One year later, let’s talk about Open RAN again

[[385310]]

This article is reprinted from the WeChat public account "Xianzao Classroom", the author is Xiaozaojun. Please contact the WeChat public account of Xianzao Classroom to reprint this article.

When I attended MWC last week, my biggest feeling was the overwhelming wave of network openness, virtualization, and intelligence.

From access networks to core networks, almost all traditional communication equipment has cloud solutions. "White box", "cloud", "lightweight" and similar words can be seen almost everywhere at the exhibition. It seems that overnight, all companies have become base station equipment vendors and core network equipment vendors, and the entire industry has entered the era of "everyone is a vendor (equipment manufacturer)".

[[385311]]

For traditional communication equipment manufacturers, this is undoubtedly a double whammy. The already fiercely competitive market, with so many new competitors, will further reduce profits and make life even more difficult.

However, for operators, this is a welcome development. The long-awaited network openness and decoupling have finally come to fruition. More and more vendors mean that they can get rid of the "holding" of a few equipment vendors and deploy networks more flexibly. Operators' network comprehensive cost (TCO) is also expected to be further reduced.

When it comes to network openness, we have to mention Open RAN.

The openness and decoupling of RAN (Radio Access Network) have always been the focus of operators. In the eyes of operators, the cloudification of RAN is more significant than the core network.

RAN expenses account for more than 60% of operators' TCO.

More than a year ago, Xiaozao introduced O-RAN to you (link). Since then, I have been closely following the development and changes of Open RAN.

Today, based on the latest information obtained from MWC, I would like to talk to you about this topic from the perspective of technology and architecture.

[[385312]]

At this MWC, a concept was mentioned repeatedly, and this concept is closely related to the Open RAN architecture, that is - RIC.

RIC is the key to the Open RAN architecture. If you understand RIC, you will understand the Open RAN architecture.

As early as when the O-RAN Alliance was founded in 2018, in order to develop open RAN specifications and standards, nine working groups (WGs) were set up to study the corresponding interfaces and technologies.

O-RAN Working Group

WG2 and WG3 are responsible for non-real-time RIC and near real-time RIC respectively.

What exactly is RIC? RAN Intelligent Controller, which is the radio access network intelligent controller.

Before we continue to introduce it, let’s take a look at the overall architecture of O-RAN.

The above figure shows the main architectural changes of 5G O-RAN compared to 4G. As can be seen from the figure, the main components of 4G LTE RAN, BBU and RRH, have become O-CU, O-DU, and O-RU in 5G O-RAN.

O-CU: Responsible for the Packet Data Convergence Protocol (PDCP) layer of the protocol.

O-DU: Responsible for all baseband processing, scheduling, radio link control (RLC), medium access control (MAC) and the upper part of the physical layer (PHY).

O-RU: The component responsible for the underlying physical layer processing, including the analog components of the radio transmitter and receiver.

O-RAN uses interoperable hardware with open protocols, replacing traditional closed interfaces and proprietary hardware and protocols, making the RAN architecture more flexible, open and decoupled.

What we usually call RAN virtualization actually mainly refers to the virtualization of O-CU and O-DU. In other words, they can be built on the x86 server platform. O-RU is the radio frequency transceiver, which is currently software radio and white box radio, and there is no way to virtualize it.

Let’s take a closer look at the O-RAN architecture, as shown in the following figure:

(Image from O-RAN Alliance)

This diagram is a bit complicated because it lists the 3GPP standard interfaces (X2, Xn, NG, E1, F1, etc.), as well as the RIC we just mentioned and the corresponding new interfaces.

I also found a picture comparing the architecture of O-RAN and 3GPP RAN, which makes it even clearer:

(Image from Ericsson)

Obviously, in Service Management and Orchestration (SMO, similar to MANO in NFV), there is a Non-Real-Time RIC. In CU, there is an additional Near-Real-Time RIC.

Non-real-time RIC is a function, not physical hardware. It is responsible for configuration management, device management, fault management, performance management, and lifecycle management of all network elements in RAN. Non-real-time RIC is responsible for processing services with latency requirements greater than 1 second, such as data analysis and AI model training.

Near real-time RIC, also known as near real-time RIC, is responsible for processing services with a latency requirement of less than 1 second (50ms-200ms), such as wireless resource management, switching decision, dual connection control, load balancing, etc.

The non-real-time RIC collects global relevant data from the RAN and application servers, performs data analysis and AI training, and sends the reasoning and strategies through the A1 interface and deploys them in the near-real-time RIC.

Near real-time RIC is responsible for collecting and analyzing real-time information from the RAN, combining it with additional or global information provided by the non-real-time RIC, and using the inference models and policies issued by the non-real-time RIC to monitor and predict changes in network and user behavior in real time, and adjust RAN parameters in real time according to policies (such as QoE targets), including adjusting resource allocation, priority, switching, etc.

Near real-time RIC includes many xAPPs. As the name implies, xAPP is an APP (application) independently deployed by a third party, which deploys AI reasoning models and policies in it, and different xAPPs are associated with different RAN functions, making the functional components of RAN flexible, programmable and scalable.

At MWC, Baicaibang, Intel, and China Mobile jointly demonstrated the "5G+AI" application scenario case based on RIC, as shown in the following figure:

In this case, the non-real-time RIC integrated with artificial intelligence pushes the algorithm to the near-real-time RIC platform through learning and reasoning. The near-real-time RIC controls the functional components of the RAN through the E2 interface, thereby accurately and reasonably scheduling and controlling the RAN. To be precise, the control target is the handover (HO) threshold, so that the UE (user terminal) can perform more reasonable handover, significantly reduce the call drop rate, and improve the user's network experience.

This case fully illustrates what the combination of 5G and AI is.

It is conceivable that RIC is not only the key to the smooth realization of comprehensive decoupling and openness of the Open RAN architecture, but also the key to AI-enabled 5G access networks.

As one of the earliest traditional equipment vendors to join O-RAN, Nokia also demonstrated its use of RIC at MWC:

As can be seen from the figure, RIC can not only be used to improve users' network experience, but also monitor network operation anomalies and even help with intelligent energy saving.

Open disaggregation is the primary goal of O-RAN and Open RAN, but not the only goal.

With the continuous construction of 5G, the operator's network has become unprecedentedly complex and large. Pure manual operation and maintenance is definitely a dead end. The only way out for operators is to seek help from AI.

Therefore, both 3GPP and O-RAN will give top priority to the combination of artificial intelligence and networks, and seriously study how to use AI to empower network construction and operation and maintenance. The "autonomous driving network" frequently mentioned in recent years (autonomous driving here has nothing to do with the Internet of Vehicles, but refers to the network "driving (managing)" itself) is actually just like this, embedding AI into traditional communication networks, managing resource allocation, identifying changes in the external environment, establishing algorithm models, generating policy results, and automatically adjusting parameters, thereby reducing manual intervention and cutting costs.

After all the hype, we still have to return to reality. What is the reality? The proportion of Open RAN architecture in the current network is far less than 10%. In other words, don't be too happy too soon.

Open RAN itself is not perfect. Ericsson once pointed out that Open RAN introduced the RIC controller and added interfaces such as A1, E2, O1, and O2, making the architecture more complex and potentially increasing security risks.

Huawei does not join the O-RAN Alliance because it believes that the energy consumption performance of the Open RAN architecture is not ideal.

Rakuten Mobile, a Japanese company that was previously the most successful in Open RAN, has also been involved in a lot of negative news recently.

Let’s wait and see where Open RAN is headed!