By Chu Peisi, nicknamed Zishuo at Alibaba.
The age of 5G is right at our doorstep now. With all of the news circulating about 5G, I’m sure you’ve got many questions in your mind: What exactly is 5G? Is it just about faster Internet speeds? How can 5G reduce average latency to milliseconds? What does network slicing mean? What exactly is going on when it comes to arguing for 5G standards and what exactly are people arguing for?
In this article, we will try to answer all of these questions, providing you with an in-depth look at 5G.
I am sure that Moore’s Law sounds familiar to you all, but Shannon’s Theorem in the field of communication is less well-known. I still remember the strict derivation of this formula I learned at a postgraduate course in Information Theory.
- When data signals are transmitted on a channel with random thermal noise, the relationship between the channel capacity Rmax and the channel bandwidth W and the signal-to-noise ratio S/N is Rmax = W*log2(1+S/N), where log2 is a base-2 logarithm.
- In simple terms, if you want to increase channel capacity, you can increase the bandwidth or the signal-to-noise ratio (SNR). Increasing bandwidth is easy to understand. However, spectrum resources are limited and cannot be allocated without restraint. Even if they are inexhaustible, there is still another key limiting factor, SNR. There are many ways to increase the SNR. For example, we could increase transmission power. However, several countries, including China, have strict restrictions on the transmission power of base stations, so we cannot increase power as we might want to. Even if we were allowed to do so, devices might not be able to meet the ultra high requirements, because high frequency amplification is not a simple matter. Other ways to increase the SNR include improving source coding and channel coding.
Here is a mind map regarding some technologies related to 5G. To provide a brief summary, 5G involves several key technologies, and each technology is designed to address specific demands in several different real world and business scenarios.
Standardization of 5G
Peak rate: 20 Gbps
There is no stipulation for bandwidth, so carrier aggregation handling up to 32 carriers can be used to meet the peak rate requirement. This is actually the peak rate of base stations, rather than of individual users, and it is shared by users within the coverage area of a base station.
- User-experienced data rate (urban area): 100 Mbps
The standard also describes the requirements for the user-experienced rate in different segments. For example, the rate requirements for “broadband access in a crowd” (such as concerts) in 5G standards are as follows: When overall user density is 500,000/km2 and the activity factor is 30%, the target downlink data rate is 25 Mbps, the target uplink data rate is 50 Mbps, the target downlink area capacity is 3.75 Tbps/km2, and the target uplink area capacity is 7.5 Tbps/km2.
- Spectral efficiency: 3 to 5 times higher than IMT-A
- The International Mobile Telecommunications Advanced (IMT-A) is a standard specification for 4G mobile communications formulated by the International Telecommunication Union (ITU). The vision of 4G was to achieve a rate of 100 Mbps at 20 MHz, so the spectral efficiency of 4G is generally considered to be 5 bps/Hz.
- According to the KPI, the improvement of spectral efficiency in 5G is quite obvious, which can more directly guide the improvement of the technology. According to Shannon’s formula, increasing bandwidth can increase speed. However, spectrum is a scarce resource and its supply can be increased to only a limited extent. If the spectral efficiency of 5G can be 3 to 5 times higher than that of 4G, 5G will theoretically reach a rate of 1.5 Gbps to 2.5 Gbps at 100 MHz. On May 9 last year, China Unicom’s first 5G base station in Guizhou went online. According to field test results, the peak downlink rate of single user equipment (UE) in the 5G network reached 1.8 Gbps at 100 MHz.
- Mobility: 500 km/h
The requirement for mobility was not well addressed with 4G, especially in scenarios such as high-speed trains. A speed of 500 km/h may cause obvious Doppler effect, making frame format processing difficult. In addition, the extremely fast speed results in frequent handovers, which also poses some challenges to data link stability.
- Latency: 1 millisecond
This target may not be possible in time division duplex (TDD) systems.
- Connection density: 1,000,000/km2
In the context of the emerging Internet of Things (IoT) industry, this indicator does sound impressive. However, the IoT standard in 5G, massive machine type communications (mMTC), is developing slowly, probably due to the absence of revolutionary products. Narrowband IoT (NB-IoT) has not reached saturation yet and cannot provide a sufficient driving force. Although the significance of IoT is widely recognized, we may still have to wait for some time until this is fully realized.
- Energy efficiency ratio: 100 times higher than IMT-A
Considering the increasing numbers of base stations and higher electricity bills paid by operators, it is time to reduce energy consumption and protect the environment.
- Traffic density: 1 Mbps/m2
This has to rely on greater bandwidth and more new technologies.
3GPP is in charge of the standardization of 5G. You can get more information about the standardization process from its official website. For more details about 5G NR physical layer protocols, you can download the document from this webpage.
The main content of NR physical layer protocol is outlined in the 3GPP TS 38.201 V15.0.0 (2017–12) document. The physical layer involves one overview document, TS 38.201, and six protocol documents: TS 38.202, TS 38.211, TS 38.212, TS 38.213, TS 38.214, and TS38.215.
A Rational View of the 5G Rate Increase
There are different claims about the 5G rate: 20 Gbps, 4.6 Gbps, and 6.5 Gbps. Why is there such a big difference, and is 4.6 Gbps definitely inferior to 6.5 Gbps?
The 5G peak download rate is 4.6 Gbps at 200 MHz below 6 GHz. Here, “below 6 GHz” means that a carrier frequency is below 6 GHz, and “200 MHz” refers to bandwidth. “4.6 Gbps” is a peak rate. According to the KPI of 5G, the spectral efficiency is 3 to 5 times higher than that of 4G. What’s the figure in 4G? The 4G spectral efficiency is 5, which means that a bandwidth of 20 MHz achieves a peak rate of 100 Mbps. Based on the KPI of 5G, we can calculate that the bandwidth of 200 MHz corresponds to a rate of 3 Gbps to 5 Gbps.
The millimeter wave rate is 6.5 Gbps at 800 MHz, which is 10 times the experienced rate of 4G LTE. The “millimeter wave” refers to a frequency band. The international mainstream frequency band is 28 GHz, which is a carrier frequency. “800 MHz” refers to bandwidth. Resources are quite abundant in the high frequency band, and 800 MHz is pretty average. The spectral efficiency is merely 1.625 times of that of 4G. This may be due to the wide bandwidth, which results in a greater OFDM subcarrier bandwidth. When the subcarrier spacing increases, the spectral efficiency drops. Regardless, though, this rate is still very impressive.
It should be noted that each of these rates is a peak rate shared by all users within the coverage of a base station. Therefore, your actual experienced rate may not be as fast as that. The base station side has scheduling algorithms to ensure fair distribution, but there may not be absolute fairness in 5G, where premium enterprise users may receive more resources. In addition, the unit of rate in the field of communication is bits per second (bps) instead of bytes per second (Bps). One byte includes 8 bits.
In summary, based on the KPI of spectral efficiency in 5G and bandwidth, you can calculate the peak rate. The peak rate reflects a total network capacity, and it is the upper limit instead of your experienced rate. Talking about rates without mentioning the corresponding bandwidth is meaningless. It is important to note that barrier frequency and bandwidth are two different things. Peak rate is related to bandwidth and spectral efficiency but not to carrier frequency. To provide an analogy, capacity of a train is determined by the number of carriages instead of its speed.
The Development of VR/AR Technologies
With the continuous expansion of augmented reality (AR) and virtual reality (VR) markets, there is bound to be significant growth in video streaming. Next-generation content formats like 6DoF will place higher requirements on networks, and the upper limit of individual data rates will leap from 200 Mbps to 1 Gbps, which will also put higher demand on bandwidth. Many participants in the AR and VR markets are eager to launch bestsellers and grab a share of the market. As a communication technology, 5G itself is designed to address problems in data transmission. VR and AR are facing many challenges in terms of experience, content source, and resources that are yet to be resolved by the industry. Of course, whoever best resolves the challenges and best coordinates with 5G will win loyal consumers and seize the market opportunity.
Let’s make some calculations on bandwidth. Currently, in China, the bandwidth allocated to China Unicom or China Telecom is 100 MHz. According to 5G requirements, assuming that a rate of 50 Mbps is required to ensure a fluent VR experience, such a bandwidth can provide a rate up to 2500 Mbps, which can support at most 50 concurrent users. Of course, these are theoretical values, and the actual value would be even lower due to signaling-related overheads. This value is still far removed from the eventual goal of allowing a crowd of viewers to watch concert performance in real time and from multiple angles. Certainly, this is just the beginning. More resources of millimeter wave frequency band will be allocated later, and broadcasting can be used to share non-interactive service traffic.
In fact, AR and VR are not exactly one in the same, and the problems faced by them are also different, but we do not strictly differentiate between them here. Anyway, it remains to be seen what the future form of this product is, Google Glass, projector, or a more advanced form of interaction? We are looking forward to seeing what fabulous innovations the future brings.
At a 5G workshop I once attended, we talked about the application of holographic projection in 5G. It is really cool to allow several viewers to see a late singer standing on the stage and singing for them. In fact, this has already been done in a small range of specific IPs. However, the promotion and development of such technologies are still limited by content, which needs to be carefully designed. The content design and creation process is time-consuming.
By contrast, physical devices can simply be rented. From today’s perspective, it is quite challenging to applying stereo scenes built by stage setting in reality on a large scale. Let’s look forward to better interaction methods like those shown in Sci-Fi movies, such as interactions through air or on mirrors. Problems faced by holographic projection are much more about the projection technology and content creation than transmission. In other words, the holographic projection can rely on optical fiber if 5G is not available. If optical fiber is finally found to be the root of problems in user experience, we can use 5G directly. However, what needs to be solved today is more the scenario itself.
Of course, the upgrade of transmission links will bring many benefits. For example, if large bandwidth and low latency transmission is implemented, various computation and rendering can be done in the cloud, and the machines in the cloud can be expanded quickly and conveniently, even at any cost. Cloud rendering provides a good direction for studies on how to create perfect user experience. In addition, cloud rendering can combine multiple isolated scenarios to form a more interesting virtual world, just like those in online games. A good VR experience not only requires bandwidth, but also is sensitive to latency. Therefore, network slicing for VR may be a compromise between large bandwidth and low latency.
The Application for Network Slicing
Network slicing is a very important technology in 5G and is quite popular with operators. Data packets can be classified by this technology, so that service levels can be established to implement differentiated services. More generally speaking, those subscribing to premium services will be assured of bandwidth and latency, while others just obtain regular services. The 4G standard also defined differentiated QoS, but only for access networks, that is, differentiated services were provided for the segment between mobile phones and base stations. Differentiated services at that time were also designed based on demand. When allocating resources to Internet service users and voice service users, base stations give priority to voice service users. With 4G, operators use Quality of Service Class Identifier (QCI) for coarse-grained service classification. The 3GPP defines a total of 9 levels: 4 Guaranteed Bit Rate (GBR) levels and 5 Non-GBR levels.
Network slicing involves two parts: network slicing in core networks and network slicing in access networks. The network slicing in core networks is closely related to virtualization technology. Network functions virtualization (NFV) and software-defined networking (SDN) have received extensive attention and research as the main technologies supporting the implementation of network slicing in core networks. The implementation of network slicing in access networks is more challenging. It has to cope with different business models and different requirements for business indicators. Network slicing and access networks need to ensure low latency, high connection density, and high reliability, as well as isolation between network slices.
Alibaba Group is exploring applications related to network slicing. In fact, there was no complete network slicing scheme with 4G. As defined in 3GPP protocols, different QCIs correspond to different levels of packet delay and packet error rates. Operators provide differentiated services for users and businesses based on QCI. According to 3GPP specifications, a typical scenario for a user with QCI = 3 is Real Time Gaming, while QCI for data services of normal users is 6. As shown in the following figure, in a stress test, the average packet delay and jitter of premium users with QCI = 3 are significantly better than those of normal users with QCI = 6. In the case of tight bandwidth resources, users with QCI = 6 cannot seize enough bandwidth resources for their services, whereas users with QCI = 3 can maintain a stable rate of 800 kbps. “Slicing” with 4G is merely a coarse QoS classification on access networks, but 5G slicing will involve a more complete end-to-end solution.
With 5G, there will be network slicing for various scenarios, including slicing for Internet of Vehicles, slicing for VR, and slicing for IoT devices. In addition, the granularity of network slicing will be finer, and QoS-based pricing may be adopted. With 5G, operators may change their billing mode, and bill the B end or both the B end and C end. We will continue paying attention to the implementation of this business, and we look forward to getting help from teams with relevant resources or technologies in the Group.
If we compare the networks constructed by operators with 4G to national highways, the networks constructed with 5G should be seen as including both normal highways and expressways. If you want better service, you may just pay for the expressway.
Mobile Edge Computing
The internal latency of the LTE network is less than 20 ms if retransmission is not considered, but this value is increased to more than 40 ms to 50 ms if an external server needs to be pinged. The transmission rate of optical fibers is 200 km/ms. 5G requires that the access network latency not exceed 0.5 ms when dealing with latency-sensitive use cases. This means that the physical distance between a 5G central machine room (or data center) and a 5G cell (base station) cannot exceed 50 kilometers. Facing the challenge of physical layer latency, we have to consider introducing mobile edge computing (MEC) and edge data centers into access networks, that is, transferring some functions of core networks and application networks down to access networks.
Edge computing is deployed on the edge of the network near things and data sources. It has core capabilities of converged networking, computing, storage, and applications. Many requirements of low latency, massive connection services, aggregation and optimization of data can be met by using the computing power and services provided by edge computing. In this way, the pressure of the load on core networks and backhaul links can be relieved. Therefore, the combination of edge computing and network slicing becomes particularly meaningful.
When network transmission latency or data security is involved, data cannot be directly transmitted to the cloud for processing in many fields. Therefore, there is a widespread trend towards edge computing. The oft-cited autonomous driving is a good example of this, because image processing in autonomous driving needs to be done at the network’s edge. In addition to mobile edge computing in this sense, operators expect that computing power will be deployed closer to access networks to support edge computing, and may eventually be directly deployed in cloud devices in base stations. This is good news for applications that are extremely sensitive to latency.
The Internet of Things (IoT) is becoming increasingly popular. As one of the three major scenarios of IoT, mMTC plays a role in imagining the future intelligent world. However, the mMTC still has some problems that urgently need to be solved. As the 5G KPI requires, connection density shall reach 1,000,000/km2, which is an exciting but misleading number. The so-called one million connections actually are not used for data transmission all at the same time. Instead, they are merely connections, and even intermittent connections or monitoring nodes that each send only one data packet a day. It can be seen that the IoT is relatively widely used in power supply metering at the moment, because metering data is basically uplink data, and it doesn’t require very frequent or real-time transmission. However, many other scenarios have very high demands on real-time or quasi-real-time two-way communication.
The super-strong data connection of the NB-IoT is not really real-time connection. Cell capacity of the NB-IoT is very large. Once an NB-IoT terminal is connected to the network, the core network and the IoT platform maintain the user session state, and the network side maintains the IP session even when the terminal is in the Power Saving Mode (PSM) or Extended Discontinuous Reception (eDRX) mode. However, such an increase in capacity is actually achieved through terminal dormancy rather than any particularly distinct technical improvement. The NB network uses 15 kHz terminal access and 180 kHz bandwidth, and the theoretical value of “concurrent users” is 12. If multiple devices attempt to access the network at the same time, not only does the noise floor increase, but devices beyond the quota also have to queue. Therefore, NB-IoT is suitable only to scenarios with low demand on network rate and latency.
Currently, DAMAI.cn (an entertainment brand in mainland China) has already applied NB-IoT to actual products. NB-IoT uses Constrained Application Protocol (CoAP) and the underlying protocol of the CoAP is User Datagram Protocol (UDP), which is unreliable. Therefore, we use the acknowledgement (ACK) response mechanism of the application layer to ensure reliable data reception.
Since our scenario attaches more importance to the data transmission rate than to power consumption, we disabled PSM and eDRX after discussing with the operator to allow data to arrive as soon as possible. However, some monitoring scenarios are sensitive to power consumption, so they need a dormant mode or other method to save as much energy as possible. This is fine in terms of purely uplink monitoring applications, but it can be difficult if you want to deliver data in near real time. This will result in limitations on use in some scenarios, and there is still a long way to go in terms of low power consumption.
NB’s coverage of scenarios also needs to be improved. For air conditioners in some buildings, the current coverage is sufficient. However, water meters and other meters are installed in enclosed environments, which is difficult for radio signals to penetrate. Consequently, many on-site water meters (such as meters in manholes and stairwells) cannot upload data, causing embarrassment to water meter manufacturers and NB-IoT technology.
The application of 5G in the mMTC scenario will continue to evolve based on NB-IoT and eMTC technologies, and we can look forward to finding better solutions to some existing problems.
Device-to-Device Communication (D2D) is actually an interesting technology that allows devices to communicate directly with each other. Of course, such communication is not completely autonomous, but is under the control of base stations. The base stations are mainly responsible for controlling signaling, so that direct communication between devices is implemented. This may lead to some social scenarios based on proximity characteristics. Vehicle-to-Vehicle (V2V) communication in the Internet of Vehicles (IoV) is a typical IoT-enhanced D2D communication application scenario. By virtue of characteristics in communication latency and proximity discovery, D2D has inherent advantages in the field of vehicle security in the IoV. In a D2D communication pattern, wireless communication can still be established between two neighboring mobile terminals, providing a guarantee for disaster rescue.
Household screen projection is a good example of D2D scenarios. However, in a world already dominated by the Wi-Fi-Direct technology, D2D has a powerful competitor.
With 4G, content distribution networks (CDNs) are basically deployed near core routers (CRs) and service routers (SRs), and their deployment positions are on the upper side. In addition, CDN nodes are sparsely deployed, and each node covers an average radius of 10 kilometers. With 5G, CDN should migrate from CR and SR to users in terms of architecture. In terms of node deployment, CDN should move towards miniaturization and high density, and decrease the average coverage radius from 10 kilometers to 1 kilometer or even less. NFV and SDN technologies in network slicing will also be applied to CDN. NFV implements network resource sharing and flexible expansion. CDN NFV implements hardware and software decoupling. SDN makes scheduling and routing control more flexible, opening up network awareness and centralized control capabilities, and providing flexible scheduling and optimized routing capabilities.
I’d like to cite a viewpoint form Chen Weiru with the Group: The next decade will witness a comprehensive and coordinated upgrade from consumer interconnectivity to industrial interconnectivity. In the future, the Industrial Internet may develop in two directions. The first direction is online and offline are integrated in the industrial connections where you are. For example, a retail business needs to refactor and fuse the online and offline sales scenarios in a digital and visual way. A supply chain business needs digitalization for online and offline convergence to ensure consistent inventory management. The second direction is full link digitalization and connectivity. Comprehensive connectivity fundamentally transforms the ecosystem, consumers, and business models. Therefore, 5G may rely on IoT technologies to implement comprehensive digitalization solutions, boosting the industrial Internet.
Alibaba Cloud’s Whitepaper: “Chinese Enterprises through 2020”
At the World Artificial Intelligence Conference in 2019, Alibaba Cloud published a whitepaper, entitled “Chinese Enterprises through 2020: Practices and Trends in Artificial Intelligence Applications.”
Alibaba Cloud has been applying AI in practical use at scale in all walks of life. Based on the large number of industrial practices, in the whitepaper, the Alibaba Cloud research center analyzed the maturity of AI applications for Chinese enterprises and the evolution in different stages. The center summarized the seven patterns that can help enterprises create and implement their value, guiding enterprises to better employ artificial intelligence.
- Supports mechanical, simple, and repetitive work
- Creates new species in digital economy era
- Breaks through the limit of human capacity
- Activates data, and innovates business and business processes
- Breaks through thinking patterns and discovers potential logic and relationships
- Provides a brand-new man-machine or service interaction pattern
- Assists humans with intelligence decision-making