Jiang Wang , Jing Xu, Yang Yang, Haidong Xu

Key Lab of Wireless Sensor Network and Communication, Shanghai 201899, China Shanghai Research Center for Wireless Communications, Shanghai 201210, China SIMIT, Chinese Academy of Sciences, Shanghai 200050, China

* The corresponding author, email: yang.yang@wico.sh

I. INTRODUCTION

Mobile communication has experienced an extraordinarily high-speed development over the past three decades. Mobile internet and Internet of Things (IoT) provide a broader development space for the next generation mobile communication system (5G). 5G will play a key role in enabling and connecting services that will become an essential part of our lives in the future. Many countries (China, USA,Europe, Korea, Japan, etc.) and standardization groups (3GPP, IEEE, ETSI, NGMN, etc.)have announced their 5G development roadmap [1]-[6]. Compared with 4G, 5G needs to support various scenarios, such as enhanced mobile broadband communication (eMBB),massive machine type connections (mMTC)and ultra-reliability & low-latency connections (uRLLC). To meet stringent performance targets, the infrastructure needs to be highly flexible and scalable to meet foreseen and uncertainty requirements. Currently, the global consensus of vision and requirements of 5G is basically reached [7][8], such as both 5G concept and technology roadmap are gradually clarified. Those diverse service requirements bring great challenges to the system design of 5G [9]-[13]. To do this, we need to revolutionize how networks are built and thus one infrastructure with more openness and flexibility should be developed. As we know, fast software iteration based on general hardware with open source software is the information technology’s (IT) classic developing mode.This mode promotes IT’s prosperity with healthy ecosystem and would speed up the convergence between IT and communication technology (CT). The focus of wireless network development should transfer from dedicated hardware to general purpose platform with open source software. That is to say, the conventional software engineers have a good entry to communication region. The boundary between IT and CT doesn’t exist anymore.The wireless network evolution becomes one open innovation process and IT developing mode would act as one more important role.

This paper presents the general purpose processor based software and hardware architectures for 5G mobile cellular networks and analyzes the advantages and challenges of GPP platform. Moreover,the performances of real-time signal processing and power consumption for the general purpose processor based hardware platform are also evaluated.

As we know, the computing complexity of wireless communication systems are increased with the system upgrade, as shown in Figure 1. Especially for 5G system, there maybe exist one sharp rise in the implementation complexity comparing with the 4G system due to the complicated spatial-time signal processing caused by massive MIMO etc. On the other hand, the semiconductor industry is constantly moving forward driven by Moore’s Law. General purpose processor (GPP) has made a lot of progress in the structure, performance and power dissipation in recent years. As shown in Figure 1, the computing capability of GPP increased several magnitudes with Dhrystone Million Instructions executed Per Second(DMIPS) during the past three decades. It provides the choice that the traditional mobile communication base station could be developed through GPP platform [12]-[15].

These new technologies of GPP include multi-core, fixed and floating point operations within a single instruction stream multiple data stream (SIMD), large-capacity on-chip cache and low-latency off-chip memory etc.By means of these new technologies, GPP can accomplish digital signal processing that is usually done by DSP, especially the baseband processing of base station apparatus. Furthermore, different from the field programmable gate array (FPGA) and digital signal processor(DSP), both physic and higher layer can be easily developed and upgraded on a GPP with high-level cross-platform programming languages, rather than hardware description languages. GPP based platform has the “FREE”advantages.

1.Fast-development.As we know, the evolution of wireless network includes requirement definition, protocol standardization and commercialization. The traditional evolution usually takes ten years because the alternative technologies should be evaluated with the complicated dedicated hardware development. However, software and GPP hardware based developing mode can support fast iteration of alternative technologies and thus speeds up the procedure significantly to next generation wireless network without the longtime of dedicated hardware’s development.

2.Re-configurability.The GPP based platform would support the software and hardware decouple, network function virtualization and flexible deployment. All GPP platform resource could be easily shared and re-orchestrated for different network functions and tasks.

3.Extensibility.Since the existence of significantly different requirements, it always takes relatively long time to extend the technology development of conventional wireless network into these vertical industries. GPP based software defined wireless network make the extension to vertical industry become more feasible and easier.

4.Economy.Due to the general purpose hardware, the network evolution (capability upgrade, capacity enlargement, etc.) can be realized without the replacement of infrastructure and thus reduce capital & operational expense significantly. Furthermore,the general purpose hardware would well support cloud infrastructure includes cloud computing, cloud storage, cloud service, etc.On the way to the convergence between IT and CT, several international groups/companies start trying. Based on LTE specification,Eurocom has developed one whole OAI (Open Air Interface) system, which includes open EPC (MME, SGW, PGW, HSS etc.) and open eNB (evolved Node B) [16]. The OAI system is based on Intel CPU and Linux operating system (OS), all functions including PHY signal processing are realized with software. To support rural coverage, Facebook announced in July 2016 that it has developed open source eNB system named OpenCellular [17]. Facebook announced that it has designed and tested an open source and cost effective, software defined wireless access platform aimed to improve connectivity in remote areas of the world. Furthermore, Facebook aims to work with telecom infra project (TIP) members to build an active open source community around cellular access technology development and to select trial locations for further validation of technical, functional, and operational aspects of the platform [18].

However, if we want to build a true GPP based base station, robust real-time digital signal processing should be achieved. In order to solve the problem of processing speed, full advantage of modern GPP featured by multi-core,high capacity and low-latency cache structure and the multi-thread parallel approach should be taken. Thus, designing of real-time signal processing on GPP involves the following key aspects: multithreading execution, high-speed data access, real time OS etc.

For multithreading execution, time-consuming module can be accelerated due to its parallelization can enhance processor resource utilization. The first step of parallelization is task decomposition that divides coarse-grained calculation process into fine-grained subtask. Then subtasks can choose up or down to accomplish vertical or horizontal integration according to data dependency analysis between subtasks. Commonly used ways of task decomposition include data decomposition,calculating process decomposition and problem decomposition.

Furthermore, dividing some long and complex communication processing into several independent or semi-independent threads can also improve the program structure and make the program easier to be processed. In the case of multi-station and multi-user, you can also use multi-threading so that each user process runs independently to reduce inter-thread interference between multiple users.

In addition to the above description, scheduling policy would also effect the real-time signal processing. In general, real-time process has absolute priority compared to non-realtime process.

II. GPP BASED OPEN 5G PLATFORM IMPLEMENTATION

2.1 Motivation of open 5G platform

As discussed above, GPP based platform has huge potential to meet the challengeable requirements of 5G. To meet the stringent performance of 5G, “Softwarization” is an overall techno-economic way for designing,implementation, deployment and operations.From the view of technology, open 5G platform should support such 5G key technologies as SDN/NFV, ultra-dense network (UDN),non-orthogonal multiple access (NOMA),and massive MIMO (Multiple-Input Multiple-Output) etc. Furthermore, open 5G platform would also support fast prototyping of customized industry applications, e.g. video surveillance, smart grid, high-speed rail etc.,as shown in Figure 2. Thus, open 5G platform would benefit from collaborative innovation with other industries. It is expected to become the solution for a unified and open developing platform for 5G in future-oriented applications and services.

2.2 Proposed software architecture of open 5G platform

One suitable and flexible software architecture is a key point of GPP based open 5G platform.Specifically, open 5G software architecture needs to cope with the following requirements to meet the open innovation and the protocol evolution.

1) Customized Network Topology.

Since all network functions can be implemented with software, the migration and re-arrangement of function module is very simple and we can adjust the network topology of the system very flexibly in accordance with our needs by re-deploying software module. For example, the evolved packet core (EPC) function not only can be allocated centralized as an independent network node to serve many base stations, but also can be distributed allocated to generate enhanced base stations.

2) Customized and Scalable System Specifications.

The system specification such as number of users and base stations depend on the ability of software processing, therefore the ability of open 5G platform would be flexibly configured according to the actual need and can support to expand/shrink its processing capability.

3) Modularized Protocol and Algorithm Function.

In pure software architecture, different protocols and algorithm functions are completely implemented by software, so updates of a variety of protocols and algorithm module only need to replace the software module. Both development and maintenance workloads of software modules are far less than those of using FPGA and ASIC (Application Specific Integrated Circuit). For a company, the old modules can be updated by those cost-effective ones with less cost, which means shorter development cycles and new products can be put into market more quickly.

4) Open External Interface

Pure software architecture of open 5G platform can provide a variety of open data interfaces conveniently. These data interfaces include protocol interface, test and maintenance interface, and performance statistics interface.

a) To provide an open protocol interface that can be connected with standard commercial core network, commercial base station and commercial terminal.

b) To provide an open test interface that supports convenient equipment maintenance functions and perfect ability of problem orientation.

c) To provide an open system interface,then you can get various statistical indicators and a variety of real-time performance statistics, such as delay, packet loss rate and call drop rate and so on.

5) Matching Evolved 5G protocols.

5G is still in the early stage, standardization and field testing phases will be done in the future before ultimately achieving commercial deployment. The open 5G platform should support fast iteration of 5G technologies.

Above functional requirements, especially the uncertainty of 5G protocols brings challenges to the design of open 5G software architecture. To alleviate the effects due to the uncertainty of 5G evolution, we apply object-oriented software to develop open 5G platform in the proposed software architecture. It’s well known that object-oriented software system has a clear overall structure,low degree of coupling between modules, high code reusability and high functional scalability. These superior characteristics provide an effective way for the large-scale software development to guarantee reliability, reusability,scalability and maintainability. Furthermore,object-oriented methods provides a convenient approach for matching between physical functions and software modules, modular packaging of protocol functions, differentiation of protocols, the inheritance and evolution of protocol functions.

a) Matching between physical functions and software modules

In accordance with the object-oriented method, the PHY, MAC, RLC, RRC protocol entities are mapped to objects of object-oriented software engineering. This kind of mapping method brings convenience to understand software architecture of the whole system which will improve the efficiency of software development and maintenance.

b) Modular encapsulation of protocol functions

Encapsulation is the foundation to ensure that software components with excellent modularity. Class in object-oriented programming is a template of object and class can contain a variety of methods and data structures that can easily encapsulate protocol function modules.Encapsulation can achieve loose coupling between modules and avoid unnecessary association between modules which is very advantageous to software development and maintenance.

c) Differentiation of protocols

Using inheritance to achieve code sharing is one of the main advantages of object-oriented programming. According to the object-oriented method, different functions with a few differences can use inheritance to share the same functional parts and simplify the system implementation.

d) The inheritance and evolution of protocol functions

Object-oriented inheritance naturally agrees with the evolution of protocols in the future.Considering the current progress of 5G, 4G(LTE) protocol will serve as a basis of 5G to a considerable extent. With the inheritance of class in object-oriented programming, the new features of protocol can be very easily realized on the basis of old protocol and there is no need to re-write those achieved features. For example, we can apply inheritance to define a new interface class S1 on the original basis and achieve different functions of 5G with respect to 4G protocol.

2.3 Proposed hardware architecture for open 5G platform

In addition to the software architecture, one powerful and efficient GPP-based hardware architecture is another important aspect to realize software defined EPC and eNB. As analyzed above, real-time signal processing is the main challenge for GPP platform. To solve this problem, the following proposals are applied in our architecture, as shown in Figure 3.

In this paper, one common commercial server based architecture is applied for open 5G platform, the key GPP hardware include CPU, Synchronous Dynamic Random Access Memory (SDRAM), acceleration card and interface control card for both transmitted and receipted radio signal processing. Moreover,the remote radio unit (RRU) includes Analog to Digital / Digital to Analog (AD/DA)converter, Radio Frequency (RF) and FPGA card for front signal processing and interface control. In GPP platform, the dealt signal is scheduled to inner core for baseband signal processing with multiple core CPU. In the proposed architecture, X86 server is used to deal with high layer protocol and low complexity physical baseband unit (BBU) processing. The complicated signal processing block such as turbo decode could be scheduled to some acceleration card to alleviate the load of CPU.

In the proposed hardware architecture,BBU and RRU are connected with optical fiber according to common public radio interface (CPRI) protocol. This heterogeneous hardware is very important to guarantee the real-time performance of physical layer signal processing, especially for eMBB scenario. The BBU module consists of the multi-core general purpose processors, which are the core part of the software base station. The BBU functions of open 5G platform are divided into:

1) Digital Baseband (DBB)

It mainly finishes baseband processing algorithms of wireless communication, including layer 1, layer 2 and layer 3;

2) Job Scheduling

It mainly completes scheduling of computing resources and guarantees the real time of baseband signal processing to the maximum extent to meet the requirements of the wireless communication.

3) Heterogeneous Computing Module with Hardware Acceleration

It is for the computer-intensive calculation types and can realize high-performance and low-latency computing by accelerating hardware to reduce the computing time and relieve the computational load on the CPU.

4) CPRI Interface in DBB

It conducts CPRI protocol analysis and is connected with the RRU by CPRI protocol.This module has a high-speed data exchange with GPP by peripheral component interconnect express (PCIe) bus and unpacks the data of RRU, then transmits the data to a general purpose processor for processing through direct memory access (DMA) technique, or delivers and packages the data processed by GPP to RRU for sending.

5) C&M

This module makes BBU control and protect itself, RRU and fiber optic connections.

In the proposed GPP based open 5G platform, RRU’s functions can be the same as the radio frequency (RF) unit of the traditional base station and it mainly includes:

1) Front-end digital signal processing, including sampling rate conversion (SRC), the up/ down conversion process (DUP/DDP), the signal amplitude crest factor reduction (CFR),digital pre-distortion (DPD) and so on.

2) Analogy-to-Digital/Digital-to-Analogy(ADC/DAC)

3) CPRI Interface conducts CPRI protocol analysis and is connected with the BBU.

4) C&M controls and protects RRU and fiber optic connections.

Fig. 3 function structure of open 5G base station

Fig. 4 Architecture of software defined mobile network

Fig. 5 Demo of software defined mobile network

Table I system configuration

III. IMPLEMENTATION AND PERFORMANCE EVALUATION OF GPP BASED SOFTWARE DEFINED MOBILE NETWORK

In this paper, one GPP based software defined LTE system and demo are presented. The presented software defined LTE network is fully developed on GPP platform based on X86 architecture, as shown in Figure 4. Both EPC and eNB are deployed in one mini Intel commercial PC (i7-5557U). The EPC and eNB share the server’s resource including operation system (OS), computing resource (CPU), storage (Memory) and drivers etc. In the software developed eNB, physical BBU, MAC/RLC/PDCP and RRC are all implemented in the GPP platform with X86 CPU instruction set to realize the PHY processing. Furthermore, the communication between RF block and BBU block is through high speed USB 3.0 interface.

Currently, this super-flexible software defined mobile network can support stable real-time signal transmissions with a bandwidth of 10MHz and a stable data rate of 32 Mbps for multiple commercial mobile terminals.The system configuration is presented in Table 1. As shown in Table 1, the platform can stably demo both TDD/FDD duplex mode and also support internet service with different commercial terminals. The EPC can support networking with multiple commercial small eNBs, X2 based handover, tracking area update (TAU) and scheduling request (SR) etc.

Besides system implementation of GPP platform, we also try some performance evaluation of GPP platform, the evaluation includes two cases, one is on real-time signal processing and the other is related to power consumption issue.

? Case I: Real-time signal processing evaluation

Due to real-time signal processing is the most challengeable problem for GPP platform,in this paper, we evaluate almost all the baseband signal processing block of eNB to find which block(s) would be the bottleneck of the processing delay performance. The test envi-ronment is presented as follows: IBM System x3400 M3 with 2.13GHz CPU, quad-core Intel Xeon E5606, 4G RAM, 256G HDD, Linux Debian 7 OS with the version 64 bits Ubuntu 14.04 Desktop. In the test, we evaluate the delay performance for different data rate and the tested blocks include (de)scrambling, (de)modulation, (de)interleaving, (de)coding etc.,the system bandwidth is 5MHz. The results of mean signal processing delay are presented in Table 2 and the turbo decoding processing delay distribution is shown in Figure 6 .

From Table 2, we can see that the Turbo decoding is the bottleneck for real-time processing. Furthermore, Figure 6 shows that the variance of turbo decoding processing delay isn’t negligible, especially for higher data rate. That is to say, the jitter of turbo decoding maybe one uncontrollable problem should be further considered.

Another test related to real-time signal processing is on the different CPU frequency, the test configuration and results are presented in Table 3. From Table 3, we can find clearly that the computing capability of CPU is very important and influent the performance significantly.So if we aim to develop the GPP based cellular network, the computing capability of CPU is the key factor we should consider firstly.

? Case II: Power consumption comparison between GPP and dedicated hardware platform

As described in section II, power consumption of GPP platform is another challenge in addition to real-time signal processing. In this section, we test the power consumption between GPP and dedicated hardware platform with different traffic and different UE number.The test configuration is presented in Table 4 and the results are shown in Figure 7.

For both platforms, the power consumption only increase slowly with the UE number due to its always full bandwidth and full power transmission for different UE number. Just as we predicted, the power consumption of GPP platform is higher than the dedicated platform for all the tested traffic modes. From Figure 7,we can see that the power consumption of GPP is almost three times compared with that of dedicated platform. The results show that the power consumption is one un-negligible factor for GPP platform in variable application, especially for power constraint scenarios. To pull GPP solution in different applications, a lower power consumption processor with higher computing capability should be further developed.

Fig. 6 processing delay distribution function of PDSCH turbo decode

Ta ble II Processing delay for different data rate

T able III Platform capability for different CPU frequency

Ta ble IV System Configuration for Power consumption test

Fig. 7 Power consumption comparison between GPP and dedicated hardware

IV. CONCLUSION AND FUTURE WORK

GPP based platform is becoming one interesting alternative solution to develop the next generation mobile communication (5G) due to its “FREE” advantages. In this paper, both software and hardware architecture are proposed for GPP platform towards 5G. Furthermore,some interesting GPP based network demo is presented and performance on signal processing and power consumption is further evaluated.The results indicated that the GPP platform can support real-time signal processing successfully while the power consumption is larger than the dedicated platform with the similar configuration. To guarantee real-time signal processing of wireless communication baseband signal,baseband processing platforms based on GPP need to further enhance the ability of processing delay and jitter in order to support higher bandwidth and more stable performance according to the current performance of baseband processing based on GPP. Our future work would focus on redefine system configuration to meet some vertical industries’ requirements,including system computing, physical layer and higher layer protocol, occupied frequency,power consumption etc.

This work is funded in part by National Natural Science Foundation of China (grant NO.61471347), National S&T Mayor Project of the Ministry of S&T of China (grant NO.2016ZX03001020-003), key program for international S&T Cooperation Program of China (grant NO. 2014DFA11640) and Shanghai Natural Science Foundation (grant NO.16ZR1435100).

Reference

[1] FCC Adopts Rules to Facilitate Next Generation Wireless Technologies, Available at: https://www.fcc.gov/document/fcc-adopts-rules-facilitate-next-generation-wireless-technologies

[2] RADIO SPECTRUM POLICY GROUP STRATEGIC ROADMAP TOWARDS 5G FOR EUROPE , Available at: http://rspg-spectrum.eu/wp-content/uploads/2013/05/RPSG16-032-Opinion_5G.pdf

[3] Communication - 5G for Europe, Available at:https://ec.europa.eu/digital-single-market/en/news/communication-5g-europe-action-plan-and-accompanying-staff-working-document

[4] 3GPP on track to 5G, Available at: http://www.3gpp.org/news-events/3gpp-news/1787-ontrack_5g

[5] SK Telecom, Ericsson and Qualcomm announce collaboration on 5G NR trials to accelerate wide-scale 5G deployments, Available at:http://www.sktelecom.com/en/press/detail.do?idx=1191

[6] Prepping for the 5G World, Available at: http://about.att.com/newsroom/prepping_for_the_5g_world.html

[7] NGMN 5G white paper, available at: https://www.ngmn.org/uploads/media/NGMN_5G_White_Paper_V1_0.pdf

[8] IEEE COMSOC WHITE PAPERS, available at:http://www.comsoc.org/whitepapers

[9] Xiaohu Ge, Tao Han, Yan Zhang, et al, “Spectrum and Energy Efficiency Evaluation of Two-Tier Femtocell Networks with Partially Open Channels,” IEEE Transactions on Vehicular Technology, 2014, 63(3): 1306-1319.

[10] Ran Zi, Xiaohu Ge, John Thompson, et al, “Ener-gy Efficiency Optimization of Frequency Radio Chains for 5G Wireless Communication Systems,” IEEE Journal on Selected Areas in Communications, 2016, 34(4): 758-771.

[11] Xiaohu Ge, Jiaqi Chen, Chengxiang Wang, et al,“5G Green Cellular Networks considering Power Allocation Schemes,” Science China Information Sciences, 2015, 59(2): 1-14.

[12] Guoqiang Mao. 5G Green Mobile Communication Networks [J]. China Communications, 2017,14(2): 183-184.

[13] Tao Y, Liu L, Liu S, et al. A Survey: Several Technologies of Non-Orthogonal Transmission for 5G[J]. China Communications, 2015, 12(10):1-15.

[14] Tao X, Hou Y, Wang K, et al. GPP-Based Soft Base Station Designing and Optimization [J].Journal of Computer Science and Technology,2013, 28(3): 420-428.

[15] B. Guan, X. Huang, G. Wu, C. Chan, M. Udayan and C. Neelam, “A Pooling Prototype for the LTE MAC Layer Based on a GPP Platform,”2015 IEEE Global Communications Conference (GLOBECOM), San Diego, CA, 2015, pp. 1-7.

[16] N. Kai, S. Jianxing, H. Zhiqiang and K. K. Chai,“LTE eNodeB prototype based on GPP platform,”2012 IEEE Globecom Workshops, Anaheim, CA, 2012, pp. 279-284.

[17] B. Guan, X. Huang, G. Wu, C. Chan, M. Udayan and C. Neelam, “A Pooling Prototype for the LTE MAC Layer Based on a GPP Platform,”2015 IEEE Global Communications Conference (GLOBECOM), San Diego, CA, 2015, pp. 1-7.

[18] Flexicell–OAI-based Cloud Radio Access Networks, Available at: http://www.openairinterface.org/?page_id=1638

[19] Facebook’s OpenCellular is a new opensource wireless access platform for remote areas，Available at: https://techcrunch.com/2016/07/06

[20] Facebook: Partnering to build the Telecom Infra Project, Available at: https://telecominfraproject.com/facebook-partnering-to-build-thetelecom-infra-project/