Over the past decade, wireless communications achieved great success and progress. People can not only use the traditional voice services, but also enjoy the high-speed data traffic. Recently, the advent of emerging Internet of things (IoT) applications further promotes wireless networks to shift the paradigm from promoting spectral efficiency to connecting everything. Therefore, in order to efficiently support the IoT applications, the fifth generation (5G) system proposes two additional usage scenarios as the supplement of enhanced mobile broadband (eMBB), namely ultra-reliable and low latency communication (URLLC) and massive machine type communication (mMTC) . The use case with high reliability and low latency is a very important application scenario of 5G. For example, some of delay-sensitive services will demand an end-to-end latency of a few milliseconds, while the fields such as wireless automation and control may in addition require reliabilities in terms of block error rates on the order of .
As a typical application scenario of IoT, vehicular networks (VNETs) are widely studied in recent years. Driven by VNETs, the automotive industry is also undergoing key technological transformations, and vehicles are connected to share various information. In order to integrate the advantages of dedicated short range communication (DSRC) and long term evolution (LTE) systems, heterogeneous vehicular networks (HetVNETs) are proposed not long ago . With the aid of HetVNETs, the experience of sharing information and cooperation driving on the road are significantly improved. However, the increasingly complex road conditions pose new challenges to HetVNETs. The future smart vehicles not only rely on the abundant sensors to deal with the safety-related applications, but also require the ultra-reliable and low latency wireless networks to exchange real-time information.
Besides the performance indicator, the increasingly complex conditions also stimulates the rapid development of communication pattern. In order to enrich the application types, the automotive and telecommunication industries put forward two novel communication patterns for the next generation VNETs (NGVNETs), namely vehicle-to-network (V2N) and vehicle-to-pedestrian (V2P) communications. Integrating with vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications, four patterns are generally referred to as vehicle-to-everything (V2X) . These four kinds of V2X communications can provide more intelligent services for end-users via cooperation. For instance, the advent of autonomous driving (AD) pushes extensive opportunities of NGVNETs . In order to enhance the reliability of AD, the predictive maintenance of sensors plays an indispensable role for AD systems. Original manufacturers can gather the huge volume of data generated by sensors, and analyze them to offer the real-time assistance of control. To this end, URLLC and diverse communication patterns become essential parts for NGVNETs.
In general, each service has its specific latency and reliability requirements. For example, safety-related applications have higher priority than non-safety-related ones, hence the requirement levels of latency and reliability are higher. Similarly, event-triggered services are more reliable and timely than periodic ones. To meet different requirement levels, various emerging techniques of physical (PHY) and media access control (MAC) layers should be investigated in detail. However, conventional wireless networks are designed with the objective of maximizing spectral efficiency, without paying much attention to the URLLC requirements. Therefore, it is necessary to study the relation between latency and reliability. By introducing URLLC applications and requirements, this paper proposes and summarizes several performance evaluation methods of latency and reliability from different viewpoints. Specifically, we discuss three methods in PHY layer from theoretical and practical viewpoints. As for MAC layer, the cases of single- and multi-node are studied respectively. Finally, some promising solutions are presented in order to tackle potential challenges.
The remainder of this paper is organized as follows. In Section II, some typical use cases of V2X are discussed, where the corresponding requirements are also analyzed. In order to accurately evaluate various techniques of URLLC, Section III proposes several performance evaluation methods from different viewpoints. Then some of advanced techniques and solutions are subsequently presented in Section IV. Finally, conclusions and future outlook are drawn in Section V.
Ii Underlying Application Scenarios
In this section, four communication patterns (V2V, V2I, V2N and V2P) and six application scenarios are discussed. As shown in Fig. 1, the scenarios are divided into two categories, where one is based on V2V and V2P patterns, and the other is based on V2I and V2N. In general, V2V communications can broadcast the messages to the surrounding vehicles to exchange V2V application information, such as velocity and location, etc. V2P communications are similar to V2V, while the main difference is that one of the tranceivers is a pedestrian. Moreover, due to the limitation of power consumption of pedestrian user, the frequency of V2P communications is lower than V2V. Unlike V2V and V2P, V2I and V2N communications focus on the functionalities of large-scale cooperation. Furthermore, V2N pattern supports the communications with evolved packet switching.
Ii-a V2V and V2P Patterns
There exist a few similar characteristics between V2V and V2P, hence we discuss them together in this subsection, and so does the next subsection. In the V2V and V2P patterns, most application scenarios are safety-related. Three main types of scenarios are studied as follows.
Ii-A1 Driving and Road Safety
As conventional vehicular applications, driving and road safety are first discussed here. With more and more vehicles being connected to the networks, the issues of traffic accidents can be eased. For instance, in order to avoid the accidents generated by blind spot, vehicles can broadcast the overtaking messages to the surrounding vehicles via V2V communications. Furthermore, some of the other use cases are also included but not limited to these, such as lane changing (via V2V) and collision avoidance (via V2V/V2P), etc.
Ii-A2 Cooperative Awareness and Control
According to the incomplete statistics, most traffic accidents are caused by the negligence of drivers. Therefore, the drivers are being remove from the closed-loop system – “vehicle, road and driver”. With the development of sensors, AD is expected to be a competitive candidate techniques to support the above operation. However, the ability of sensors on a single AD vehicle is limited. Thus the AD reliability can be effectively improved by multi-vehicle cooperative awareness. Moreover, utilizing ultra-reliable and low latency NGVNETs, we can also integrate the different control functions of AD vehicles into one control plane, which is convenient to monitor and manage them. On the other hand, V2P communications can be seen as the most reasonable way to let pedestrians interact with AD vehicles in this case.
Ii-A3 Mobility as a Service
With the rapid development of mobile Internet, mobility as a service (MaaS), as an emerging mode, is proposed recently . Unlike the above safety-related use cases, MaaS is non-safety-related one. Generally speaking, MaaS offers end-users a solution to find the most appropriate means of transports and telecommunications. For example, in order to enjoy more efficient and comfort travel, pedestrians can broadcast their actual needs to the surrounding vehicles via V2P communications, so that they can quickly find the suitable vehicles. Similarly, (AD) vehicles can find the pilots to enjoy accurate navigation through V2V/V2P communications. Besides, due to the characteristics of flexible mobility, (AD) vehicles can be used as mobile relays to meet the communication requirement in the poor coverage areas.
Ii-B V2I and V2N Patterns
The application scenarios of V2V and V2P patterns mainly focus on small-region. On the contrary, most cases of V2I and V2N concentrate on large-region cooperation. Here three main types of application scenarios are also discussed, i.e.,
Ii-B1 Traffic Efficiency
It is well known that traffic congestion and environmental pollution have become important global issues in the transport industry. To overcome these problems, NGVNETs are indispensable elements in the future. In general, the traffic flow is similar to fluid dynamics, the macro control of flow adopting V2I (V2N) is more efficiently than that of V2V. Then, with the help of reasonable routes, the issues of congestion and pollution can be eased. Furthermore, V2I (V2N) communications can also deal with the problems of broadcast storm caused by the emergency vehicles.
Ii-B2 Periodic Report
Periodic report as an important and typical scenario in the V2I and V2N patterns, is listed here. There exist two aspects for periodic report, namely manual driving and AD vehicles. As for manual driving vehicles, not only the machinery state messages but also the road condition information should be reported to the infrastructures, which helps department of transportation comprehensively coordinate the road safety. While for AD vehicles, besides the above messages, the huge volume of data generated by sensors should be sent simultaneously. Then the services of predictive maintenance can be offered by the powerful background servers and advanced analysis algorithms.
Ii-B3 Social Entertainment on the Road
This scenario can be regarded as the eMBB services in NGVNETs. Hence, the basic applications of eMBB can be shifted here. Despite this, some particular features in the vehicular environment should be considered, such as regional and self-organization characteristics. This case may study in other literatures in detail .
Ii-C Requirement Levels
Different scenarios result in different requirements of latency and reliability. Obviously, concerning the reliability of AD vehicles, cooperative awareness and control have the highest level among these cases. However, social entertainment on the road is not sensitive to latency and reliability, but requires high data rate. Various requirement levels of latency and reliability are summarized in Table I.
|Pattern||Typical scenario||Latency||Reliability||Data rate||
|V2V and V2P||Driving and road safety||Ultra-high||Ultra-high||Low||
|Cooperative awareness and control||Ultra-high||Ultra-high||Medium|
|Mobility as a service||Medium||Medium||High|
|V2I and V2N||Traffic efficiency||High||High||Low||
|Social entertainment on the road||Low||Low||High|
Iii Performance Evaluation Methods for URLLC
The conventional wireless networks are mainly designed with the objective of promoting spectral efficiency, without paying much attention to the URLLC requirements. In order to meet the URLLC requirements, it is necessary to study the relation between latency and reliability. In this section, several performance evaluation methods of latency and reliability are discussed from the viewpoints of PHY and MAC layers.
Iii-a PHY Layer
In order to fully understand latency and reliability in PHY layer, we study the relation between them from theoretical and practical viewpoints.
Iii-A1 Theoretical Viewpoint
Shannon’s ground-breaking work is the creation of the abstract model of communication, which converts many practical engineering problems into appropriate mathematical problems. One of the outstanding parts among Shannon’s achievements is to study how much information can be transmitted reliably through channel, and it is also referred to channel coding. Hence, we discuss the relation between latency and reliability based on the information-theoretic viewpoint.
The formula of channel capacity tells us the maximum number of bits transmitted by channel with no error. However, in order to achieve this capacity, there raise two limitations namely latency and complexity. If we want to transmit the information with channel capacity, we must pay infinite coding length (latency) and huge complexity. Obviously, it goes against the requirements of URLLC. Therefore, the researchers study the new results in recently years to deal with these two limitations, then we have the following approximation 
where is the channel capacity, is the so-called channel dispersion, and denotes the inverse of the Gaussian function. and
is the block length and error probability, and they can be seen as the proxies for latency and reliability, respectively. Unlike the classical formula of channel capacity, Equ. (1) clearly shows that the relation between latency, reliability and capacity. Here let us give an example to illustrate it. Consider an additive white Gaussian noise channel with signal-to-noise ratio (SNR) equaling 0 dB . We can calculate the capacity by Shannon’s formula, i.e. . This result shows that we can communicate at rates 0.5 bits per channel use with no error by the infinite number of channel uses. But this result is always limited by latency requirements for the practical engineering. Hence, we decrease the capacity by a factor 0.9 and tolerate a error probability , whereupon the minimum number of channel uses is reduced from infinity to . Obviously, it is beneficial for the URLLC requirements. The above formula is based on the time-invariant channel. With the regard to the real-world wireless channel, the results of block-fading and time-variant channel can be found in .
In general, we discuss the outage capacity instead of the channel capacity (also referred to as the ergodic capacity) in wireless communications. Therefore, the outage probability is modeled as the reliability here. Our research is based on the orthogonal frequency division multiplexing (OFDM)-related system model. The time duration and frequency bandwidth of one resource element are defined as and . Then the relation between latency and reliability is
where denotes the total bandwidth, is the packet size (the number of bits) and .
is the probability density function (PDF) of instantaneous SNR, and is related to the characteristics of channels.
Iii-A2 Practical Viewpoint (BLER)
Although the error probability and outage probability have closed-form expressions, there still exist some gaps between theoretical analyses and practical uses. In particular, it is hard to directly use the above results in the practical systems, such as LTE-related systems. Hence, in order to tackle this question, we remodel the reliability as block error rate (BLER), which is more actual for the existing systems. Since there adopts the simulation method, the relation between latency and reliability is considered as an interface function. Further, a total of five parameters are treated as the inputs, and BLER as the proxy of reliability is regarded as the output.
As illustrated in Fig. 2, these five input parameters are the packet size, total bandwidth, latency, coding rate and average SNR. The usage and idea of interface are described in detail as follows. First, various SNR-BLER curves are obtained by the means of simulation according to all kinds of typical input parameters. Then, the average SNR can be calculated by curve fitting, which is based on the actual system states and corresponding requirements (including and BLER, etc.). At last, in order to meet this average SNR, the appropriate transmission schemes are provided, which is composed of the modulation and coding scheme (MCS) and configuration of multi-antenna, etc. Overall, compared with outage probability, BLER is a more efficient and practical performance metric for real use.
Iii-B MAC Layer
PHY layer mainly focuses on transmission latency also called air interface latency, while MAC layer pays much more attention to scheduling latency. In this subsection, we discuss scheduling latency from the viewpoints of single-node and multi-node, respectively.
Iii-B1 Single-node Viewpoint
The relation between latency and reliability of single-node is similar with the system reliability. Hence, first of all, let us elaborate the background of system reliability theory. Generally, there are three main branches of reliability, namely hardware reliability, software reliability and human reliability. Here we mainly concern with the first one: the reliability of components and systems. In the theory, the reliability function is described as a probability that the time to failure (random variable) is greater than a value.
Inspired by the above contents, we can easily shift this paradigm to MAC layer. Then the reliability is modeled as the probability that the scheduling latency does not exceed an expected threshold. Taking advantage of the cumulative distribution function (CDF) of latency, the general expression for latency and reliability can be given by
where is the PDF of scheduling latency. The conceptual illustration of Equ. (3) is plotted at Fig. LABEL:sub@fig:3a as follows. As shown in Fig. LABEL:sub@fig:3a, there is a tradeoff between latency and reliability. The general expression of latency and reliability offers us a guide to future research and design, that is, our optimization goal is to let the curve move left as far as possible.
On the other hand, many mathematical theories can be also used for analyzing the scheduling latency, such as queueing theory, stochastic network calculus and Markov process, etc. Besides, an empirical CDF can be acquired by the means of simulation.
Iii-B2 Multi-node Viewpoints
The above analysis is based on the case of single-node, without the consideration of multi-node cooperation. As for the case of multi-node, we can reuse Equ. (3) to evaluate the performance of the entire network. In particular, with the aid of graph theory, the physical networks are transformed to the abstract models. As shown in Fig. LABEL:sub@fig:3b, the abstract models consist of parallel, serial or hybrid systems. Then the latency and reliability can be calculated by the combination of Equ. (3). Moreover, utilizing the outage probability and BLER, we can jointly evaluate the latency and reliability of PHY and MAC layers. It is noted that more complicated factors should be considered in the transformation, such as the connectivity of node, etc.
Iv Promising Solutions for URLLC
The application scenarios of URLLC pose new challenges to the techniques of PHY and MAC layers. To deal with these challenges, several promising solutions are discussed in this section. Moreover, network architecture design is also considered here, such as network slicing.
Iv-a Flexible Transmission Solutions of PHY Layer
Error probability indicates the coding performance, while outage probability focuses on the performance of transmission techniques. It is well know that multiple-input multiple-output (MIMO) is a key technique for OFDM-related systems. MIMO has two modes namely diversity and multiplexing. Diversity is generally seen as a means to enhance reliability. Hence, with the aid of outage probability, the effects of MIMO to latency and reliability are investigated at first.
Fig. 4 shows the relation of latency and reliability with various configurations of multi-antenna, where the channel is Rayleigh fading. Maximum ratio transmission (MRT) and maximum ratio combining (MRC) are adopted here. It is noted that MRT in multiple-input single-output has the same performance with MRC in single-input multiple-output, thus we only plot the case of single-input multiple-output. As shown in Fig. LABEL:sub@fig:4a, the outage probability monotonically decreases with the increase of latency. It means that there is a tradeoff between latency and reliability. Furthermore, the outage probability decreases with the increasing number of antennas. As for the given latency and , the outage probability decreases with the increasing SNR. Similarly, the latency also decreases with the increasing SNR. To this end, the high average SNR is conducive to enhance latency and reliability. Many 5G emerging techniques can be adopted for URLLC to improve the average SNR, such as massive MIMO, millimeter wave communications and hybrid beam-forming, etc . At last, the correlation of channel is considered. Compared with Fig. LABEL:sub@fig:4a, Fig. LABEL:sub@fig:4b clearly illustrates that the correlation can reduce the reliability.
Besides the above techniques, another promising solution is the novel numerology design in order to efficiently support URLLC. As shown in Fig. LABEL:sub@fig:5a, due to low latency, the numerology of URLLC has the largest subcarrier spacing, leading to the shortest transmission time interval (TTI). The new numerology should keep good backward compatibility with existing LTE/LTE-advanced systems.
With the interface function mentioned in Section III-A, we can define a flexible PHY layer interface to meet various service requirements. Many emerging PHY layer techniques can be involved in the interface, such as massive MIMO and new waveforms. For instance, Fig. LABEL:sub@fig:5b illustrates the curves between SNR and BLER based on the frame structure of LTE/LTE-advanced. As can be seen from Fig. LABEL:sub@fig:5b, if we want to transmit the packet with latency 0.5ms and reliability , the average SNR should be greater than or equal to +8.6 dB, and the modulation scheme should be 64-quadrature amplitude modulation (QAM). According these conclusions, the flexible air interface can adopt appropriate techniques to meet these requirements, such as multi-antenna diversity. To this end, the interface function of latency and reliability can efficiently promote the implement of flexible PHY layer transmission.
Iv-B Advanced Scheduling Schemes of MAC Layer
Broadcast is able to meet the requirements of V2V and V2P services, especially for those ultra-reliable and ultra-low latency applications, such as cooperative awareness and control, etc. In order to eliminate the additional latency and overhead of scheduling, we usually adopt carrier sense multiple access (CSMA)-related broadcast strategies (like DSRC system). Hence, there inevitably exist broadcast storms, and result in the performance degradation. Recently, several contention-based strategies are studied, where the most representative ones are spares code multiple access (SCMA) and multi user shared access (MUSA) . Their core ideas are to achieve non orthogonal access in code domain, and prevent transmission collision. In summary, the contention-based mechanism is the promising solution for broadcast services.
Besides the broadcast, another method for V2X applications is to adopt LTE-based device-to-device communications. Usually, one transmission is scheduled by the base stations or access points with a request-grant procedure in the LTE, which brings the extra overhead and latency. Therefore, grant-free multiple access techniques should be studied in depth. As a temporary transition, semi-persistent scheduling (SPS) scheme is discussed in the recent research . In the SPS, the resource pool of schedule assignments and its associated data resource pool are jointly employed with the frequency division multiplexing manner rather than the time division multiplexing one. Obviously, this way is beneficial to decrease latency and improve resource utilization.
Another key performance indicator of scheduling is reliability. Taking advantage of performance evaluation method mentioned in Section III-B, we can compare the performance of different algorithms efficiently and accurately. On the other hand, because of the characteristics of self-organization in V2X, some distributed algorithms are the emphases of future research, as well as open-loop centralized algorithms .
Iv-C Network Slicing
The current wireless network utilizes a relatively monolithic framework to carry all kinds of services, such as mobile data from smart phones, high-speed vehicles and embedded machine-to-machine (M2M) devices, etc. With the rapid development of mobile Internet in recent years, many emerging services pose challenges to not only various PHY and MAC layer techniques, but also network architectures. In order to efficiently support the vertical industry applications (such as mMTC and NGVNETs, etc.), and manage network functionalities and infrastructures, end-to-end network slicing is regarded as a promising solution to these challenges.
In general, the intrinsic concept of network slicing is network virtualization. The physical networks are sliced into multiple virtual networks, which is the so-called network slicing. Each slicing is designed and optimized for the specific requirements and service. Moreover, network slicing can also be extended to the radio access networks, not just in the core networks .
As illustrated in Fig. 6, integrating the above various techniques, several network slicing are proposed for different requirement levels. For example, as for driving and road safety as well as cooperative awareness and control, an ultra-reliable and ultra-low latency slicing is designed with the shortest TTI and grant-free access. Compared with V2V and V2P, V2I and V2N are mainly centralized, hence massive MIMO and efficient centralized scheduling can be adopted in the base stations or access points. Furthermore, MaaS and social entertainment on the road can be support by the mMTC and eMBB slicing, respectively. Network slicing can not only maximize the use of the previous infrastructures, but also keep good backward compatibility. Therefore, network slicing provides a feasible evolutional roadmap from 4G to 5G, and offers better services to end-users.
URLLC, as one of the three main scenarios in the future, is paid much attention by various industries, especially the automotive industry. In order to evaluate the emerging 5G techniques for URLLC, we mainly proposed and summarized several performance evaluation methods in this paper. In PHY layer, utilizing the outage probability and BLER, the relation between latency and reliability were illustrated from theoretical and practical viewpoints, respectively. Besides, the capacity formula of finite blocklength was also used for assessing the coding performance. As for MAC layer, inspired by the system reliability theory, we discussed the general expression for latency and reliability with the help of the CDF of latency. Based on the above methods, the promising solutions to deal with potential challenges have been proposed at the end of this paper. Finally, more research are still needed to address many other challenges of URLLC in the future.
-  IMT Vision - Framework and overall objectives of the future development of IMT for 2020 and beyond, ITU-R Std. M.2083-0, Sep. 2015.
-  K. Zheng, Q. Zheng, P. Chatzimisios et al., “Heterogeneous vehicular networking: A survey on architecture, challenges, and solutions,” IEEE Commun. Surveys Tuts., vol. 17, no. 4, pp. 2377–2396, Fourthquarter 2015.
-  “Study on LTE-based V2X services,” 3GPP, Tech. Rep. TR 36.885, Jun. 2016.
-  K. Zheng, Q. Zheng, H. Yang et al., “Reliable and efficient autonomous driving: The need for heterogeneous vehicular networks,” IEEE Commun. Mag., vol. 53, no. 12, pp. 72–79, Dec. 2015.
-  “5G-PPP white paper on automotive vertical sector,” White Paper, 5G-PPP, Oct. 2015.
-  T. Luan, R. Lu, X. Shen et al., “Social on the road: Enabling secure and efficient social networking on highways,” IEEE Wireless Commun., vol. 22, no. 1, pp. 44–51, Feb. 2015.
-  Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel coding rate in the finite blocklength regime,” IEEE Trans. Inf. Theory, vol. 56, no. 5, pp. 2307–2359, May 2010.
-  G. Durisi, T. Koch, and P. Popovski, “Toward massive, ultrareliable, and low-latency wireless communication with short packets,” Proc. IEEE, vol. 104, no. 9, pp. 1711–1726, Sep. 2016.
-  M. Rausand and A. Høyland, System reliability theory: Models, statistical methods, and applications. Hoboken, New Jersey: John Wiley & Sons, 2004.
-  “User equipment (UE) radio transmission and reception,” 3GPP, Tech. Rep. TS 36.101, Sep. 2016.
-  J. Andrews, S. Buzzi, W. Choi et al., “What will 5G be?” IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1065–1082, Jun. 2014.
-  C. Bockelmann, N. Pratas, H. Nikopour et al., “Massive machine-type communications in 5G: Physical and MAC-layer solutions,” IEEE Commun. Mag., vol. 54, no. 9, pp. 59–65, Sep. 2016.
-  S. hui Sun, J. ling Hu, Y. Peng et al., “Support for vehicle-to-everything services based on LTE,” IEEE Wireless Commun., vol. 23, no. 3, pp. 4–8, Jun. 2016.
-  C. She, C. Yang, and T. Q. S. Quek, “Radio resource management for ultra-reliable and low-latency communications,” IEEE Commun. Mag., vol. 55, no. 6, pp. 72–78, Jun. 2017.
-  K. Samdanis, X. Costa-Perez, V. Sciancalepore et al., “From network sharing to multi-tenancy: The 5G network slice broker,” IEEE Commun. Mag., vol. 54, no. 7, pp. 32–39, Jul. 2016.