and autonomous adaptability into cellular networks
2. To reduce capital and operation expenditure
3. To enhance network performance in terms of network capacity, coverage, offered service/experience, etc.
SON solutions for RAN can be divided into three categories: Self-Configuration, Self-Optimization and Self-Healing.
1. Self-Configuration
Self-configuration is a process of bringing a new network element (eNB) into service with minimal human operator intervention.
a. PnP (Plug and Play): dynamic plug-and-play configuration of newly deployed eNBs. The eNB will by itself configure the Physical Cell Identity (PCI), transmission frequency and power, leading to faster cell planning and rollout.
· Basic setup: Backhaul configuration and interface setup including S1, X2
· autoPCI: Auto physical cell identifier assigns cell IDs, avoiding issues of ID conflicts and confusion
· Automatic inventory and software update
b. ANR (Automatic Neighbor Relation) configures the neighboring list in newly deployed eNBs and defines mobility paths for handover between cells
2. Self-Optimization
It includes optimization of coverage, capacity, handover and interference.
a. MLB (Mobility Load Balancing). The MLB is the SON function in charge of managing cell’s congestion through load transfer to adjacent cells which have spare resources.
b. MRO (Mobility Robustness Optimization). The MRO is designed to guarantee proper mobility to maximize handover success rates, i.e. proper handover in connected mode and cell re-selection in idle mode.
c. RACH. RACH optimization aims at optimizing the random access channels in the cells based on UE feedback and knowledge of its neighboring eNBs RACH configuration to minimize access attempts.
d. (e)ICIC. (enhanced Inter-Cell Interference Coordination). ICIC aims to minimize interference among cells using the same spectrum, manages resource reuse across a cluster of cells, improving signal quality, which assists cell range extension.
e. CCO (Coverage and Capacity Optimization), CCO is a SON function that aims to design self-optimizing algorithms that achieve optima trade-offs between coverage and capacity, aims to enhance overall signal strength and quality
f. ES (Energy Saving) adjusts power of cells according to traffic demand, switching off when not needed. Aims at providing the quality of experience to end users with minimal impact on the environment. The objective is to optimize the energy consumption, by designing Network Elements (NE) with lower power consumption and temporarily shutting down unused capacity or nodes when not needed.
3. Self-Healing
a. Self-recovery of NE software: If the NE software failed due to load earlier software version and/or configuration, the most important thing to ensure that the NE runs normally by removing the faulty software, and restoring the configuration.
b. Self healing of board faults: This use case aims to solve hardware failures in the NE
c. Cell Outage management this use case is split into several functions:
d. COD (Cell Outage Detection). The main objective here is to detect a cell outage through the monitoring performance indicators, which are compared against the thresholds and profiles
e. COC (Cell Outage Compensation). This use case aims at alleviating the outage caused by the loss of cell from service, adjusts neighbor power, tilt, etc. to mitigate the localized outages.
f. COR (Cell Outage Recovery). Automates recovery from detected cell outages
g. CDD (Cell Degradation Detection). Identifies symptoms of poor performance and flags issues
High Level examples of SON Functions
Machine Learning Enabled SON and NM (4G)
The classes of problems that need to be addressed when managing the network autonomously are:
· Variable estimation or classification: the task aims at e.g. estimating the QoS or the QoE of the network, at predicting performances or behaviors of the network, by learning from the analysis of data obtained from past behaviors of the network
· Diagnosis of network fault or misbehaviors: aim at detecting issues ongoing in the network, which may be associated with faults and anomalous setting of network parameters
· Dimensionality reduction: the network continuously generates a huge amount of data. It is convenient to eliminate the noise present in the data base and extract useful information, by reducing the dimensionality of data
· Pattern identification, grouping: aim at identifying patterns, group of nodes with similar characteristics, according to some kind of criteria. An objective may be to apply to them similar optimization approaches
· Sequential decision problems for online parameter adjustment: online adjust network parameters, with the objective to meet certain performance targets
Data Generated by Mobile Cellular Networks
A huge amount of data is currently already generated in mobile networks during normal operation by control and management functions. This kind of data can be exploited to find patterns and extract useful information from them,
1. Call/Charging Data Records (CDR): they provide a comprehensive set of statistics at the service, bearer and IMS levels. These records are typically stored for offline processing. The records are generated in correspondence with high-level service events (e.g., start of a call)
2. Performance management functionality: it covers aspects of the performance of the RAN, such as, FCAPS, radio resource control and utilization, performance of the various bearers (both on radio part and in the backhaul), idle and connected mode mobility
3. Minimization of Drive Test (MDT): this data refers to the radio measurements of both idle and connected mode mobility, coverage items, such as, power measurements and radio link failure events, and can be associated with position information of the UE performing the measurement
4. E-UTRA Control Plane protocols and interfaces: such as RRC, S1-AP, X2-AP protocols, concerning aspects such as cell coverage, user connectivity, mobility in idle and connected mode, intercell interference, resource management, load balancing, among others
5. Data plane traffic flow statistics: which can be gathered at various points of the network, like the eNB, or the PDN gateway (PGW) and Serving gateway (SGW). The internet protocol flow information expert (IPFIX) is an example of standardized format to exchange this kind of statistics.
Overview of ML based Network management
1. MLB Mobility Load Balancing
2. MRO: Mobility Robustness Optimization
3. CCO: Coverage and Capacity Optimization
4. ICIC: Inter-Cell Interference Coordination
5. ES: Energy Savings
6. COC: Cell Outage Compensation
7. COD: Cell Outage Detection
8. SON Conflicts Coordination
9. MDT: Minimization of Drive Test
10. Core Networks
11. SDN
SON 5G Evolution
In 5G, SON is expected to be coherent functionality that integrates all SON functions across network (RAN, CN), network management layers and RATs with efficient coordination between the centralized and distributed components. Such solutions should be adapted to virtualized networks.
The use of AI techniques in the network supervisory system could help solve some of the problems of future network deployment and operations. ETSI has therefore set up a new ‘Industry Specification Group’ on Experiential Networked Intelligence’ (ISG ENI) to develop standards for a network Supervisory assistant system.
ONAP support for Network Optimization and Optimization Framework
· The first one is for the optimal placement of 5G virtual network functions
· The second one is for performance optimization, where SON is a key functionality
The need for SON in 5G
To grow capacity and/or usage experience, 5G has three radio dimensions it could evolve, more available spectrum, higher levels of cell re-use (densification) and improved spectral efficiencies and radio resource management. These are some improvements in modem/modulation design, but these are almost optimum in LTE already.
Moreover, a 5G system, from a holistic perspective, will require better integration of the radio system in synergy with the network backhaul and associated internet content and application servers.
SON Features for 5G NR
Release 15 is the first 3GPP release describing the new 5G NR standard. However, it barely specifies any SON function. Specification of SON for 5G BR are laid in release 16 and later versions.
LTE SON should be evolved to be applicable in 5G NR networks.
· Network monitor mode
· Automatic Neighbor Relation (ANR)
· Inter-Cell Interference Coordination (ICIC)
· Coverage and capacity optimization (CCO)
· Physical Cell Identity (PCI) optimization
· Mobility Robustness Optimization (MRO)
· Transmit Power Management.
1. Network Monitor Mode (Small Cell)
In a small cell, one of the cornerstone capabilities that enables SON is the ability to listen to the radio environment in the vicinity of the deployment and optimize the RAN deployment parameters accordingly. This capability, known as network monitoring or network listening, involves the small cell acting as a passive UE receiver, receiving the RF signal as well as searching and decoding system parameters from any nearby cell received with sufficient signal strength. This capability will continue to be key in 5G small cells as well and will have added detection responsibilities in line with the increased complexity of the 5G signal, such as multiple bandwidth parts.
2. Automatic Neighbor Relation (ANR)
· Measurement Report in Inter-RAT ANR Cell discovery procedure: in addition to the NCGI(s), TAC(s) and PLMN ID(s), 5G UEs will also report the RANAC(s) and the NR frequency band(s) of the related neighbor cell. RANAC is the code of the RAN Notification Area, which is a new NR feature introduced to support the new RRC_INACTIVE mode. This new feature should be considered to support fast RRC re-establishments during mobility events.
· ANR Measurements Configuration. For example, since 5G NR has no cell-specific reference signals, the reporting configuration must specify the type of reference signals that shall be measured (CSI or SS). Furthermore, in addition to RSRP and RSRQ, 5G NR also defines SINR as an L1 measurement. The network can then use this new measurement — for example, to exclude neighbor cells whose signal quality does not satisfy pre-given quality criteria.
3. Inter-Cell Interference Coordination (ICIC)
Similarly to LTE, 5G NR defines an interface between gNBs, called the Xn interface, to exchange interference coordination information. 5G NFV might facilitate in the future the deployment of centralized SON functions, wherein a central SON network function would control the maximum power that gNBs may transmit on each soft frequency reuse (SFR) sub-band. However, defining those ICIC sub-bands will be difficult in 5G due to the different numerologies used in neighbor cells, which can lead to overlapping bandwidth parts across cells that are nearby
4. Coverage and Capacity Optimization (CCO)
The objective of the CCO function is to maximize the cell size while minimizing the interference caused to neighbor cells.
· CRS (Cell-Specific Reference Signal) evolution in 5G: LTE cells achieve SON by self-configuring the power of the cell-specific reference signals, which affects the power with which all physical channels are transmitted (in LTE, the power of each and all PHY channels is defined relative to the power of the reference signals). The lack of CRS in 5G NR makes this approach invalid. In fact, 5G NR may allocate less power to reference signals than LTE used to, thus leaving more power available for data transmission. This new scheme needs to be considered by the power budget.
· High Frequency band in 5G: 5G NR support for frequency bands above 6 GHz which suffer higher free-space propagation attenuation than the bands below 6 GHz. Consequently, unless gNBs operating in these bands support significantly higher transmit powers than their below 6 GHz competitors, 5G cells in these bands will be smaller (and the network denser). This higher network density has implications regarding the number of neighbors gNBs that need to be considered when configuring the transmit powers of the physical channels.
· mmWave Small Cell: the higher number of antennas that massive MIMO introduces in 5G NR increases the power requirements of all RAN nodes. Small cell products have higher power restrictions than Macro cell. It is unlikely that sub-6 GHz small cells use more than four transmit antennas. However, mmWave small cells will not only need to cope with higher path losses, but also with up to 256 transmit antenna elements. This puts mmWave small cells in a position where only very high-density deployments are possible
5. Physical Cell Identity (PCI) optimization:
· PCI ID. LTE has 504 physical layer cell IDs while 5G NR defines a total of 1008 unique PCIs grouped in three sets of 335 PCIs each. The total available PCI doubled, however, NR networks will most likely be denser than LTE networks, which may increase the number of cells which overlap, so the collision reduction factor is unknown. Larger PCI set is essential to support the higher densities for which NR is designed, especially in deployments above 6 GHz
· CRS (Cell-specific reference signal) obsoleted in NR. The SON autoPCI function in LTE networks has to avoid selecting PCIs sharing the same PCI mod 3 as those from neighboring cells. This is to prevent LTe’s reference signals from neighboring cells interfering with each other in frame-synchronized networks (e.g. TDD). Since 5G NR has no cell-specific reference signals, the requirement to avoid PCI mod 3 collisions no longer applies in 5G. the autoPCI function has therefore more flexibility when choosing a PCI in frame-synced networks (e.g. TDD)
6. Mobility Robustness Optimization (MRO): the XnAP specification does not include procedures for the purpose of optimizing the mobility parameters (e.g. handover thresholds). Any support for MRO in Rel-15 5G NR deployments must therefore be coordinated centrally through virtual SON functions.
SON and Orchestration in 5G era
From SON to Open MANO
In oder to manage hyperdense and heterogenous cell networks, that traditional SON will need to be enhanced with advanced algorithms which make it more dynamic, and which extend its reach beyond the access network to cover core, backhaul and IT assets. It will eventually leverage Machine Learning and AI techniques to evolve beyond just automating the optimization of the network, and to add a hefty dose of intelligence to those automated decisions.
Network Slicing
Network slicing involves the instantiation and operation of multiple parallel logical networks operated within a single physical network. Network slices can be created for service specific use cases, and each slice can be viewed as a ‘service’, with its own specific requirements in terms of bandwidth, QoS, latency, security, etc. Each network slice with an operator network has a lifecycle that includes preparation, instantiation, activation, monitoring and modification, and de-commissioning.
It has been postulated that many of the SON functions are applicable to network slice as well. The SON concepts of automated configuration, reconfiguration, optimization and healing can be translated into network slice instances (NSIs). Network performance data can be analyzed and used to drive these slice management functionalities.
Some of these challenges include:
· Bandwidth management across slices
· RAN management across slices
RIC
As part of the industry-wide push towards open platforms, the RAN Intelligent Controller (RIC) was conceptualized within O-RAN framework. It is clear that the RIC framework has significant overlap with what has traditionally been considered as SON. With this development, we are seeing a potentially significant movement towards multi-vendor SON, wherein disparate SON functions are provided by different third-party SON vendors. Such a scenario presents significant challenges. It is very likely that this transformation to a true multi-vendor framework will happen gradually. Initially, it is likely that there will be one dominant SON implementation, with some additional third-party applications provided as extensions. Such integration of ‘native SON’ with third-party SON presents its own set of challenges. Gradual movement from native SON/RRM to RIC architecture is likely.
RIC Functionality
· Non-real-time functionality
o QoE Cross-layer guarantee
o Load balancing (MLB)
o Multi-cell massive MIMO BF opt.
o Data collection for AI/ML analytics
o Customized algorithm deployment
· Near-real-time RIC functionalities
o Admission control (RRM)
o Bearer admission and modification (RRM)
o Mobility management (MRO)
o Load balancing (MLB)
o Inter-cell interference coordination (ICIC) (ICIC)
o Multi-DU/cell radio Resource management (RRM)
Automation using Machine Learning and Big Data Analytics
Big data analytics allows us to collect massive amounts of data and apply millions of calculations on this data to extract useful insights from it. In the context of RAN virtualization, where capacity and coverage are decoupled and can scale orthogonal to one another, big data analytics gives us the capability to shape the RAN dynamically based on the patterns observed in deployment.
· Dynamic sectorization in a virtualized RAN
· Value-added services
· Anomaly detection
· Elastic scaling of virtualized RAN
I. Dynamic sectorization in a virtualized RAN
In a virtualized RAN deployment, the sector boundaries are defined by the set of radio points that belong to the sector. By having some radio points belonging to one sector and others belonging to a different sector and so on we have multiple sectors. By changing the association between radio points and the baseband units (RU vx CU/DU in ORAN), we can dynamically vary the sector areas covered by each sector. With big data analytics and machine learning we can create spatio-temporal models for user movement and traffic density and use this model to dynamically adapt the sector boundaries to the changing conditions throughout the dat. This can help in reducing inter-cell interference as well as the number of handovers.
II. Value added services
By identifying the accurate location of the user, and by using big data analytics, we can then build a spatio-temporal database of user density patterns, user movement patterns, and data traffic patterns at a deployment. This data will be very useful to the operator to fine tune their network deployment, as well as to enable value-added services to the venue owners. For example, techniques such as crowd pattern mining can be used to trigger alarms when the crowd movement patterns deviate from the norm by a threshold which can serve as indicator for an emergency at the venue. Also in a retail environment, applications like shop floor movement correlations can be very helpful to understand which shops people frequently visit together in one trip. Similarly, knowing user density patterns will help the shops place their advertisements and offers at the most trafficked locations.
III. Anomaly Detection
Time series analysis of KPIs can be done to detect whenever any KPI deviates from the ‘normal’ range beyond a certain threshold — and then trigger alarms. These thresholds cannot be static as the KPI may have a variation through the day. For example, a certain rate of handovers seen in the system at a certain time of day may be ‘normal’, whereas it may be abnormal at other times. By building time series model for normal behavior of the KPIs using big data analytics and machine learning, we will be able to reduce false alarms as well as reduce the chances if missing alarms.
IV. Elastic Scaling of Virtualized RAN
The software running at baseband unit will be architected such that it can be horizontally scaled up or down as per the requirement. The virtualized RAN will also be NFV-compliant and the NFV orchestration framework will be able to elastically scale up or down the number of entities based on policies: time of day-based policies or process/memory consumption-based policies, for example. Bug data analytics could be employed to build time series models for how the demand varies in time and establish repeated cycles of demand growing and reducing over time. The NFV orchestrator can then consult this constantly learning model rather than a static policy and more efficiently scale up or down a network based on accurate demand prediction. This network orchestration function is covered by ORAN WG6
Conclusion
Release 15 is the first 3GPP release to describe 5G NR standard, however, specification for SON only started from Release 16. With 5G NR evolution, traditional SON function needs to be updated accordingly.
It is essential that the NG SON be developed as an integral part of the next generation wireless network. In 5G, SON has to be integrated across network (RAN and CN), NM layer and RATs with efficient coordination between the centralized and distributed components.
RIC framework has a significant overlap with what has traditionally been considered as SON.
Technologies coming RAN information, virtualized network and AI/ML will be leveraged to solve some of the problems of future network deployment and operation.
The evolution of 5G SON through 3GPP Release 16 and 17 has brought significant advancements, empowering networks with enhanced automation, intelligence, and optimization capabilities. These key features not only improve network efficiency and performance but also pave the way for innovative applications and services across various industries, heralding a new era of connectivity and communication. As 5G networks continue to evolve, SON will remain at the forefront, driving efficiency and enabling the full potential of this transformative technology.
References
· SON Evolution for 5G Mobile Networks, Wiley
· 3GPP TS 37.320
· 3GPP TS 32.500, 32.50x
APPENDIX
Key Features Introduced in 3GPP Release 17
1. Energy-Efficient SON
Energy efficiency has been a key focus in Release 17 SON, employing intelligent algorithms to optimize energy consumption without compromising network performance. This includes sleep mode optimization and adaptive power management.
2. RAN-Assisted Positioning
Release 17 enhances SON capabilities by leveraging RAN (Radio Access Network)-assisted positioning techniques, enabling more accurate and reliable location-based services. This improves overall network performance and enables new applications relying on precise location data.
3. Automated Traffic Offloading
SON in Release 17 introduces intelligent traffic offloading mechanisms, automatically redirecting traffic to less congested cells or alternative networks, ensuring better load balancing and improved user experience.
4. Cognitive Network Management
Cognitive SON capabilities have been bolstered in Release 17, enabling networks to self-learn, reason, and make informed decisions in real-time, thereby improving adaptability and responsiveness to dynamic network conditions.
