This work was supported by the National Science and Technology Major Project of the Ministry of Science and Technology of China under Grant No. 2009ZX03003.
1 Background and Objectives
Traditional Universal Mobile Telecommunications System (UMTS) networks are primarilydesigned for voice services. But in the development from 3G, Long Term Evolution (LTE) to future 4G technologies, packet data services act as a driving force in the transition of network architectures. Future communication networks will not only converge various heterogeneous wireless networks, but also integrate wireless networks with wired Internet. With diverse Internet applications and new Internet services being introduced, the capacities of future Radio Access Networks (RANs) must be sufficiently large. As a result, future network architectures should optimize their data service features to satisfy users’ diversified and customized communication demands.
In 3GPP, the LTE project aims to improve performance of existing 3G technologies in order to provide a high packet rate, high spectrum efficiency and short delay. Compared with UMTS networks, LTE networks make great progress toward all-IP flattened architecture. However, more efficient architectures are needed to keep up with the trends of future applications and to offer users a high-speed broadband experience. Meanwhile, existing cellular networks still allocate radio resources based on cells, which makes resource sharing among Base Stations (BSs) difficult in access networks. The networks are also unable to adapt to the
non-uniform traffic rule, thus leading to high network Capital Expenditures (CAPEX) and low equipment utilization. In LTE- or LTE-A-based Orthogonal Frequency Division Multiplexing (OFDM) systems, in order to guarantee spectrum efficiency, cells are networked at the same frequencies. However, the cell-based radio resource allocation scheme is likely to bring inter-cell interference and causes performance of cell-edge nodes to deteriorate dramatically. To avoid the constraints of single cell-based network architectures on network performance, distributed architectures are proposed: distributed configuration of radio resources can minimize the impact of channel interference on network performance, while the distributed antenna system is used in RANs. With distributed deployment of antennas and reasonable allocation of network resources, RAN capacity can be further increased.
Moreover, wireless networks are characterized by particular dynamics: The fading characteristics of radio links change over time, users keep moving in the networks, and users choose services according to their respective preferences. The dynamic characteristics of wireless networks makes them unsuitable for large-scale centralized information processing. In joint signal processing of multiple cells, a distributed approach should be adopted. Distributed signal processing can quickly adapt to changes in network status, reduce network overheads, and improve network robustness, scalability, self-organization, and
self-configurability. These all improve overall network performance.
2 Requirements for Distributed RAN
To achieve inter-BS cooperation, distributed resource allocation, and joint information processing, future RAN architecture should possesses the following features or functions:
- Enable spectrum resource sharing and dynamic resource configuration.
- Share hardware resources and have a resource cooperation mechanism.
- Support multiple modes on the same hardware platform and allow multi-mode BSs to coexist.
- Be flattened architecture with IP-oriented system design for better convergence with Internet.
- Provide ubiquitous coverage and access.
- Consistently deliver wireless access and high-rate data services.
3 Distributed RAN Architecture
To meet the above requirements, we propose a new RAN architecture, which uses distributed radio technologies to deliver cost-effective and high-performance services. In this architecture, the Radio Frequency (RF) unit and Baseband Unit (BBU) of a BS are separated. By shortening the distance between antennas and users, the network’s capacity, energy efficiency and coverage are enhanced. Moreover, with multi-cell Multiple Input Multiple Output (MIMO) technology being used, distributed BBUs are deployed, and the system’s spectrum efficiency and cell-edge node performance are greatly improved.
In current 3G and future LTE systems, distributed BSs, comprising of Remote Radio Units (RRUs) and BBUs, will become more popular. Often, a distributed BS comprises a powerful BBU and several RRUs, which can be flexibly networked to cover a large geographic area. In such a BS, BBU and RRUs are separated and connected with Gigabit optical fibers. Each RRU is configured with transmitting/receiving devices to convert RF signals into digital Intermediate Frequency (IF) signals. Functions such as baseband processing, radio resource management, and network management are implemented by the BBU.
BSs in cooperative RANs operate differently to traditional distributed BSs, where BBUs and RRUs are statically connected. In cooperative radio systems, each RRU does not belong to any specific BBU and all BBUs form a pool with virtualization and high-speed transmission technology. Virtualization technology enables physical resources to be optimally allocated, while load balancing policy makes overall utilization of physical devices more efficient. The joint signal processing coordinates transmission of different RRUs in a distributed way and be finally completed by a BBU of the pool. Several RRUs can achieve high spectrum efficiency when a pool implements joint signal processing. In a cooperative RAN system, a mobile terminal can select suitable RRU to access based on the strengths of received signals. Meanwhile, several mobile terminals can be reasonably paired according to channel correlation to form a virtual MIMO system for joint transmission/reception. This improves system spectrum efficiency further.
The BBU pool can include BBUs with different processing capabilities, and improve resource utilization by means of effective load balancing mechanisms. In addition, the pool can easily employ MIMO technologies, for example, Cooperative Multi-Point (CoMP) transmission/reception in LTE-A, to enable the system to obtain high performance gains.
The advantages of a distributed BBU pool include:
(1) Saving on the costs of housing equipment and peripherals. The number of equipment rooms is greatly decreased, thus construction costs or rents can be saved. Because circuits are shared, investments in Global Positioning System (GPS) devices, control circuits, and interface circuits between BS and Base Station Controller (BSC) are saved.Centralized control and maintenance utilizes a single-point monitor, thus saving human resource expenses, and the energy consumption of the system is reduced.
(2) With a distributed BBU pool, high-density RRUs are employed, the coverage of each cell is shrunk, and the distance from each RRU to UE is shortened. Therefore, each cell requires lower transmitting power than a traditional BS. This means a UE will consume less energy in transmitting signals and its energy-saving will be enhanced.
(3) As a central control point, it can obtain any data and Channel State Information (CSI) within its coverage. Consequently, joint transmission/processing algorithms can be easily scheduled and deployed to eliminate inter-cell interference and improve cell-edge performance.
(4) It supports centralized baseband processing, enabling channel sharing to be implemented on a larger scale than traditional BS systems, and traffic to be balanced among multiple sites. Hence, it can better cope with tide effects and save baseband resources.
(5) As a UE moves, its serving cell must change and handovers take place. Because the BBU pool covers a larger area than a traditional BS system, it reduces the number of handovers and improves network performance.
4 Key Technologies and Challenges for Distributed RAN
In cooperative radio systems, each RRU does not belong to any specific BBU, and all BBUs form a pool using virtualization technology and high-speed transmission technology. Virtualization technology enables optimal allocation of physical resources, while load balancing policy makes overall utilization of physical devices more efficient. Such a BBU pool can implement joint signal processing for several RRUs to achieve high spectrum efficiency.
(1) Cooperative MIMO
Cooperative MIMO technologies[2-5] can use interference signals as useful signals, thereby reducing inter-cell interference and improving the system’s spectrum utilization. Figure 1 illustrates cooperation in different ranges.
Research on cooperative MIMO focuses on:
- Air interface measurement.
- Channel feedback and reference signal design.
- Inter-BS channel information.
- Sharing and distribution mechanisms for data and scheduling information.
(2) High-Speed Transmission System
In order to capture interference signals, cooperative RRUs are deployed in a cross-BS way, as shown in Figure 2. Consequently, high-speed transmission networks are required between RRUs and BBUs for remote data transmission.
Research on high-speed transmission systems focuses on:
- Analysis of deployment scenarios and system-level bandwidth requirements.
- Development of bandwidth reduction technologies to reduce transmission costs. Because the Cpri/Ir interface between RRU and BBU is at a gigabit-level in LTE, in the future, the coverage of each cell will need to be smaller with a greater number of RRUs employed.
- Development of switching interfaces to ensure radio resources are dynamically allocated to BTSs in need (as the Cpri/IrCpri/Ir interface is the final interface).
- Time synchronization and frequency synchronization among multiple points and in wide areas.
(3) Dynamic Radio Resource Configuration
Research on dynamic radio resource allocation focuses on:
- Distributed radio resource configuration optimization technology (power, time, and space).
- Dynamic multi-antenna configuration technology.
- User pairing technology.
(4) Inter-BS Load Balancing
Traditional networking methods are not adaptable to the non-uniform traffic rule, limiting the full utilization fo network resources. Also, traditional network architectures restrict the deployment and application of new technologies, such as distributed radio technology, MIMO, and interference suppression algorithms. Hence, they hinder wireless networks from improving performance.
Research on inter-BTS load balance focuses on:
- Deployment of distributed BSs and connections of their network elements.
- Load balance mechanisms.
- Devices and related signaling procedures.
- Handover procedures.
(5) Software Defined Radio and Virtualization Technologies
Hardware architectures of existing BSs include almost all types of chips, such as Application-Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), Digital Signal Processor (DSP), Network Processor (NP) and Central Processing Unit (CPU). This heterogeneity makes BS hardware resources difficult to manage and share. As a result, the device utilization in some regions is quite low and capacity expansion is needed.
With the quick growth of computation, many companies have introduced their own soft BS systems based on a single-CPU hardware structure. These BS systems support multiple modes (e.g. TD-SCDMA, and TD-LTE) by means of software configuration to meet network demands. Moreover, with virtualization technologies from IT field, these BS uniform hardware platforms enable resources of several BSs to be managed and shared in a unified way. However, there are a series of issues to be studied before CPU is used for digital signal processing.
Research on SDR and virtualization technologies focuses on:
- Issues related to IT-based soft BS, e.g. a chip’s power consumption, hardware accelerator design, scheduling and synchronization.
- Multi-mode software configuration technology.
- Soft BS-based virtualization technology, which implements system resource allocation and management.
(6) Application Layer Optimization Technologies for BSs
The flattening of network architecture enables the network to better meet the demands of users. As the initial access point to the network, the BS has the advantage of capturing users’ behaviors, analyzing these behaviors and selecting optimization policies for the application layer. Specifically, with Deep Packet Inspection (DPI), the BS can conduct statistical analysis of users and select various application Caches (e.g. Web and Video) based on their service characteristics. This will improve user experience. In addition to Internet applications, the geographical features of BS make it quite suitable for directly employment in region-specific enterprise services or end user services.
Research on application layer optimization technologies for BSs focuses on:
- Analysis of optimized hardware implementation platform
- Analysis and research of BS DPI and user behaviors
- Optimization of applications such as diversified region-specific service innovations or caches
5 Performance Evaluation
5.1 Downlink Transmission System Model
Downlink joint processing means several cooperative BSs coordinate transmitted signals of all users via the preprocessing matrix in order to suppress the interference on user signals in advance. Currently, there are two main download preprocessing algorithms: Zero Forcing (ZF)[6-7] and Block Diagonalization. In the CoMP system, the joint preprocessing can be performed in a centralized manner with a central processing unit. These cooperative BSs form a cooperative BS set, serving a UE group that uses the same frequency at the same time. The central processing unit implements joint signal preprocessing to mitigate inter-cell interference and subsequently improve system spectrum efficiency, especially the cell-edge user throughput.
Let us assume there are nt transmit antennas at each cooperative BS and nr receive antennas at each UE. One cooperative BS set consists of M cooperative BSs, which serve N UEs using the same frequency at the same time. In the downlink, the M cooperative BSs and the N UEs can form a (Nnr )×(Mnt ) virtual MIMO system. The downlink transmission system model is shown in Figure 3.
The performance of the CoMP downlink transmission scheme is evaluated by system level simulation. In the evaluation, the Time Division Duplexing (TDD) frame structure with 10 milliseconds radio frame length and 1 millisecond subframe length is applied and we assume that all subcarriers are transmitted with equivalent power. Compared with a conventional
non-cooperation system, where Rel. 8 codebook-based precoding scheme is applied, the CoMP transmission scheme enhances the average downlink cell spectrum utilization by 38%, and
cell-edge user spectrum utilization by 70%.
5.2 Uplink Transmission System Model
Uplink joint processing means several cooperative BSs jointly detect the received signals of all users. MIMO detection algorithms for uplink processing include Minimum Mean Square Error (MMSE) detection and Successive Interference Cancellation (SIC). As shown in Figure 4, we assume that there are nt transmitting antennas at each UE and nr receive antennas at each cooperative BS. One cooperative BS set consists of M cooperative BSs, which serve N UEs that use the same frequency at the same time. In the uplink, the M cooperative BSs and the N UEs can form a (Nnr )×(Mnt ) virtual MIMO system.
Similarly, a system level simulation is used to evaluate the performance of the CoMP uplink transmission scheme. In the evaluation, the TDD frame structure with 10 milliseconds radio frame length and 1 millisecond subframe length is applied and it is assumed that all the subcarriers are transmitted with equivalent power.
The main feature of next generation distributed access networks is that RF units are remotely and distributively deployed, while BBU is centralized, offering services to a large number of RRUs. The distributed deployment of RF units results in small cell radii, and consequently improves the performance of the air interface and reduces energy consumption. The centralized management of BBU not only reduces construction and maintenance costs, but also cancels inter-cell interference and greatly improves the system’s spectrum efficiency if cooperative MIMO technologies are employed. Distributed access network is still a new concept. China Mobile’s research on this enables it to press ahead in the field of infrastructure networks, accelerating technical innovation quickly, and constructing low-cost high-performance networks to serve end users.
 3GPP TR36.913. Requirements for Further Advancements for E-UTRA (LTE-Advanced) (Release 8) [S]. 2008.
 3GPP R1-083410. Text Proposal for RAN1 TR on LTE-Advanced [S]. 2008.
 JOUNG Jingon, KIM Eun Yong, LIM Sung Hoon, et al. Capacity evaluation of various multiuser MIMO schemes in downlink cellular environments [C]//Proceedings of IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’06), Sep 11-14, 2006, Helsinki, Finland, Piscataway, NJ, USA: IEEE, 2006: 1-5.
 LIN Zihuai, VUCETIC B, MAO Jian. Ergodic capacity of LTE downlink multiuser MIMO systems [C]//Proceedings of IEEE International Conference on Communications (ICC’08), May 19-23, 2008, Beijing, China. Piscataway, NJ, USA: IEEE, 2008: 3345-3349.
 ANDREWS J G. Interference cancellation for cellular systems: a contemporary overview [J]. IEEE Wireless Communications, 2005, 12(2): 19-29.
 SPENCER Q H, SWINDLEHURST A L, HAARDT M. Zero-forcing methods for downlink spatial multiplexing in multiuser MIMO channels [J]. IEEE Transactions on Signal Processing, 2004, 52(2):
 CHOI L U , MURCH R D. A transmit preprocessing technique for multiuser MIMO systems using a decomposition approach [J]. IEEE Transactions on Wireless Communications, 2004, 3(1): 20-24.
 ZHANG Jun, CHEN Runhua, Andrews J G, et al. Coordinated multi-cell MIMO systems with cellular block diagonalization [C]//Proceedings of the 41st Asilomar Conference on Signals, Systems and Computers (ACSSC’07), Nov 4-7, 2007, Pacific Grove, CA, USA. 2007: 1669-1673.
The next generation Radio Access Network (RAN) is evolving in a distributed style with network architecture flattening out. Base stations in the next generation RAN cooperatively optimize the radio resource allocation to improve spectral efficiency. In addition, base-band pool technology reduces the CAPEX and OPEX of RAN by greatly improving the efficiency of resource usage and decreasing the space requirements for equipment.