O-RAN NGRG RESEARCH REPORT
O-RAN next Generation Research Group (nGRG)
Research Report
O-RAN Native AI Architecture Description
Report ID: RR-2023-02
Contributors:
CSCN
CTC
VTT
Lenovo
CICT
Release date: 2023.12
<RR-2023-02> 1
O-RAN NGRG RESEARCH REPORT
Authors
CSCN Sun Junshuai [email protected]
CTC Wang Qingtian [email protected]
VTT Chen Tao [email protected]
Lenovo Yang Shuigen [email protected]
CICT Sun Wanfei [email protected]
Reviewers
NVIDIA Lopamudra Kundu [email protected]
Nokia Niraj Nanavaty [email protected]
Qualcomm Douglas Knisely [email protected]
Dell George Ericson [email protected]
Disclaimer
The content of this document reflects the view of the authors listed above. It is not
reflecting the view of the of O-RAN ALLANCE as a community. The materials and
information used for this document have been prepared or assembled by the above-
mentioned authors, and are intended for informational purposes only. The above-
mentioned authors shall have no liability for damages of any kind including without
limitation direct, special, indirect, or consequential damages that may result from the
use of this document subject to any liability which is mandatory due to applicable law.
The information in this document is provided ‘as is’, and no guarantee or warranty is
given that the information is fit for any particular purpose.
Copyright
The content of this document is with the above-mentioned authors. Copying or
incorporation into any other work of part or all of the document in any form without
the prior written permission of the authors is prohibited, save that you may:
- Print or download extracts of the document on for your personal use; or
- Copy the document for the purpose of sending to individual third parties for
their information provided that you acknowledge the authors as the source of
the material and that you inform the third party that these conditions apply to
them and that they must comply with them.
<RR-2023-02> 2
O-RAN NGRG RESEARCH REPORT
Executive summary
Native Artificial Intelligence (AI) is the enabler technology for 6G. The RAN
Intelligence Controller (RIC) of O-RAN is the potential approach for native AI. For
novel scenarios in the future, AI can help to improve the quality of user experience
and network efficiency. The report analyzes the requirements of Native AI in 6G and
discusses the principles and features of native AI, followed by a recommended future
framework for native AI.
<RR-2023-02> 3
O-RAN NGRG RESEARCH REPORT
Table of Contents
Authors ..................................................................................................................... 2
Reviewers ................................................................................................................. 2
Disclaimer ................................................................................................................. 2
Copyright .................................................................................................................. 2
Executive summary.................................................................................................. 3
1 Definitions and Abbreviations .......................................................................... 4
1.1 Definitions ..................................................................................................... 4
1.2 Abbreviations ................................................................................................ 5
2 The Architecture of O-RAN: Overview ............................................................. 6
2.1 Overall Architecture of O-RAN ...................................................................... 6
2.2 Non-RT RIC .................................................................................................. 6
2.3 Near-RT RIC ................................................................................................. 7
2.4 E2 Interface ................................................................................................... 7
3 Requirements of Native AI Architecture of O-RAN ......................................... 8
3.1 The Concept of Native AI .............................................................................. 8
3.2 The Requirements of Native AI in 6G ............................................................ 8
3.3 Programmable RAN toward Native AI ........................................................... 9
3.4 Centralized and Distributed AI..................................................................... 10
4 General Principles and Features of Native AI Architecture of O-RAN ........ 11
4.1 General Principles of O-RAN Native AI Architecture ................................... 11
4.2 Technical Features of O-RAN Native AI Architecture.................................. 11
5 Native AI Architecture of O-RAN .................................................................... 12
5.1 Collaboration control ................................................................................... 14
5.2 Centralized AI and DT ................................................................................. 14
5.3 Distributed AI and DT .................................................................................. 15
6 Conclusion ....................................................................................................... 16
References .............................................................................................................. 17
1 Definitions and Abbreviations
1.1 Definitions
E2: E2 is a logical interface connecting the NRT-RIC with an E2 Node as defined in
[1].
<RR-2023-02> 4
O-RAN NGRG RESEARCH REPORT
Near-RT RIC: O-RAN near-real-time RAN Intelligent Controller: a logical function
that enables near-real-time control and optimization of RAN elements and resources
via fine-grained data collection and actions over E2 interface. It may include AI/ML
workflow including model training, inference and updates.
Non-RT RIC: O-RAN non-real-time RAN Intelligent Controller: a logical function
within SMO that enables non-real-time control and optimization of RAN elements
and resources, AI/ML workflow including model training, inference and updates, and
policy-based guidance of applications/features in Near-RT RIC.
O1: Interface between management entity and O-RAN managed elements, for
operation and management, by which FCAPS management, PNF (Physical Network
Function) software management, File management shall be achieved.
O2: Interface between management entities and the O-Cloud for supporting O-RAN
virtual network functions.
A1: Interface between Non-RT-RIC and the Near-RT RIC functions.
1.2 Abbreviations
3GPP 3rd Generation Partnership Project
5GC 5G Core
AI Artificial Intelligence
CN Core Network
ML Machine Learning
NRT-RIC Near Real Time Radio Intelligence Controller
nRT-RIC non Real Time Radio Intelligence Controller
RAN Radio Access Network
RIC Radio Intelligence Controller
RRM Radio Resource Management
SMO Service Management and Orchestration
TN Transport Network
<RR-2023-02> 5
O-RAN NGRG RESEARCH REPORT
2 The Architecture of O-RAN: Overview
This chapter refers to O-RAN specification 0, and describes the standardized
architecture of O-RAN, that serves as an important baseline for the research of the
future RAN.
There are three parts as the basic characteristics of O-RAN that are different from
RAN defined by 3GPP, etc. NRT-RIC, nRT-RIC, and O-Cloud & SMO (Service
Management and Orchestration) are shown in Figure 1.
2.1 Overall Architecture of O-RAN
Figure 1 provides an overview of the O-RAN architecture. The SMO utilizes four key
interfaces to connect to O-RAN network functions and to the O-Cloud. The
interfaces are the A1, O1, O2, and Open-Fronthaul-M-Plane. Figure 1 also illustrates
that the O-RAN network functions can be VNFs (Virtualized Network Functions), i.e.,
VMs (Virtual Machines) or Containers, sitting above the O-Cloud and/or PNFs
(Physical Network Functions) utilizing customized hardware. All O-RAN network
functions are expected to support the O1 interface for management by the SMO.
Figure 1: High-Level Architecture of O-RAN
2.2 Non-RT RIC
Non-Real Time RAN Intelligent Controller (Non-RT RIC or nRT-RIC) is the
functionality internal to the SMO in O-RAN architecture that provides the A1 interface
to the Near-Real Time RIC (NRT-RIC).
The primary achievement of nRT-RIC is to support intelligent RAN optimization by
providing policy-based guidance, ML (Machine Learning) model management and
enrichment information to the NRT-RIC function so that the RAN can be optimized. It
<RR-2023-02> 6
O-RAN NGRG RESEARCH REPORT
can also intelligently configure radio resource management function in non-real-time
interval (i.e., greater than 1 second).
nRT-RIC can use data analytics and AI/ML training/inference to determine the RAN
optimization actions for which it can leverage SMO services such as data collection
and provisioning services of the O-RAN nodes.
2.3 Near-RT RIC
Near-Real Time RAN Intelligent Controller (Near-RT RIC or NRT-RIC) is a logical
function that enables near real-time control and optimization of E2 nodes’ functions
and resources via fine-grained data collection and actions over the E2 interface with
control loops in the order of 10 ms-1s. The NRT-RIC hosts one or more xApps via
E2 interface to collect near real-time information (e.g., on a per UE basis or a per cell
basis) and provide value-added services. The NRT-RIC’s control over the E2 nodes,
and the SMO manages each NRT-RIC via its A1 interface.
The RRM functional allocation between the NRT-RIC and the E2 node is subject to
the capability of the E2 node exposed over the E2 interface by means of the E2
Service Model in order to support the use cases described in reference [2]. The E2
Service Model describes the functions in the E2 node which may be controlled by the
NRT-RIC and the related procedures, thus defining a function-specific RRM split
between the E2 node and the NRT-RIC. For a function exposed in the E2 Service
Model, the nRT-RIC may e.g., monitor, suspend/stop, override or control the
behavior of the E2 node via policies.
In the event of a NRT-RIC failure, the E2 Node will be able to provide services but
there may be an outage of certain value-added services that may only be provided
using the NRT-RIC.
2.4 E2 Interface
E2 is a logical interface connecting the NRT-RIC with the DU, CU-CP, CP-UP, and
eNB RAN components that support the E2 interface.
•An E2 Node is connected to only one NRT-RIC.
•An NRT-RIC can be connected to multiple E2 Nodes.
The protocols over E2 interface are based exclusively on Control Plane protocols.
The E2 functions are grouped into the following categories:
•NRT-RIC Services (REPORT, INSERT, CONTROL and POLICY, as described in
[2]).
• NRT-RIC support functions, which include E2 Interface Management (E2 Setup, E2
Reset, Reporting of General Error Situations) and NRT-RIC Service Update (i.e.,
capability exchange related to the list of E2 Node functions exposed over E2).
<RR-2023-02> 7
O-RAN NGRG RESEARCH REPORT
3 Requirements of Native AI Architecture of O-RAN
3.1 The Concept of Native AI
Native AI refers to the concept of inherent and built-in AI capability, available or
exposed for network or services. Native AI does not just represent a shift in how AI is
implemented, but also signifies a profound restructuring of the AI mechanism itself.
Different scenarios require different AI mechanisms. If there are loose causal
correlations between different parts of the system, the federated learning [3] model
of artificial intelligence should be adopted. On the other hand, if there are clear
causal relationships between different parts of the system, the split mode of AI can
be utilized. This split AI mode is particularly effective for the intelligent function of
RAN and allows for native AI capabilities within the RAN.
Hence, the concept of native AI involves splitting the entire AI system that serves the
RAN into multiple subsystems or components based on the specific objectives of the
service. Each component is then integrated into the service function of the RAN, to
provide a cohesive system. The split AI approach can utilize a distributed
architecture where different parts of the system handle model training, derive
inference, and carry out data processing, under the centralized control or
orchestration. The distributed AI components can be implemented on both the
network and the UE, depending on the specific requirements for interaction between
the RAN and the UE.
3.2 The Requirements of Native AI in 6G
Native AI is one of the key characteristics under discussion for 6G. In order to
enable native AI in 6G, a number of elements, including computing power, data, AI
algorithms, and functionalities of RAN, TN, CN and Management should be taken
into consideration.
Computing power is the core element of native AI, and 6G applications and services
are expected to increase the demand for robust and accelerated computing
resources to meet the requirements of these applications and services. The
advancement of computing capabilities also plays a vital role in propelling AI
development. Powerful computing resource can significantly enhance the accuracy
and adequacy of AI model training while reducing time costs. In order to
accommodate the evolving scenarios and network functionalities of 6G networks, the
development of novel computing architectures and infrastructures for handling
extensive data volumes and intricate algorithms is required. This can entail
specialized hardware, including graphics processing units (GPUs), field-
programmable gate arrays (FPGAs), and application-specific integrated circuits
(ASICs), NPUs (Neural Processing Units), data processing units (DPUs). It can also
include implementing distributed computing architectures capable of managing the
vast quantities of data generated by 6G networks and applications.
Data is another crucial element of native AI in 6G. AI relies on large volumes of data
to learn and improve its performance over time, and 6G networks will need to
provide a reliable and secure platform for collecting, storing, processing and
<RR-2023-02> 8
O-RAN NGRG RESEARCH REPORT
retrieving this data. This may involve the development of new data storage and
processing technologies, such as edge computing [4] and cloud computing [5]. They
can handle the large volumes of data required as well as generated by AI
applications in real time.
AI Algorithms are also a key component of native AI in 6G. AI algorithms are the
underlying code that allows AI systems to learn from data and make decisions based
on that learning. Development of new algorithms may be necessary to handle the
unique challenges presented by wireless communication networks, such as variable
signal strength, latency, and bandwidth constraints. This may involve the use of
machine learning techniques, such as deep learning and reinforcement learning, that
can adapt to changing network conditions and optimize performance over time.
Finally, the functionalities of 6G RAN are a critical element of native AI. 6G networks
will need to be designed to support the real-time processing and analysis of large
volumes of data, as well as the seamless integration of AI applications with existing
network infrastructure.
3.3 Programmable RAN toward Native AI
In the context of 6G, the integration of native AI needs an efficient and convenient
approach to incorporate AI elements seamlessly. Programmable RAN [6] emerges
as a promising solution to drive the development of native AI. By leveraging
interfaces like E2SM-KPM [1], programmable RAN enables the collection and
interaction of data functions from RAN. This collected data can then be processed
and analyzed to uncover meaningful correlations, thereby enhancing the network's
overall capabilities. Furthermore, programmable RAN facilitates the adoption of
relevant RAN frameworks, enabling the import and update of AI models, ultimately
catering to the diverse requirements of various scenarios.
Programmable RAN encompasses three key components: Parameter, Data, and
Behavior. Programmable parameter facilitates the seamless adaptation of
parameters between AI models and the software-defined RAN functions through a
programmable framework and open interface. Programmable data involves the
construction of data sets for AI model training and the exploration of data
relationships within the RAN functions. Additionally, data can be securely provided to
third parties through relevant secured methods. Programmable behavior allows for
the modification of RAN actions through the utilization of different AI models. To
facilitate this process, a robust framework is required to deploy and manage these
models effectively. The framework should encompass a comprehensive set of
programmable interfaces and function modules, enabling seamless integration and
operation. Furthermore, the programmable behavior aspect empowers the RAN to
dynamically update and replace AI models as needed, ensuring flexibility and
adaptability in response to changing network requirements.
To enable the implementation of programmable RAN, it is essential to progressively
open he traditionally closed protocol stack within the RAN. This involves enhancing
the functionality at the protocol stack level and standardizing and generalizing the
newly opened interfaces. In the context of the ongoing evolution of native AI,
<RR-2023-02> 9
O-RAN NGRG RESEARCH REPORT
programmable RAN catalyzes for advancing toward a more open and intelligent RAN
vision. By embracing programmability, the RAN can effectively adapt to dynamic
network requirements, foster innovation, and leverage the full potential of native AI.
3.4 Centralized and Distributed AI
For 6G networks, it is anticipated that numerous scenarios or applications could be
created anywhere in the network and at any given or arbitrary time. Native AI, as the
critical enabler of 6G, should handle those requirements with quick response.
Therefore, native AI will manifest in two ways in the 6G:, centralized, and distributed
AI.
Centralized AI has a central system that can collect all kinds of required information
and data in devices or clouds. It has abundant and powerful computing resources for
data processing, AI model training and inferring, and so on. However, centralized AI
is unable to process the latency sensitive services or applications requiring quick
response. In that context, distributed AI is a great complement to centralized AI.
Distributed AI [7] can be deployed anywhere on a 6G network, e.g., radio access
network, or core network. Distributed AI is not only efficient with data privacy and
bandwidth, but also is responsive to the application’s actions. Distributed AI has the
important characteristic of being either data-driven or application-driven. This is
because the cached AI models always serve as regional applications. Centralized AI
can help distributed AI to train the complex AI models offline.
Both centralized and distributed AI have their advantages. Centralized AI controls
data and resources, making it suitable for applications where data centralization or
centralized decision-making brings significant benefits. Distributed AI, on the other
hand, leverages the computing power of multiple network elements or devices,
facilitating collaboration while addressing privacy concerns. The choice among
centralized AI, distributed AI and coexistence of both depends on the specific
requirements of the applications, data distribution, and privacy.
<RR-2023-02> 10
O-RAN NGRG RESEARCH REPORT
4 General Principles and Features of Native AI Architecture of
O-RAN
4.1 General Principles of O-RAN Native AI Architecture
The general principles refer to the conditions and objectives that should be
considered in designing O-RAN native AI architecture.
• The O-RAN native AI architecture should be based on the architecture of O-
RAN as the starting point.
• The final native AI architecture may not converge into one and there can be
several potential solutions that describe options to support native AI into O-
RAN.
4.2 Technical Features of O-RAN Native AI Architecture
To achieve the target of native AI in the architecture of O-RAN, the architecture
should have the following technical features：
• Layered Distributed AI: In the future network, the AI function will be flexibly
planned and deployed in the O-RAN logical nodes, rather than being largely
centralized. The O-RAN native AI architecture is also required to support a
more layered distributed AI function.
• Cross-domain AI collaboration: in the 6G timeframe, AI functions are likely to
exist not only in the RAN domain, but also in the network domains like core
network, UE, etc. To support this, the O-RAN native AI architecture needs to
have the ability to carry out the AI task collaboratively among different logical
nodes and across different domains, such as RAN domain and UE domain, or
RAN domain and core network domain.
• Guaranteed AI service: In the 6G timeframe, AI may be a service provided to
customers (i.e., user equipment, network element) in the O-RAN system, i.e.,
AIaaS, and the service quality should be guaranteed. Therefore, the O-RAN
native AI architecture should have the ability to provide AI service and
guarantee the quality of the service.
<RR-2023-02> 11
O-RAN NGRG RESEARCH REPORT
5 Native AI Architecture of O-RAN
In the 4G/5G system, the incorporation of AI is achieved as a potential add-on, which
means AI is deployed with the served network functionalities and is outside of
network elements logically. The evaluation method for AI model training should be
provided from RAN, and the trained AI models or policies would be sent directly to
RAN. There are several unavoidable challenges to this approach.
Firstly, a huge amount of real-time and finer-granularity metrics should be sent out
from RAN to add-on AI entity which may have a large capacity impact on the
transport network and RAN. This brings not only the high cost of RAN, but also adds
to the challenges of interoperability among different vendors.
Secondly, effective inference deeply depends on effective metrics. Otherwise, the
policies or commands inferred by the AI algorithm are not available to RAN.
Therefore, the full potential of AI network cannot be realized.
In response to the above problems, a new native AI architecture should be proposed.
In the native AI architecture, huge amounts of metrics should be avoided to be sent
outside of RAN, and metrics for training and inference of AI should be converged into
the control flow and data flow of RAN. In this research report, a solution using
distributed native AI and digital twin network (DTN) [8] is studied in Figure 2.
In the solution, distributed AI system and distributed DT (digital twin) system are
defined. In this way, the functionalities of AI and DT are converged into
functionalities of CN, TN, RAN and UE respectively. The distributed system of AI and
DT is composed of centralized AI/DT part and distributed AI/DT part. The centralized
AI part configures and controls the distributed AI part, and Distributed AI part reports
measurements to the centralized AI part. Distributed AI part is converged into
serving functionalities of CN, TN, RAN and UE, and accelerates them. The instance
of AI algorithms is decided by AI serving objectives which means different serving
scenarios with different AI algorithms. Distributed DT system is adjoined to
distributed AI system in order to bring offline or online simulation to training or being
inferred of AI. Distributed AI part and distributed DT part locally process the metrics
of RAN so that ZMR (Zero Measurement Report) is achieved. Distributed AI system
and distributed DT system run on the cloud platform (O-Cloud) and leverage
computing power from it.
<RR-2023-02> 12
O-RAN NGRG RESEARCH REPORT
Figure 2: Native AI Architecture of RAN
As shown in Figure 2, there are AI/DT functionalities in CN/TN/RAN and UE
respectively.
Centralized AI and DT run on SMO platform and configure the system end to end
which means centralized AI and DT configure large-scale models or policies.
For configuring UE, the signaling is carried over the Uu interface [9], which may be
considered for 6G specification by 3GPP.
<RR-2023-02> 13
O-RAN NGRG RESEARCH REPORT
5.1 Collaboration control
The proposed 6G native AI network will provide intelligence, data, computing,
management, and other services, e.g., connection services. Enabling this wide range
of services requires addressing a diverse set of requirements, which is difficult to
meet with traditional wireless networks. Especially, collaboration control between
different networks is a new challenge.
Firstly, for differentiated services in different scenarios, the collaboration control
function located in the centralized AI/DT will decompose the intent into service
requirements on evolved network domains, e.g., UE, radio access network, transport
network and core network. The service requirements will affect connection
requirements, AI algorithm requirements, data requirements, and computing
requirements. The decomposed service requirements are sent to the corresponding
collaboration control function in the network domain.
Based on the service requirements from the collaboration control function located in
the centralized AI/DT, the collaboration control function located in the network
domain will decompose the service requirements into the network function
requirements, connection requirements and resource requirements, in order to
establish associations and allocate resources. In addition, the network functions,
connection, and resources are adjusted adaptively according to the service changes.
Secondly, the collaboration control function will be deployed closer to the network
function. Therefore, it will have more real-time management capabilities. For the
services with high real-time requirements and low complexity, it will be processed by
the collaboration control function located in the UE, radio access network, transport
network or core network. For the services with low real-time requirements, larger
areas and high complexity, it will be processed by the collaboration control function
located in the centralized AI/DT (cloud platform).
5.2 Centralized AI and DT
Native AI can be an important characteristic of 6G, and it can drive 6G network
intelligently. The power of AI can be used not only to accelerate the network
functionalities and optimize radio resources, but also to run the entire network
including CN, TN and RAN as an overall intelligent system.
DT brings the infrastructure to run native AI in 6G. The computing and processing of
AI can be run into digital twin domain as online as well as simulation.
The AI and DT systems of the proposed 6G architecture consist of the environment
for OAM (orchestration and management) [2], acceleration of algorithms, and flexible
routing etc. encompassing an end-to-end system including UE side and network side.
The AI/DT system is a distributed system, of which the centralized AI and the
centralized DT are the basic components, and the distributed AI and distributed DT
are complimentary components to serve objects on-demand in a distributed manner.
Centralized AI and centralized DT are an organic whole.
Centralized AI entities run for:
<RR-2023-02> 14
O-RAN NGRG RESEARCH REPORT
a) orchestrating, managing, deploying and controlling all the distributed AI entities.
b) executing functions of centralized AI entities such as training models and
processing data with varying time scales (such as 10ms, 100ms, 1m, etc. or some
arbitrary periods), making policies or decisions etc.
c) supporting other requirements.
Centralized DT entities run for:
a) orchestrating, managing, deploying and controlling all the distributed DT entities.
b) executing functions of centralized DT entities such as supporting centralized AI to
verify the accuracy and validity of policies and commands it produces.
c) producing, managing and running the digital twin entities that centralized AI
served.
Centralized AI and centralized DT are implemented in the network side and running
on a cloud platform.
5.3 Distributed AI and DT
Distributed AI and distributed DT are the complimentary entities of the AI system and
DT system besides centralized AI and centralized DT.
Distributed AI entities run for:
a) receiving and executing commands from centralized AI entities.
b) serving functions of the network.
c) processing measurement reports and metrics locally.
Distributed DT entities run for:
d) receiving and executing commands from centralized DT entities.
e) serving distributed AI entities.
f) providing online simulation for distributed AI entities.
Distributed AI and distributed DT run on CN, TN, BS and UE respectively on
demand.
<RR-2023-02> 15
O-RAN NGRG RESEARCH REPORT
6 Conclusion
The report provides an introduction to a proposed native AI architecture based on
the O-RAN architecture. It begins with an overview of the architecture in section 2,
followed by the presentation of requirements in section 3. In section 4, the principles
and features of the architecture are proposed. Additionally, the report delves into the
details of the proposed native AI architecture in section 5.
<RR-2023-02> 16
O-RAN NGRG RESEARCH REPORT
References
[1] O-RAN E2 General Aspects and Principles (E2GAP) 4.01. Available :
https://orandownloadsweb.azurewebsites.net/specifications
[2] O-RAN Architecture Description 10.0. Available:
https://orandownloadsweb.azurewebsites.net/specifications
[3] Li T, Sahu A K, Talwalkar A, et al. Federated learning: Challenges, methods,
and future directions[J]. IEEE signal processing magazine, 2020, 37(3): 50-60.
[4] Al-Ansi A, Al-Ansi A M, Muthanna A, et al. Survey on intelligence edge
computing in 6G: Characteristics, challenges, potential use cases, and market
drivers[J]. Future Internet, 2021, 13(5): 118.
[5] Tomkos I, Klonidis D, Pikasis E, et al. Toward the 6G network era:
Opportunities and challenges[J]. IT Professional, 2020, 22(1): 34-38.
[6] Bonati L, Polese M, D’Oro S, et al. Open, programmable, and virtualized 5G
networks: State-of-the-art and the road ahead[J]. Computer Networks, 2020, 182:
107516.
[7] Cao L. Decentralized ai: Edge intelligence and smart blockchain, metaverse,
web3, and desci[J]. IEEE Intelligent Systems, 2022, 37(3): 6-19.
[8] Groshev M, Guimarães C, Martín-Pérez J, et al. Toward intelligent cyber-
physical systems: Digital twin meets artificial intelligence[J]. IEEE
Communications Magazine, 2021, 59(8): 14-20.
[9] NR; NR and NG-RAN Overall Description; Stage 2 (Release 17), 3GPP TS
38.300 V17.6.0, 2023.
<RR-2023-02> 17
