February 16, 2024

5G networks are essential to digital transformation (DX) and advanced applications. Nevertheless, with most applications in recent years, productivity drops drastically if communication quality declines for even a moment. In response to this issue, NEC presented a technology that analyzes the state of applications and wireless quality in real time and autonomously optimizes the 5G network. This is believed to ensure reliability and real-time performance required for remote control of robots and automobiles. We spoke with our researchers about the details of this technology.

Autonomous optimization of 5G networks in user units

― What kind of technology is the technology that autonomously optimizes 5G radio access networks?

Iwai: With most advanced applications, even a momentary drop in communications quality can compromise the productivity of the application. The radio access network (RAN) autonomous optimization technology analyzes and predicts the state of applications and wireless quality in real time with AI and can smartly and autonomously optimize the RAN. We developed this technology with an eye to applying it to use cases where real-time performance and reliability must be ensured, such as remote control of robots and automobiles. This technology can improve the productivity of applications by controlling the RAN in user units instead of in conventional base station units, which embodies NEC’s innovativeness.

Additionally, this technology is in compliance with the standard specifications of the O-RAN Alliance, which is promoting open RAN interfaces. With this technology, you can equip AIs with the RAN Intelligent Controller (RIC), which is defined in the standard specifications of the O-RAN Alliance, to enable smart autonomous optimization of versatile RAN functions. You can apply this to both public and private 5G networks, wide and narrow areas, regardless of whether it is outdoor or indoor―a whole variety of settings.


Takahashi: We researched and developed this technology to promote DX and advanced application development. When I was in charge of business development a couple years ago, I did exhaustive research on the latest cases of industrial DX and IoT. I have visited developers of cutting-edge large-scale robot warehouses in Europe and such and discussed with them from morning to night for several days. What I found surprising then was that, while many application developers design applications based on current wireless communications standards, early adopters that develop forward-looking applications design with what they want to realize with the application at the forefront first and then deal with any wireless communications issues. The value to be offered comes first and then they solve the wireless communications issues through trial-and-error to stabilize performance. That, as a result, improves on-site productivity. In the case of warehouses, this leads to increased number of customers and the items that can be handled. This can then further increase sales and raise stock prices. However, the solutions at the time were complicated; an overall customization of the communications infrastructure, including network and devices, was necessary for just that specific application.

While such innovative application development was necessary, needing to customize communications infrastructure for each application did not contribute to scaling up NEC’s business. It was easy to assume that applications will continue to diversify with the progress of DX, so we had the awareness that some new way was necessary.

Around that time, a trend aimed at open RAN, utilization of software, and intelligence was starting to pick up in the market, which gave us the inspiration to smartly control RAN standard functions using AI. We worked on the R&D with the thought that we could make a system that autonomously and efficiently adapts RAN to various applications.

Reducing the frequency of excessive latency to 1/50 or less of conventional occurrences

― How does the technology work specifically?

Onishi: The key point of this technology is to smart-control RAN according to the conditions on site. Industrial sites have intermixed data transmissions of different applications with different communication requirements, such as transmitting large-volume camera video data and massive amounts of small robot control packets. In addition, movements of objects on site may block or interfere with wireless communication. The communications quality problem occurs in a complex entwinement of these transmissions and conditions. Therefore, we developed a system that accurately analyzes the state of applications and changes in wireless quality to appropriately control the RAN.

The most critical indicator in wireless communications at industrial sites is transmission latency. In robot remote control, when transmission latency exceeds the threshold, the system may be stopped for safety purposes. It was necessary to build a reliable system with minimum latency.

Broadly speaking, there are two types of transmission latency: retransmission delay and queuing delay. Retransmission delay occurs when retransmission is necessitated by transmission error due to a sudden degradation in signal strength. Queuing delay is caused by latency time due to numerous users attempting transmission at the same time.

Generally speaking, more radio resources are allocated to eliminate retransmission delays. This, however, results in straining radio resources throughout the entire system, making it prone to queuing delays. To work around such trade-off in transmission latency, we created a RAN autonomous optimization system that can efficiently resolve this problem.


Takahashi: Aside from the retransmission delay and queuing delay trade-off in wireless communications, many other problems occur from the enmeshed relationship among RAN control parameters and inter-user resource conflicts. We pursued a RAN autonomous optimization technology that handles all these issues at once.

Onishi: Specifically, we developed an algorithm that analyzes the state of applications and wireless quality in real time and derives the RAN control parameters values and excessive latency risks as a probability. Then we also developed a machine learning-based RAN control method that determines a decent control policy based on that prediction result.

Controlling the RAN while learning on-site data by leveraging machine learning can effectively solve complicated issues at work sites. Nevertheless, while machine learning generally delivers high performance, it is also characterized by the difficulty in guaranteeing 100% reliability. Since it was a vital requirement for this technology to ensure high reliability, it was necessary to create a system that enhances stability. What we came up with then was a combination of a machine learning-based control engine and a logic-based control engine.


Takahashi: While running machine learning, the system verifies that at the same time. Any time it detects a risk of the accuracy being compromised, it switches to a logic-based engine. For example, supposing the accuracy of machine learning is 100-percent, if the system detects an exceptional condition, it provisionally switches over to the stably 80-percent logic-based algorithm and then switches back to the 100-percent machine learning. This technology successfully ensures the stability of RAN control by switching engines. Simulation results also demonstrated that the frequency of excessive latency in transmission can be reduced to 1/50 or less of that of conventional methods. This means that it can increase the continuous operation time of robots without stopping them by fifty-fold, which is a quite meaningful effect.

On another note, this system determines the control policy upon its understanding of the current state based on changes in wireless quality. The ability to predict changes in wireless quality with high accuracy is a key feature of this system. This part was taken on by Mr. Nishikawa, who has years of experience in the R&D of wireless quality.


Nishikawa: The learning-based wireless quality analysis technology , presented in March 2023, is a technology that analyzes wireless quality, identifying the cause of degraded wireless quality and detecting where the problem is using AI. While the learning-based wireless quality analysis technology focused on wireless quality and communication throughput, the issue that we wanted to address with the new technology was transmission latency and wireless quality for each packet. This was the point of challenge.

To analyze transmission latency and wireless quality for each packet, we need to understand momentary wireless quality at each point in time an individual packet is transferred over the RAN. To determine the next control, we need accurate prediction of the wireless quality in the next moment. RAN is equipped with a function that prevents transmission data errors that can be caused by changes in wireless quality and ensures error-free communications. For this technology, it was important to predict sudden, rapid drops in wireless quality that could not be covered with that function.

Therefore, we used the latest AI technologies to develop a mechanism that precisely predicts steep drops in wireless quality from time-series wireless quality data. This technology very accurately predicts drops in wireless quality in 100-millisecond cycles and passes on that information to the control system. This enables the control system to preemptively set RAN control parameters, thereby making possible strict control that completely eliminates excessive latency even in the case of massive packet flow.


Iwai: This technology is planned to be one of the featured exhibits of the NEC booth at the Mobile World Congress (MWC) Barcelona 2024, the world’s largest connectivity event, held in February.

Bringing forward the future of Beyond 5G

― What are some use cases you have in mind?

Takahashi: Currently, we are anticipating applications to robot control for manufacturing or warehouses, and controlling truck platooning in logistics. These applications are still in simulation phase, but we were able to confirm the effects. Preparation for demonstration verification is underway. As a specific milestone, we are currently working on R&D with the aim to deliver this technology to society in 2025.

What I want to emphasize here is that this technology can realize in advance some use cases that were previously considered impossible without waiting for the completion of the Beyond 5G infrastructure.

― Specifically, what kind of use cases?

Onishi: A use case that involves meticulous motion control of machine tools, especially in an environment where wireless quality can fluctuate significantly, needs a Beyond 5G framework that can eliminate transmission latency on the hundred-microsecond order. However, applications in the runup to such a use case, such as remote control of robots and automobiles as well as collaborative robot control, can perhaps be achieved with this technology.


Nishikawa: This technology can also be used in cases where multiple applications co-exist. While current standard 5G specifications can function well as long as they are customized to communication requirements for a single applications, they do not simultaneously satisfy different communication requirements, such as low latency, high reliability, and high-speed and large-volume transmissions. Without waiting for Beyond 5G to be completed, this technology can be applied to use cases where a diverse range of applications with different communication requirements co-exist. It can facilitate a gradual increase in co-existing applications until Beyond 5G is in full-fledged operation.


Takahashi: I agree. Not only can this technology lift a number of barriers in wireless communication, but it also enables automatic adaptation to a wide variety of applications. This, we believe, will increase freedom in developing advanced applications in the future and accelerate social development. We will continue R&D toward its practical application.

Multiple patents pending (as of January 30, 2024)

Go back to Featured Technologies