Efficient Network Structure of 5G Mobile Communications Kwang-Cheng Chen1(B) , Whai-En Chen2 , Wu-Chun Chung3 , Yeh-Ching Chung3 , Qimei Cui4 , Cheng-Hsin Hsu3 , Shao-Yu Lien5 , Zhisheng Niu6 , Zhigang Tian6 , Jing Wang6 , and Liqiang Zhao7 1

National Taiwan University, Taipei, Taiwan [email protected] 2 National Ilan University, Yilan City, Taiwan [email protected] 3 National Tsing Hua University, Hsinchu, Taiwan [email protected], {ychung,chsu}@cs.nthu.edu.tw 4 Beijing University of Post and Telecommunications, Beijing, China [email protected] 5 National Formosa University, Huwei, Taiwan [email protected] 6 Tsing Hua University, Beijing, China {niuzhs,zgtian,wangj}@tsinghua.edu.cn 7 Xidian University, Xi’an, China [email protected]

Abstract. 5G mobile communications requires system and network considerations from many aspects. Instead of high spectral efficient physical layer communication, We introduce efficient network structure supplying a new design paradigm to meet user experience, spectral efficiency, and energy efficiency, under wide range of services and applications on top of 5G mobile communication networks, with virtual networking over software defined networking facilitating.

1

Introduction

5G mobile communications and subsequent issues on wireless communication systems and wireless networks emerge as the most important technology challenge at this time. It is widely addressed regarding 1,000 times of data rate [11]. However, in this paper, we wish to supply a different view on efficient network structure to facilitate 5G mobile communications as a paradigm shift. Tradition network design considers efficiency to satisfy quality of services (QoS) based on connection-oriented traffic like voice and video. However, the traffic has significantly changed in 5G and actually also for 4G. Internet traffic like web browsing dominates in mobile communication networks, particularly social media. In this paper, instead of common targets on aggregated data rate and quality of service (QoS) metrics, we would like to focus on the efficient network structure of 5G mobile communications, in particular the quality of user experience (QoE), spectrum efficiency, and energy efficiency that is c Springer International Publishing Switzerland 2015  K. Xu and J. Zhu (Eds.): WASA 2015, LNCS 9204, pp. 19–28, 2015. DOI: 10.1007/978-3-319-21837-3 3

20

K.-C. Chen et al.

related to battery at any specific user equipment (UE) or any device/machine [3]. Researchers have focused on the spectrum efficiency at physical layer, while the aggregated data rate can not really reflect user’s demand. We consider more important to satisfy network throughput for a specific user equipment or a machine. In other words, to satisfy quality of user experience (QoE) is more important in the 5G systems and networks. As Internet traffic dominates in 5G networks, latency is a critical factor of users’ QoE, working together with cloud computing. These considerations form a unique view to design the efficient network structure of 5G mobile communications. The efficient network structure is even more complicated by considering another extreme of traffic from Internet of Things (IoT) that is usually composed of small packets from tremendous number of devices and machines, as the foundation of cyber-physical-socio systems. In this paper, we are indicating the approach by information-centric processing to establish spectral and energy efficient network structure for 5G.

2

Cloud Based Service Architecture

Although the heterogeneous cloud radio access networks (H-CRAN) have shown the potentials in spectrum efficiency and energy efficiency enhancements to support emerging applications, the second characteristic of providing ultralow latency services is still a huge challenge. To enhance the end-to-end latency performance, a proper resource scheduling/allocation scheme is to reduce the transmission delay in the air interface. Adopting an access control policy is also effective to reduce latency in wired/wireless backhaul. However, the unacceptable signaling overheads in the air interface still impose large data exchange delay to existing mobile networks. In addition to latency resulted from signaling overheads, the H-CRAN further induces two new sorts of latency that may be severer than that in the air interface. The first one is latency in resource optimization. Radio resource optimization is a widely discussed issue, and the existing results reveal that the computational complexity is increased along with the number of available radio resources, the number of devices, and the number of eNBs. Unfortunately, it is projected that the number of devices will exponentially increase in the following decade. To support the increasing number of devices, the number of available radio resources should be increased as well as the number of eNBs in the H-CRAN. This growing computational complexity may eventually obstruct the latency performance. The second one is latency in the routing/paging procedures to forward data to a mobile device in the H-CRAN. In the existing mobile network design, the routing and paging procedures assume that each mobile device may communicate with any other mobile devices and servers. Therefore, fixed routing/paging information with a hierarchical information inquiry scheme is adopted. However, this design fully ignores the fact that a mobile device may frequently communicate with the mobile devices or web servers within its social network, while rarely exchange data with terminals/servers outside its social network. The existing routing/paging procedures

Efficient Network Structure of 5G Mobile Communications

21

Fig. 1. Could services and applications on top of 5G mobile communication networks

may therefore result in a severer delay than that in the air interface. We can thus obtain the design paradigm of low-latency network architecture [1]. Figure 1 depicts an illustration of the integration in which mainly consists of five parts: IoT devices and machines, UniCloud platform, Cloud-RAN, core network and applications. In the device domain, M2M devices are able to communicate with each other via heterogeneous wireless techniques like WiFi and Bluetooth; or even communicate with application servers via 4G LTE and next 5G communications. We attempt to not only propose efficient algorithms to approach green M2M communications, but also introduce a HetNet GW to leverage multiple wireless communications for legacy devices that might not be equipped with a 4G/5G module. With mature techniques in cloud computing, both CRAN and core network are desired to be hosted on UniCloud to sustain more flexible and scale-out communication services. The baseband unit of an eNB is able to be deployed on a set of virtualized resources of UniCloud based on the separation design of RRU and BBU in Cloud-RAN. Moreover, virtualized core components of a vEPC are able to be deployed on UniCloud with the techniques of network function virtualization. Accordingly, 5G communications and data transmissions between radio access network and backhaul data network could be processing on the cooperative 5G cloud platform.

3

New Domain/Layer Network Structure

The objective of this network architecture is to explore the sustainable development model for future mobile communication network. The sustainability could be expressed in 3 dimensions:

22

K.-C. Chen et al.

1. Sustainable business model As we known, telecommunication operator is facing the scissor effect of floor revenue growth and explosive traffic (then cost) growth. To seek for a healthier situation, the telecommunication industry has to change the business model in two sides. Firstly, more valuable services have to be provided to mobile internet industry for new revenue resource; Secondly, we want to leverage the Moore’s Law of IT infrastructure (including both hardware and software) to lower the OPEX and CAPEX. Previously, mobile internet service provider stay in one end of mobile network and users stay in the other end of mobile network, so the mobile network is saw as pipe or connectivity. With the competition pressure from WiFi and Bluetooth, mobile networks value (revenue) can not be valued more. However current mobile internet service provider are building their own application network (cloud computing) to serve more and more users, mobile network should sell networking services to mobile internet service provider to be paid more. With the effort from IC vendor like Intel, the hardware cost or performance obey the Moore’s Law, so we can use not so many hardware deployment to support exponential growing users/services. Actually, the software obeys Moore’s Law also. Open source or cloud computing make the cost (R&D investment) does not follow the exponential growing of users/services. Comparing to OPEX, CAPEX in telecommunication industry does not occupy remarkable ratio in revenue. Therefore OPEX should also leverage the Moore’s Law of IT software infrastructure. 2. Sustainable network architecture According to cooper’s law, mobile network throughput per unit area (bps/m2) increased 1000000 times for past 45 years, it says, mobile network throughput per unit area doubled every 27 months. For past 45 years, we used 25X spectrum and improve the spectral efficiency 25X, and the cellular density increase 1600X. However, for next 1000X traffic increasing in next 10 years, we have only 3X more spectrum and 6X spectral efficiency, so the cellular density have to be increased 56X. That means the focus for technical development of mobile network has to change from improving the spectral efficiency to refine the architecture to accommodate exponential scale increasing. 3. Sustainable evolution mode Because of the objective of business model shift and the CAPEX saving, the evolution mode of one generation per 10 years has to a changed to smoother and smarter one. The network architecture should encourage such evolutional mode. All these requirements lead to a kind of open network architecture. Open network services to mobile internet service providers; Open network virtualization interface to COTS IT infrastructure; Open network protocol architecture to different player (operator, system vendor, mobile virtual network operator (MVNO), network optimization service provider, third-party innovational developer etc.) who concern different aspects of network and operating, different RAT, different vertical industry scenario, different region, different technical trial, etc. We therefore can organize the entire mobile network into 4 domains and 4 layers as Fig. 2.

Efficient Network Structure of 5G Mobile Communications

23

Fig. 2. Domain and layer structure of 5G mobile communications.

For different cross points of domains and layers, corresponding schemes can be developed with the same goal of open network architecture: • • • • • • •

Open Open Open Open Open Open Open

Operation & Business : Networking/Operation as a Service Cellular Networking : HCA (Hyper Cellular Architecture) FrontHaul : Coverage Subsystem Infrastructure :YaRAN ( Yet another RAN) Network Protocol : NoStack (Not only Stack) Terminal Architecture : Nostack4UE & YaRAN4UE Air Interface Waveform : Soft Defined Air Interface

Here, our focuses are HCA, YaRAN and NoStack. These technical solutions are mapped in to various system components of a 5G mobile network. Hypercellular network suggests a network concept in which signaling and data are decoupled at the air interface to mitigate the signaling overhead. Consequently, the need to rely on cellular operation can be alleviated and control overhead could be subsequently reduced to allow energy efficient operation of base stations.

4

Information Centric Wireless Network Design

Due to the booming of Internet application, social media and data generated by IoT has become the major portion of modern communication networks [4,5]. Consequently, the attractive network design of 5G mobile communication relies on interaction of cyber-social systems, to fully utilize the spectrum and to smooth hybrid heterogeneous networks (multi-band systems, cellularWiFi, macro-micro-femto cell networks, etc.). On the other hand, to provide a compatible wireless system, one fundamental concept of a cyber-physical-social

24

K.-C. Chen et al.

network can be implemented by introducing a new information processing network between physical network and application/social network. The information processing network focuses on how to match wireless resource to most popular or needed social content and increase the total performance of the whole system. It consists of two main function: information cache and content-based (socialbased) grouping. In the part of information cache, the information processing network determines what content should be cached. Due to limited cached space and efficiency issue, the information processing network can only cache those most popular contents to decrease the end-to-end delay between cloud server and consumer. Another issue is how to group social network into a sub-network to form a file sharing network to reduce the burden of the core-network. By forming a file sharing network among consumers according to their social relationship, core network need not to transfer data to consumers separately. Instead, only those key consumers access core network to retrieve information and share them among their social networks. Consequently, 5G networking actually has a kind of cyber-physical-socio scenario that is illustrated as Fig. 3. To achieve the goal of information-sharing among tens or thousands of devices, such as machine-to-machine network [2], network grouping according to their social relationship is necessary. That is, information processor should focus on how to match wireless resource to most popular or needed social content and increase the total performance of the whole

Fig. 3. The cyber-physical-social networking scenario in 5G

Efficient Network Structure of 5G Mobile Communications

25

system. To find the balance between overhead of inferring and performance, Least Absolute Shrinkage and Selection Operator (LASSO) is an immediate tool to determine community members associated with different users, instead of finding the exact isolated social structure. Ideally, we can group sensors into almost disjoint clusters to simplify the algorithms for efficient transmission and device management. In practice, this situation does not always arise, and the algorithm to solve such problems is NP-hard. Instead, we choose the most dependent nodes of a given sensor to form its community and control the size of the community and accuracy of data recovery by l1 -norm optimization. A fraction of the sensor measurement can thus be eliminated to save transmission energy. A LASSO problem can be formulated as 1 min τ |ui |1 + ||x[−i] ui − xi ||22 2

(1)

where xi is the observation signal and x[−i] is the sensing matrix which does not include ith community. The nodes corresponding to non-zero entries of ui are collected together to form the community structure. Here, we focus on minimize mean square error. E(||x[−i] ui − xi ||22 ) and minimize the community to reduce feedback overhead simultaneously. With the proper setting, which indicates the weighting of community size, we can form a small community of nodes with expected error less than pre-defined error level. Of course, leveraging sparse signal processing [2,5] is another plausible approach to tackle general ultra dense networks. An important characteristics of Internet is that most of transferring data flows are generated by seldom file sources. However, in the current communication architecture, these popular data flows are processed separately. Such large amount of data suffer from severe end-to-end delay while passing through core network of communication system. Here, we propose tomography and in-network computations to find the most popular information flows in the mobile communication systems [6,7]. The tomography is a tool that we can predict the structure of the whole system by observing the feedback of stimulation. We successfully apply this technology into cognitive radio network (CRN). CRN tomography is shown in Fig. 4. The message flow originally exists in CRN, θ is the unknown information of interests, and y represents observation taking values in an observation set Γ which may be a set of vectors. The purpose of CRN tomography is to infer θ according to y and full/partial information about s. There are two strategies to infer the value of θ: passive inference and active inference. For passive monitoring, we can obtain the inference result of θˆ by θˆ = Φp (y, Υ [s),

(2)

where Υ (., .) is the passive inference. For active monitoring, the probing signal is transmitted to induce or enhance the correlation between target information and observation. The probing signal p may be completely or partially known beforehand, and we use the information extraction function Λ[p] to characterize

26

K.-C. Chen et al.

Fig. 4. Scheme of radio network tomography.

a priori knowledges of p. Therefore, we can formulate the active tomography problem as θˆ = ΦA (y, Γ[p], Υ [s]).

(3)

By stimulating the system with probing signal p and the resulting signal y, we can know the parameters of the system. After understanding the structure of the system, we can easily identify those popular information in the network. Based on earlier study on social network properties of mobile communication networks and through in-network computations and cache [8] these information in suitable routers or devices via information from social relationship, we can significantly reduce the end-to-end delay. This step actually involves tradeoff between computing/storage and communication/networking, which is another frontier in communication networking to facilitate IoT traffic of small packets and massive social media delivery in an efficient way.

5

Virtual Network Realization over Software Defined Networking Platform

Building up a platform that can integrate the social information and the resource in physical network is the key technology for future communication networks. SDN has been proposed as a possible solution to integrate network resource in different network layers. The main features of SDN include: separation of control and data planes; centralized and programmable control planes of network equipment; support of multiple, isolated virtual networks. The SDN provides a programmable framework to facilitate network configuration of more flexibility. In other words, via the SDN structure, the network is not only for data transmission but also for information flows of different applications. That is, SDN converts a realistic network into a virtual network and therefore more flexibility than before. To serve 5G traffic of both IoT and social media, we propose new architecture of SDN in following. Figure 5 illustrates the architecture of 5G wireless communications, from which, we can find that there are two main data flows: social

Efficient Network Structure of 5G Mobile Communications

27

Fig. 5. Implementation of virtual network by software-defined network.

media data flow and IoT data flow, have been integrated into SDN. With the help of SDN, we can build up two virtual networks to serve video and IoT traffic flow separately and reach the optimal performance. In this section, we firstly focus on how the property of complex network can influence the implementation of realistic communication systems. Second, we aim to develop the communication theoretical breakthroughs and new networking design paradigm through the networking research of information centric networking among this architecture, and hence reach the effective information delivery so as to greatly enhance the spectral-efficient and energy-efficient in communication systems and networks. For example, we have successfully developed an algorithm to control the information over a network. How to implement such a successful control scheme into our platform is a promising research direction. All in all, the research results shall serve as helpful tool in networked data processing problems in different scientific areas, thus to facilitate the information-centric processing for communication and social networks. Consequently, we can fully take advantage of the concepts developed with system examples in [9], to revolutionize the system design to accommodate totally different traffic types, via the effective management of radio resource to PHY, and thus to facilitate CRAN compatible with cloud computing applications and services. This task is further fertilized by implementation of CRAN protocol stack on cloud computing and on top of software radio implementation for both core network and hyper-cellular access network [10].

28

6

K.-C. Chen et al.

Conclusion

This paper suggests the efficient network structure and realization of 5G mobile communications by considering different aspects. Cloud services and applications are therefore possible on top of this efficient network structure. We expect such an approach to generally meet the diverse goal of the 5G mobile communications. Acknowledgement. This study is conducted under the “Energy-Efficient Mobile Communication Systems Project” of the Institute for Information Industry which is subsidized by the Ministry of Economic Affairs of the Republic of China.

References 1. Lien, S.Y., Hung, S.C., Chen, K.C., Liang, Y.C.: Ultra-low latency ubiquitous connections in heterogeneous cloud radio access networks. To appear in the Special Issue on Heterogeneous Cloud Radio Access Networks, IEEE Wireless Communications 2. Chen, K.C., Lien, S.Y.: Machine-to-machine communications: technologies and challenges. Ad Hoc Netw. 18, 3–23 (2014) 3. Han, N.D., Chung, Y., Jo, M.: Green data centers for cloud-assisted mobile ad hoc networks in 5G. IEEE Netw. 29, 70–76 (2015) 4. Condoluci, M., Dohler, M., Araniti, G., Molinaro, A., Zheng, K.: Toward 5G densenets: architectural advances for effective machine-type communications over femtocells. IEEE Commun. Mag. 53, 135–141 (2015) 5. Wunder, C., Boche, H., Strohmer, T., Jung, P.: Sparse signal processing concepts for efficient 5G system design. IEEE Access 3, 195–208 (2015) 6. Lien, S.Y., Cheng, S.M., Shih, S.Y., Chen, K.C.: Radio resource management for QoS guarantees in cyber-physical systems. IEEE Trans. Parall. Distrib. Syst. 23(9), 1752–1761 (2012) 7. Yu, C.K., Cheng, S.M., Chen, K.C.: Cognitive radio network tomography. IEEE Trans. Veh. Technol. 59(4), 1980–1997 (2010) 8. Niesen, U., Shah, D., Wornell, G.W.: Caching in wireless networks. IEEE Trans. Inf. Theory 58(10), 6525–6540 (2012) 9. Lien, S.Y., Chen, K.C., Liang, Y.C., Lin, Y.H.: Cognitive radio resource management for future cellular networks. IEEE Wirel. Commun. 21(1), 70–79 (2014) 10. Zhao, T., Yang, P., Pan, H., Deng, R., Zhou, S., Niu, Z.: Software defined radio implementation of signaling splitting in hyper-cellular network. In: SIRF. ACM (2013) 11. Andrews, J.G., Buzzi, S., Choi, W., Hanly, S.V., Lozano, A., Soong, A.C.K., Zhang, J.C.: What will 5G be? IEEE J. Sel. Areas. Commun. 32(6), 1065–1082 (2014)