C-RAN The Road Towards Green RAN White Paper Version 3.0 (Dec, 2013)
China Mobile Research Institute
Table of Contents C-RAN............................................................................................................................................... i The Road Towards Green RAN ..................................................................................................... i 1 Introduction ............................................................................................................................ 3 1.1 Background ......................................................................................................................... 3 1.2 Vision of C-RAN .................................................................................................................. 4 1.3 Objectives of this White Paper ....................................................................................... 4 1.4 Status of this White Paper ............................................................................................... 5 2 Challenges of Today’s RAN ............................................................................................... 6 2.1 Large Number of BS and Associated High Power Consumption .............................. 6 2.2 Rapid Increasing CAPEX/OPEX of RAN.......................................................................... 7 2.3 Interference in LTE networks .......................................................................................... 9 2.3 Explosive Network Capacity Need with Falling ARPUs............................................. 13 2.4 Dynamic mobile network load and low BS utilization rate ..................................... 14 2.5 Growing Internet Service Pressure on Operator‟s Core Network.......................... 14 3 Architecture of C-RAN ....................................................................................................... 16 3.1 Advantages of C-RAN ..................................................................................................... 18 3.2 Technical Challenges of C-RAN ..................................................................................... 20 4 C-RAN deployment scenarios ........................................................................................ 23 4.1 TD-SCDMA C-RAN deployment ..................................................................................... 23 4.2 TD-LTE C-RAN deployment ........................................................................................... 26 5 Technology Trends and Feasibility Analysis ...................................................................... 30 5.1 Wireless Signal Transmission on Optical Network.................................................... 30 5.2 Dynamic Radio Resource Allocation and Cooperative Transmission/Reception . 39 5.3 Large Scale Baseband Pool and Its Interconnection ........................................................... 42 5.4 Open Platform Based Base Station Virtualization ................................................................ 43 5.5 Distributed Service Network ................................................................................................. 47 6 Recent Progress .................................................................................................................. 49 6.1 C-RAN Field Trials ............................................................................................................... 49 6.1.1 TD-SCDMA and GSM Field Trial ................................................................................ 49 6.1.2 TD-LTE C-RAN Field Trial ........................................................................................... 55 6.2 Cooperative radio technologies under C-RAN ........................................................... 57 6.3 PoC development on C-RAN BBU pooling .................................................................. 60 6.4 Progress on C-RAN virtualization ................................................................................. 69 6.5 Edge Applications on C-RAN ......................................................................................... 74 i
China Mobile Research Institute
7 Evolution Path....................................................................................................................... 78 8 Global landscape of C-RAN activities .......................................................................... 81 9 Conclusions ........................................................................................................................... 82 Acknowledgements ............................................................................................................... 84 Terms and Definitions .......................................................................................................... 85 References................................................................................................................................. 88
ii
China Mobile Research Institute
1 Introduction 1.1 Background Today‟s mobile operators are facing a strong competition environment. The cost to build, operate and upgrade the Radio Access Network (RAN) is becoming more and more expensive while the revenue is not growing at the same rate. The mobile internet traffic is surging, while the ARPU is flat or even decreasing slowly, which impacts the ability to build out the networks and offer services in a timely fashion.. To maintain profitability and growth, mobile operators must find solutions to reduce cost as well as to provide better services to the customers. On the other hand, the proliferation of mobile broadband internet also presents a unique opportunity for developing an evolved network architecture that will enable new applications and services, and become more energy efficient. The RAN is the most important asset for mobile operators to provide high data rate, high quality, and 24x7 services to mobile users. Traditional RAN architecture has the following characteristics: first, each Base Station (BS) only connects to a fixed number of sector antennas that cover a small area and only handle transmission/reception signals in its coverage area; second, the system capacity is limited by interference, making it difficult to improve spectrum capacity; and last but not least, BSs are built on proprietary platforms as a vertical solution. These characteristics have resulted in many challenges. For example, the large number of BSs requires corresponding initial investment, site support, site rental and management support. Building more BS sites means increasing CAPEX and OPEX. Usually, BS‟s utilization rate is low because the average network load is usually far lower than that in peak load; while the BS‟ processing power can‟t be shared with other BSs. Isolated BSs prove costly and difficult to improve spectrum capacity. Lastly, a proprietary platform means mobile operators must manage multiple none-compatible platforms if service providers want to purchase systems from multiple vendors. Causing operators to have more complex and costly plan for network expansion and upgrading. To meet the fast increasing data services, mobile operators need to upgrade their network frequently and operate multiple-standard network, including GSM, WCDMA/TD-SCDMA and LTE. However, the proprietary platform means mobile operators lack the flexibility in network upgrade, or the ability to add services beyond simple upgrades. In summary, traditional RAN will become far too expensive for mobile operators to keep competitive in the future mobile internet world. It lacks the efficiency to support sophisticated centralized interference management required by future heterogeneous networks, the flexibility to migrate services to network edge for innovative applications and the ability to generate new revenue from revenue from new services. Mobile operators are faced with the challenge of architecting radio network that enable flexibility. In the following sections, we will explore ways to address these challenges.
China Mobile Research Institute
3
1.2 Vision of C-RAN The future RAN should provide mobile broadband Internet access to wireless customers with low bit-cost, high spectral and energy efficiency. The RAN should meet the following requirements:
Reduced cost (CAPEX and OPEX)
Lower energy consumption
High spectral efficiency
Based on open platform, support multiple standards, and smooth evolution
Provide a platform for additional revenue generating services.
Centralized base-band pool processing, Co-operative radio with distributed antenna equipped by Remote Ratio Head (RRH) and real-time Cloud infrastructures RAN (C-RAN) can address the challenges the operators are faced with and meet the requirements. Centralized signal processing greatly reduces the number of site‟s equipment room needed to cover the same areas; Co-operative radio with distributed antenna equipped by Remote Radio Head (RRH) provides higher spectrum efficiency; real-time Cloud infrastructure based on open platform and BS virtualization enables processing aggregation and dynamic allocation, reducing the power consumption and increasing the infrastructure utilization rate. These novel technologies provide an innovative approach to enabling the operators to not only meet the requirements but advance the network to provide coverage, new services, and lower support costs. C-RAN is not a replacement for 3G/B3G standards, only an alternative approach to current delivery. From a long term perspective, C-RAN provides low cost and high performance green network architecture to operators. In turn operators are able to deliver rich wireless services in a cost-effective manner for all concerned. C-RAN is not the only RAN deployment solution that will replace all today‟s macro cell station, micro cell station, pico cell station, indoor coverage system, and repeaters. Different deployment solutions have their respective advantages and disadvantages and are suitable for particular deployment scenarios. C-RAN is targeting to be applicable to most typical RAN deployment scenarios, like macro cell, micro cell, pico cell and indoor coverage. In addition, other RAN deployment solution can serve as complementary deployment of C-RAN for certain case.
1.3 Objectives of this White Paper The objective of this white paper is to present China Mobile‟s vision of C-RAN and provide a research framework by identifying the technical challenges of C-RAN architecture. We would like to invite both industry and academic research institutes to join the research to guide the vision into reality in the near future.
4
China Mobile Research Institute
1.4 Status of this White Paper This document version 3.0 is an update on previous version 2.5 released in October 2011. It is not yet fully complete and there may still be some inconsistencies. However, it is considered to be useful for distribution at this stage. It is expected that new research challenges might be added in future versions. Comments and contributions to improve the quality of this white paper are welcome.
China Mobile Research Institute
5
2 Challenges of Today’s RAN 2.1 Large Number of BS and Associated High Power Consumption As operators constantly introduce new air interface and increase the number of base stations to offer broadband wireless services, the power consumption gets a dramatic rise. For example: in the past 5 years, China Mobile has almost doubled its number of BS, to provide better network coverage and capacity. As a result, the total power consumption has also doubled. The higher power consumption is translated directly to the higher OPEX and a significant environmental impact, both of which are now increasingly unacceptable. The following figure shows the components of the power consumption of China Mobile. It shows the majority of power consumption is from BS in the radio access network. Inside the BS, only half of the power is used by the RAN equipment; while the other half is consumed by air condition and other facilitate equipments. Obviously, the best way to save energy and decrease carbon-dioxide emissions is to decrease the number of BS. However, for traditional RAN, this will result in worse network coverage and lower capacity. Therefore, operators are seeking new technologies to reduce energy consumption without reducing the network coverage and capacity. Today, there are quite a number of „amendment‟ technologies that helps reduce BS‟ power consumption, such as the software solutions which save power through turning off selected carriers on idle hours like midnight, the green energy solutions which offer solar, wind and other renewable energy for base station‟s power supply according to local natural conditions, and the energy-saving air conditioning technology which combined with the local climate and environment characteristics, reduce the energy consumption of the air conditioning equipment, etc. However, these technologies are supplementary methods and cannot address the fundamental problems of power consumption with the number of increasing BS. In the long run, mobile operators must plan for energy efficiency from the radio access network architecture planning. A change in infrastructure is the key to resolve the power consumption challenge of radio access network. Centralized BS would reduce the number of BS equipment rooms, reduce the A/C need, and use resource sharing mechanisms to improve the BS utilization rate efficiency under dynamic network load.
6
China Mobile Research Institute
Transmission, 15%
Other Support Equipment, 3%
Management office, 7%
Cell site, 72%
Channel, 6%
Air Conditioners, 46%
Major Equipment, 51%
Fig. 2-1 Power Consumption of Base Station
2.2 Rapid Increasing CAPEX/OPEX of RAN Over recent years, mobile data consumption has experienced a record growth among the world‟s operators as subscribers use more smart phones and mobile devices, like tablets. To satisfy this consumer usage growth, mobile operators must significantly increase their network capacity to provide mobile broadband to the masses. However, in an intensifying competitive marketplace, high saturation levels, rapid technological changes and declining voice revenue, operators are challenged with deployment of traditional BS as the cost is high, the return is not high enough. Average Revenue Per User (ARPU) are all affecting mobile operators‟ profitability. They become more and more cautious about the Total Cost of Ownership (TCO) of their network in order to remain profitable and competitive.
Fig. 2-2: Increasing CAPEX of 3G Network Construction and Evolution
Analysis of the TCO
The TCO including the CAPEX and the OPEX results from the network construction and operation. The CAPEX is mainly associated with network infrastructure build, while OPEX is mainly associated with network operation and management.
China Mobile Research Institute
7
In general, up to 80% CAPEX of a mobile operator is spent on the RAN. This means that most of the CAPEX is related to building up cell sites for the RAN. The historical CAPEX expenditure of 2007-2012 forest are shown in Fig.2-2. Because 3G/B3G signals (deployed frequency 2GHz have higher path loss and penetration loss than 2G signals (deployed frequency 900MHz), multiple cell sites are needed for the similar level of 2G coverage. Thus, the dramatic increase was found in the CAPEX when building a 3G network. The CAPEX is mainly spent at the stage of cell site constructions and consists of purchase and construction
expenditures.
Purchase
expenditures
include
the
purchases
of
BS
and
supplementary equipments, such as power and air conditioning equipments etc. Construction expenditures include network planning, site acquisition, civil works and so on. As shown is Fig.2-3, it is noticeable that the cost of major wireless equipments makes up only 35% of CAPEX, while the cost of the site acquisition, civil works, and equipment installation is more than 50% of the total cost. Essentially, this means that more than half of CAPEX is not spent on productive wireless functionality. Therefore, ways to reduce the cost of the supplementary equipment and the expenditure on site installation and deployment is important to lower the CAPEX of mobile operators.
Fig. 2-3: CAPEX and OPEX Analysis of Cell Site OPEX in network operation and the maintenance stage play a significant part in the TCO. Operational expenditure includes the expense of site rental, transmission network rental, operation /maintenance and bills from the power supplier. Given a 7-year depreciation period of BS equipment, as shown in Fig. 2-4, an analysis of the TCO shows that OPEX accounts for over 60% of the TCO, while the CAPEX only accounts for about 40% of the TCO. The OPEX is a key factor that must be considered by operators in building the future RAN. The most effective way to reduce TCO is to decrease the number of sites. This will bring down the cost for the construction of the major equipment; and will minimize the expenditure on the installation and rental of the equipment incurred by their occupied space. Fewer sites means the corresponding cost of supplementary equipment will also be saved. This can significantly decrease the operators‟ CAPEX and OPEX, but results in poorer network coverage and user experience in the traditional RAN. Therefore, a more cost-effective way must be found to minimize the non-productive part of the TCO while simultaneously maintaining good network coverage.
8
China Mobile Research Institute
Fig. 2-4 TCO Analysis of Cell Site
Multi-standard environment
It is understood that the large number of legacy terminals, 2G, 3G, and B3G infrastructure will coexist for a very long time to meet consumers‟ demand. Most of the major mobile operators worldwide will thus have to use two or three networks (Table 1) [1]. In the new economic climate, operators must find ways to control CAPEX and OPEX while growing their businesses. The base station occupies the largest part of infrastructure investment in a mobile network. Multi-mode base station is expected as a cost efficient way for operators to alleviate the cost of network construction and O&M. In addition, sharing of hardware resources in a multi-mode base station is the key approach to lower cost.
Table 1. Multi-Network Operation of Major Mobile Service Providers Cellular Technologies
Vodafone
TMobile
√
√
√
WCDMA One
France Telecom
Verizon
SK Telecom
Telstra
China Unicom
√
√
√
√
√
√
TD-SCDMA
CDMA EVDO
China Mobile
&
2000
&
√
GSM GPRS EDGE
√
√
LTE
√
√
√
√
√ √
√
2.3 Interference in LTE networks LTE is designed to operate with frequency reuse factor (FRF) of one to improve spectrum efficiency, which is different from both 2G and 3G network with FRF larger than one. OFDM and SC-FDMA are the essential downlink and uplink transmission technologies for LTE. The orthogonality among different sub-carriers eliminates the intra-cell interference. However, since all the cells operate on the same frequency band, the inter-cell interference from and to the
China Mobile Research Institute
9
adjacent cells becomes unavoidable, which leads to low-throughput performance. How to avoid and eliminate inter-cell interference becomes an important researching subject for LTE. In the inter-cell interference tests in the trial networks, the comparison tests in terms of SINR and single-user throughput have been done on the condition of different system loads. The results are illustrated in Figure 2-5 and Figure 2-6. Comparing to 0% load case, the downlink average SINR is decreased by 5.33dB and 8.28dB respectively, and the downlink throughput is decreased by 40% and 55% respectively in case of 50% load and 100% load.
Fig. 2-5 SINR Changes under different loadings
10
China Mobile Research Institute
Fig. 2-6 DL Throughput Changes under different loadings The co-channel interference in LTE are mainly attributed to two patterns: 3 or more adjacent cells overlap and PCI mode 3 conflict. For the interference induced by the PCI model 3 conflict pattern, the handover is not obviously affected. It is observed that the handover success rate is decreased by 2 percent at most. The reason is that the SINR of the target cell is too low which causes Radom Access Process to fail when the UE receives the handover command. However, the pattern has much impact on the traffic performance. In case of 0% load, the cell edge throughput is degraded by 4%~18%. In case of non-zero loading, the CRS SINR and the cell edge throughput are little affected (0.5~2dB decrease for CRS SINR and less than 10% decrease for cell edge throughput). The interference due to 3 or more adjacent cell overlapping has much higher impact on cell edge throughput. It is found that when the number of neighboring cells with 6dB less than the serving cell decreases from 3 to 2, then there is a noticeable increase on user throughput with 30% improvement on average. It is also found that the interference from intra-cell has more impact than neighboring cells. Switching off intra-cell can have a big increase on user throughput (58% on average) while only 4% throughput improvement on average is observed when switching off the neighboring cells. In addition, reducing the number of neighboring cells or their transmission power can also help to improve the system performance.
In LTE networks, it is very common of coverage overlapping with neighboring cells. In our test, we defined adjacent cell as the cell which RSRP is at most 10dB less than the serving cell and made a statistical results on the number of adjacent cells. The result is shown in Figure 2-7. It can be seen that in high-density urban area with inter-cell distance of from 300 to 500 meters,
China Mobile Research Institute
11
the probability for a UE to find one or more adjacent cells is as high as 71.8%. In some cases, the UE can even find 6 adjacent cells. Cell sectorization technology is usually used for 3 intra-site cells to set them to different orientation. It is clear that on the cell edge, overlapping is unavoidable for coverage sake. According to the statistics shown in figure 2-8, the probability is 30.1% for UEs to detect the signals coming from the intra-site adjacent cells. At the same time, the probability is 1.4% for UEs to simultaneously detect the signals coming from the intra-site 3 cells.
Fig. 2-7: The statistics of the number of adjacent cells in large-scale network (RSRP is lower than the main cell within 10dB)
Fig. 2-8: The statistics of the number of adjacent cells loaded an eNB in large-scale network (RSRP is lower than the main cell within 10dB) Through the comparison tests, it can be seen that how to reduce the co-channel interference is the major problem and challenge for large-scale LTE networks. At present, there are many interference coordination technologies such as ICIC, CoMP etc. However the gain from those technologies is limited under traditional distributed architecture. On the contrary, a centralized
12
China Mobile Research Institute
C-RAN architecture can facilitate their implementation and fully exploit their gain on system performance.
2.3 Explosive Network Capacity Need with Falling ARPUs Data rate of mobile broadband network grows significantly with the introduction of air-interface standards such as 3G and B3G; this in turn speeds up end user‟s mobile data consumption. Some forecasts indicated the number of people who access mobile broadband will triple in next several years, after LTE and LTE-A are deployed.
These findings reflect the fact that the
increasing bandwidth of wireless broadband triggers the increase in mobile traffic, because the mobile users can use a variety of high-bandwidth services, such as video-based applications. This new trend will become a serious challenge to future RAN. Based on the forecast data [2], global mobile traffic increases 66-fold with a compound annual growth rate (CAGR) of 131% between 2008 and 2013. The similar trend is observed in current CMCC network. On the contrary, the peak data rate from UMTS to LTE-A only increases with a CAGR of 55%. Clearly, as shown in Fig. 2-9, there is a large gap between the CAGR of new air interface and the CAGR of customer‟s need. In order to fill this gap, new infrastructure technologies need to be developed to further improve the performance of LTE/LTE-A.
Fig. 2-9 Mobile Broadband Data-rates/Traffic Growth On the other hand, the revenue of mobile operators is not increasing at the same pace as the network capacity they provide. Mobile operators‟ voice volumes are steadily increasing and the data volume grows quickly, but revenues are not and ARPUs are even falling in some case. In order to face the slow growth in revenue, operators are forced to constantly hold down costs – notably operating costs. That means mobile operators must find a low cost, high-capacity access network with novel techniques to meet the growth of mobile data traffic while keeping a healthy, profitable growth.
China Mobile Research Institute
13
2.4 Dynamic mobile network load and low BS utilization rate One characteristic of the mobile network is that subscribers are frequently moving from one place to another. From data based on real operation network, we noticed that the movement of subscribers shows a very strong time-geometry pattern. Around the beginning of working time, a large number of subscribers move from residential areas to central office areas for work; when the work hour ends, subscribers move back to their homes. Consequently, the network load moves in the mobile network with a similar pattern,so called "tidal effect". As shown in Fig.2-10, during working hours, the core office area‟s Base Stations are the busiest; in the nonwork hours, the residential or entertainment area‟s Base Stations are the busiest.
Fig. 2-10 Mobile Network Load in Daytime Each Base Station‟s processing capability today can only be used by the active users in its cell range, causing idle BS in some areas/times and oversubscribed BS in other areas. When subscribers are moving to other areas, the Base Station just stays in idle with a large of its processing power wasted. Because operators must provide 7x24 coverage, these idle Base Stations consume almost the same level of energy as they do in busy hours. Even worse, the Base Stations are often dimensioned to be able to handle a maximum number of active subscribers in busy hours, thus they are designed to have much more capacity than the average needed, which means that most of the processing capacity is wasted in non-busy time. Sharing the processing and thus the power between different cell areas is a way to utilize these BS more effectively.
2.5 Growing Internet Service Pressure on Operator’s Core Network With the hyper-growth of smart phones as well as emerging 3G embedded Internet Notebook, the mobile internet traffic has been grown exponentially in the last few years and will continue to grow more than 66x in the next 5-6 years. However because of increasingly competition between mobile operators, the projected revenue growth will be much lower than the traffic growth. There will be a huge gap between the cost associated with this mobile internet traffic and the revenue generated, let alone the mobile operators needing to spend billions of dollars
14
China Mobile Research Institute
to upgrade their back-haul and core network to keep up with the growing pace. This is a huge common challenge to all the mobile operators in the wireless industry. The exponential growth of mobile broadband data puts pressure on operators‟ existing packet core elements such as SGSNs and GGSNs, increasing mobile Internet delivery cost and challenging the flat-rate data service models. The majority of this traffic is either Internet bound or sourced from the Internet. Catering to this exponential growth in mobile Internet traffic by using traditional 3G deployment models, the older 3G platform is resulting in huge CAPEX and OPEX cost while adding little benefit to the ARPU. Additional issues are the continuous CAPEX spending on older SGSNs & GGSNs, the higher Internet distribution cost, the congestion on backhaul and the congestion on limited shared capacity of base stations. Therefore, offloading the Internet traffic, as close to the base stations as possible, can be an effective way to reduce the mobile Internet delivery cost.
Fig. 2-11 Wireless traffic on a commercial 3G Meanwhile it is interesting to understand how people are using today‟s mobile internet. A recent research paper [3] published by one major TEM may give us a glimpse of the most popular mobile applications. It is surprising to see that people are gradually using mobile internet just like they use the fixed broadband network. Content services which include content delivered through web and P2P are actually dominating the network traffic. Fig.2-11 is an example of wireless traffic on a commercial 3G operator. Considering this usage pattern, do we have better choice than just blindly spending billions of dollars to upgrade back-haul and the core network?
China Mobile Research Institute
15
3 Architecture of C-RAN We believe Centralized processing, Cooperative radio, Cloud, and Clean (Green) infrastructure Radio Access Network (C-RAN) is the answer to solve the challenges mentioned above. It‟s a natural evolution of the distributed BTS, which is composed of the baseband Unit (BBU) and remote radio head (RRH). According to the different function splitting between BBU and RRH, there are two kinds of C-RAN solutions: one is called „full centralization‟, where baseband (i.e. layer 1) and the layer 2, layer 3 BTS functions are located in BBU; the other is called „partial centralization‟, where the RRH integrates not only the radio function but also the baseband function, while all other higher layer functions are still located in BBU. For the solution 2, although the BBU doesn‟t include the baseband function, it is still called BBU for the simplicity. The different function partition method is shown in Fig.3-1. Antenna
Solution 2 Solution 1 GPS Core network
Main Control & Clock
Baseband processing
BBU
Digital IF
…
Transmitter /Receiver
…
PA & LNA
… …
RRU
Fig. 3-1 Different Separation Method of BTS Functions Based on these two different function splitting methods, there are two C-RAN architectures. Both of them are composed of three main parts: first, the distributed radio units which can be referred to as Remote Radio Heads (RRHs) plus antennas which are located at the remote site; second, the high bandwidth low-latency optical transport network which connect the RRHs and BBU pool; and third, the BBU composed of high-performance programmable processors and real-time virtualization technology.
16
China Mobile Research Institute
Virtual BS Pool
L1/L2/L3/O&M
L1/L2/L3/O&M
L1/L2/L3/O&M
Fiber
RRH
RRH
RRH
RRH
RRH
RRH
RRH
Fig. 3-2 C-RAN Architecture 1: Fully Centralized Solution Virtual BS Pool
L2/L3/O&M
L2/L3/O&M
L2/L3/O&M
Fiber or Microwave
RRH/L1
RRH/L1
RRH/L1
RRH/L1
RRH/L1
RRH/L1
RRH/L1
Fig. 3-3 C-RAN Architecture 2: Partial Centralized Solution The „fully centralized‟ C-RAN architecture, as shown in figure 3-2, has the advantages of easy upgrading and network capacity expansion; it also has better capability for supporting multistandard operation, maximum resource sharing, and it‟s more convenient towards support of multi-cell collaborative signal processing. Its major disadvantage is the high bandwidth requirement between the BBU and to carry the baseband I/Q signal. In the extreme case, a TDLTE 8 antenna with 20MHz bandwidth will need a 10Gpbs transmission rate.
China Mobile Research Institute
17
The other type of C-RAN is to centralize partial BBU functions which include collaborative function, L2 and L3 scheduling, and wireless resource allocation. As shown in Figure 3-3, the feature of this architecture is “small centralization” with partial BBU functions centralized into one central point which is connected with the remained remote BBU via dark fiber or PTN networks. With such architecture, the central point can schedule the wireless resource in each cell on a global level and even realize the joint transmission or joint reception on PHY layer to improve cell edge performance. The data bandwidth between the central point and remote sites is small, which minimizes the change on existing transport networks. The major disadvantage of this architecture is that it still requires remote equipment rooms. One-body type base station is not preferred from the perspective of system management and future upgrade. In addition, the delay on information exchange can have an impact on the system performance improvement.
With either one of these C-RAN architectures, mobile operators can quickly deploy and make upgrades to their network. The operator only needs to install new RRHs and connect them to the BBU pool to expand the network coverage or split the cell to improve capacity. If the network load grows, the operator only needs to upgrade the BBU pool‟s HW to accommodate the increased processing capacity. Moreover, the „fully centralized solution‟, in combination with open platform and general purpose processors, will provide an easy way to develop and deploy software defined radio (SDR) which enables upgrading of air interface standards by software only, and makes it easier to upgrade RAN and support multi-standard operation. Different from traditional distributed BS architecture, C-RAN breaks up the static relationship between RRHs and BBUs. Each RRH does not belong to any specific physical BBU. The radio signals from /to a particular RRH can be processed by a virtual BS, which is part of the processing capacity allocated from the physical BBU pool by the real-time virtualization technology. The adoption of virtualization technology will maximize the flexibility in the C-RAN system. Both solutions described above are under development and evaluation. They could be properly deployed in different networks depending on the situation of the network. The following discussion will focus on the „Fully Centralized Solution‟.
3.1 Advantages of C-RAN The benefits of the C-RAN architecture are listed as follows:
Energy Efficient/Green Infrastructure C-RAN is an eco-friendly infrastructure. Firstly, with centralized processing of the C-RAN architecture, the number of BS sites can be reduced several folds. Thus the air conditioning and other site support equipment‟s power consumption can be largely reduced. Secondly, the distance from the RRHs to the UEs can be decreased since the cooperative radio technology can reduce the interference among RRHs and allow a higher density of RRHs.
18
China Mobile Research Institute
Smaller cells with lower transmission power can be deployed while the network coverage quality is not affected. The energy used for signal transmission will be reduced, which is especially helpful for the reduction of power consumption in the RAN and extend the UE battery stand-by time. Lastly, because the BBU pool is a shared resource among a large number of virtual BS, it means a much higher utilization rate of processing resources and lower power consumption can be achieved. When a virtual BS is idle at night and most of the processing power is not needed, they can be selectively turned off (or be taken to a lower power state) without affecting the 7x24 service commitment.
Cost-saving on CAPEX &OPEX Because the BBUs and site support equipment are aggregated in a few big rooms, it is much easier for centralized management and operation, saving a lot of the O&M cost associated with the large number of BS sites in a traditional RAN network. Secondly, although the number of RRHs may not be reduced in a C-RAN architecture its functionality is simpler, size and power consumption are both reduced and they can sit on poles with minimum site support and management. The RRH only requires the installation of the auxiliary antenna feeder systems, enabling operators to speed up the network construction to gain a firstmover advantage. Thus, operators can get large cost saving on site rental and O&M.
Capacity Improvement In C-RAN, virtual BS‟s can work together in a large physical BBU pool and they can easily share the signaling, traffic data and channel state information (CSI) of active UE‟s in the system. It is much easier to implement joint processing & scheduling to mitigate inter-cell interference (ICI) and improve spectral efficiency. For example, cooperative multi-point processing technology (CoMP in LTE-Advanced), can easily be implemented under the CRAN infrastructure.
Adaptability to Non-uniform Traffic C-RAN is also suitable for non-uniformly distributed traffic due to the load-balancing capability in the distributed BBU pool. Though the serving RRH changes dynamically according to the movement of UEs, the serving BBU is still in the same BBU pool. As the coverage of a BBU pool is larger than the traditional BS, non-uniformly distributed traffic generated from UEs can be distributed in a virtual BS which sits in the same BBU pool.
Smart Internet Traffic Offload Through enabling the smart breakout technology in C-RAN, the growing internet traffic from smart phones and other portable devices, can be offloaded from the core network of operators. The benefits are as follows: reduced back-haul traffic and cost; reduced core network traffic and gateway upgrade cost; reduced latency to the users; differentiating service delivery quality for various applications. The service overlapping the core network also supplies a better experience to users.
China Mobile Research Institute
19
3.2 Technical Challenges of C-RAN The centralized C-RAN brings lots of benefits in cost, capacity and flexibility over traditional RAN, however, it also has some technical challenges that must be solved before deployment by mobile operators.
Radio over Low Cost Optical Network In C-RAN architecture 1, the optical fiber between BBU pool and RRHs has to carry a large amount of baseband sampling data in real time. Due to the wideband requirement of LTE/LTE-A system and multi-antenna technology, the bandwidth of optical transport link to transmit multiple RRHs‟ baseband sampling data is 10 gigabit level with strict requirements of transportation latency and latency jitter.
Advanced Cooperative Transmission/Reception Joint processing is the key to achieve higher system spectrum efficiency. To mitigate interference of the cellular system, multi-point processing algorithms that can make use of special channel information and harness the cooperation among multiple antennas at different physical sites should be developed. Joint scheduling of radio resources is also necessary to reduce interference and increase capacity. To support the above Cooperative Multi-Point Joint processing algorithms, both end-user data and UL/DL channel information needs to be shared among virtual BSs. The interface between virtual BSs to carry this information should support high bandwidth and low latency to ensure real time cooperative processing. The information exchanged in this interface includes one or more of the following types: end-user data package, UE channel feedback information, and virtual BS‟s scheduling information. Therefore, the design of this interface must meet the realtime joint processing requirement with low backhaul transportation delay and overhead.
Baseband Pool Interconnection The C-RAN architecture centralizes a large number of BBUs within one physical location, thus its security is crucial to the whole network. To achieve high reliability in case of unit failure, in order to recover from error, and to allow flexible resource allocation of BBU, there must be a high bandwidth, low latency, low cost switch network with flexible, extensible topology that interconnects the BBUs in the pool. Through this switch network, the digital baseband signal from any RRH can be routed to any BBU in the pool for processing. Thus, any individual BBU failure won‟t affect the functionality of the system.
Base Station Virtualization Technology After the baseband processing units have been put in a centralized pool, it is essential to design virtualization technologies to distribute/group the processing units into virtual BS entities. The major challenges of virtualization are: real-time processing algorithm implementation, virtualization of the baseband processing pool, and dynamic processing capacity allocation to deal with the dynamic cell load in system.
20
China Mobile Research Institute
Service on Edge Unlike service in a data center, distributing services on the edge of the RAN has its unique challenges. In the following research framework part, we try to summarize these challenges into the following three categories: services on the edge‟s integration with the RAN, intelligence of DSN, and the deployment and management of distributed service.
China Mobile Research Institute
21
22
China Mobile Research Institute
4 C-RAN deployment scenarios The C-RAN deployment scenarios differ at different stages of 2G/3G/4G constructions. For GSM network, the need for C-RAN deployment is limited and thus the main strategy is to maintain the network reliability and stability. For TD-SCDMA, it has already provided country wide coverage in most of the cities. Future network expansion will mainly focus on rural area and the remaining few cities. The main construction strategy is to improve hot-spot and weak-spot coverage.For 4G, CMCC just finished the large-scale field trials in the past few years and only a few cities have the TD-LTE coverage. It can be foreseen that in the coming few years TD-LTE deployment will be our main target. This chapter will describe different C-RAN deployment scenarios for 3G and 4G, respectively.
4.1 TD-SCDMA C-RAN deployment A typical TD-SCDMA site has 3 sectors with 3 carriers per sector. The mainstream equipments support three RRU cascade. The utilization efficiency of TD-SCDMA carriers is low due to the severe network tidal effect. At the same time, there is still existing much area with weak coverage in the current TD-SCDMA networks. On the other hand, as the number of subscribers is increasing fast, the high-density area will require more sites to absorb the traffic, which in turn increase the difficulty of site selection. In addition, there are also some other special area such as expressway, railway, street and riverway in which the handover success rate is relatively low due to a large number of fast handover. For these scenarios, the centralization of BBU deployment can help to address the above-mentioned issues, i.e. to deal with tidal effect effectively, to improve the utilization efficiency of carriers, to reduce the difficulty of site selection and to improve handover success rate. .
4.1.1 Scenario 1: Capacity and coverage improvement using Pico-RRU for weak-spot and hot-spots In this scenarios, C-RAN is used to provide hot-spot coverage or improve weak coverage in some area. The new BBUs can be installed in macro site room and connected with remote RRU via fiber.
China Mobile Research Institute
23
)
))
)
)
))
))
))
)
)
))
)
))
)
))
)
))
BBU Pool
)
))
)
))
Fig. 4-1: Capacity and coverage improvement using Pico-RRU for weak-spot and hot-spots With the increased difficulty on site acquisition and pressure on forced removal of existing equipment rooms by proprietors, many area in high density urban cities are of weak coverage. To address this issue, installation of BBU pool in the center equipment room and the small RRUs will take more important roles . It is recommended that the centralization equipment room should be owned by operators themselves to avoid impact by possible site relocation in the future. At the same time, the so-called “multi-RRU co-cell” technology can be used to improved the network quality. Generally there can be a vertical three-layer network deployment mode: basic coverage by macro base stations, capacity and coverage supplement by micro RRUs in the outdoor and traffic asorbion via indoor solution.
The characteristic of this scenario includes two key parts: BBU pool centralized in the existing macro-site and 2-antennas Pico-RRU with low transmit power on the remote site. The scale of centralized carriers is decided by area characteristics such as the traffic volumn. In addition, the fiber from the last-mile pipeline can be utilized or it can be installed hanging over the building.
4.1.2 Scenario 2: Area with tidal effect The tidal effect in such area is evident. Examples include campus city, industrial parks, dormitory area, commercial districts, residential area and so on.
24
China Mobile Research Institute
BTS
BTS
BTS
Residential area
BTS
BTS BTS
BTS
BBU pool
BTS
Industrial area
Fig. 4-2 Tidal effect in residential and industrial area Making use of construction of new area or re-construction of old area, the transport facilities can be deployed to enable BBU centralization with dark fiber. Deployment of centralized BBU pool can deal with tidal effect. In addition, the usage of carrier live migration can help to save the overall number of carriers and improve the system performance-power ratio by dynamic resource allocation.
4.1.3 Scenario 3: Region with massive fast handover Such scenarios include the area such as the highway, railway, streets and riverway. For the users moving fast through the regions, it is easy for a call to drop due to delay on mobile signal measurement or fast handover. To address this, some technologies with optimization on fast handover such as multi-carrier co-cell can be used in the centralized BBU pool.
BBU pool
RRU
Fig. 4-3 Frequent handover in railway coverage area The scale of BBU pool is quite dependent on the available resource of fiber pipeline. The remote RRU can be installed on the lampposts with power supply using either DC remote supply or local supply. The BBU pool can be installed in the outdoor cabinet or simple equipment room in an embellished way.
China Mobile Research Institute
25
4.2 TD-LTE C-RAN deployment The construction of TD-LTE network is our current focus. From previous large-scale field trials, we have accumulated a lot of experience and solved many key problems. However there are still some issues left. On one hand, due to the co-site deployment of TD-LTE with 2G/3G systems it is found that the TD-LTE antennas are usually either too high or too low and intercell distance is very close. All these lead to severe interference by large overlap among cells and as a result the system performance deteriorates a lot. Since LTE is more sensitive to the interference than 2G/3G. Some 2G/3G sites are not suitable for TD-LTE deployment. This, in other words, means that new sites are needed. In fact, it is estimated that around 30% and 5%~10% new sites are needed for TD-LTE D band and F band deployment respectively. Doubtless, the addition of new sites adds the difficulty on site selection. C-RAN is deemed as an efficient way to help network construction with the advantages of reducing interference, saving cost, speeding up site construction and lowing down difficulty in site selection.
4.2.1 Scenario 1: HetNet with C-RAN Similar to 3G, the need for improvement of weak-spot and hot-spot coverage still exists in TDLTE. There are three reasons for this. 1. The wall penetration ability of D-band is worse than F band. As a result, in the dense urban, there will be more area with weak coverage caused by building shelter. 2. In TD-LTE data rate is one of the most important measurement to user experience. If we use the minimum data rate to define the cell edge, then in order to provide high-quality service the cell size will be smaller than 2G/3G networks. 3. In some urban area, there exist super hot spots which is of extremely high data traffic. To absorb the traffic, multiple small cells can be deployed with seamless coverage. The C-RAN deployment method in TD-LTE is similar to in 2G/3G networks. Considering the relative abundance of the frequency resource at the initial stage, it is preferred that the small cells use different frequency bands from the macro cells. After the introduction of the Carrier Aggregation technology, it will be easy to implement the C/U split to further improve the overall capacity. Reusing the same frequency bands between the macro and small cells can be considered when the need for higher capacity becomes urgent. No mater what kind of frequency scheme is used, the deployment of C-RAN can facilitate the cooperation between macro and small cells. At the same time, due to people‟s more attention to the environment, the concern on radio radiation has become the first reason that prohibits the deployment of wireless equipments. Because of this recently in large cities such as BeiJing and ShangHai, we encountered many obstacles when upgrading 2G/3G sites to 4G. Even more, some sites under construction were forced to be removed because of residents‟ complaint during site construction. On the other hand, some 2G/3G sites do not have sufficient reserved space to accommodate TD-LTE .
26
China Mobile Research Institute
Installation of RRU and antennas needs reconstruction on the rooftop in original sites, which instead, makes the civil work more difficult. As the result, in the predictable future, there will appear large area of blind or weak coverage in the urban cities. To address this, small cells are needed to provide continuous seamless coverage , which imposes new requirements on wireless equipments, including: 1. Smaller transmission power and miniaturization for RRU as well as smaller size for antennas. RRU and antennas with smaller size can reduce the public concern on the radio radiation. And RRUs of low power consumption will match the requirements of the environment-friendly policies from government and save the time for installation permission. The current transmission power is 5w per channel for an outdoor RRU. It is estimated through link budget calculation that in case of typical inter-cell distance of 100 meters, the needed transmission power can be smaller. 2. Collaborative radio support with BBU pool. Some technologies, such as multi-RRU co-cell and generalized MIMO can help to reduce the interference and thus to improve system performance. In this way, the network will consist of at least two layers. One is the macro cell for basic coverage, and the other is the small cell to absorb the hot-spot traffic. It is estimated that the ratio of macro to micro RRUs is between 1:3 and 1:6.
4.2.2 Scenario 2: Combination with the construction of integrated service access zone Integrated Service Access Zone (ISAZ) is a new method to plan and construct the transport infrastructure with target at household wideband wireline customers, group wired customers as well as BS access needs. The idea of ISAZ is to divide a city into several smaller zones with each of area of 3~5 square kilometers. For each zone the transport resource will be planned overally and comprehensively. Some good examples of ISAZ include university campus, hi-tech science parks, residential area, exhibition parks and industrial parks. According to our current planning, an ISAZ usually consist of 1~2 transport access ring ( may have more rings in some big cities) with each ring of 6~8 mobile macro equipment rooms. In some cases the maximum number of wireless macro equipment rooms can be 12. Considering that current macro base stations typically have 3 sectors with each sector of one 20MHz TD-LTE carrier, then the total number of TD-LTE carriers is between 24 and 36 in one access ring. It could becomer higher to 50~70 in the future when sectors are upgraded with two carriers. In the cities to be deployed with TD-LTE, combination with ISAZ is a promising scenario for CRAN deployment. The basic idea is to make full use of the relatively rich transport resources such as fiber, duct and pipeline. Then the BBUs within the same ISAZ can be centralized to the aggregation site (which can be possibily the aggregation office in the transport network) with remote site deployed with RRU. Dark fiber is now widely used in our C-RAN trials due to its maturity. With CPRI compression and bi-direction single fiber technologies, one fiber core can support one 20MHz TD-LTE carrier with 8 antenna. We then therefore suggested to reserve at
China Mobile Research Institute
27
least 48 fiber cores for C-RAN centralization in the ISAZs with sufficient fiber, taking into account the potential centralization scale. In the future the usage of fiber can be further reduced with the introduction of WDM equipments. After the centralization of BBU, the collaborative radio technologies (e.g. JT/JR) can be further adopted in the BBU pool to enhance the system performance. There are three construction methods under this scenario. A. Scenario a: If the TD-LTE equipements can‟t be installed in existing 2G/3G sites, then the new BBUs can be centralized into aggregation office of ISAZ and a new remote site with outdoor stand-by power supply is necessary for RRU installation. B. Scenario b: If the TD-LTE equipments can be installed in existing 2G/3Gsites, then the new BBUs can be centralized into aggregation office of ISAZ and the RRU can be installed in the existing 2G/3G remote sites. Stand-by power resource for RRU is also required. C. Scenario c: If the TD-SCDMA BBU can be upgraded to TD-LTE, then it is not necessary to deploy C-RAN. However, if the network suffers from severe interference from neiboring cells, then C-RAN centralization can be used for introduction of collaborative radio technologies to address the issue.
4.2.3 Scenario 3: Comibination of the two scenarios above. There is no conflict between the two above-mentioned scenarios, i.e. HetNet and ISAZ . In fact, in the highly dense urban with ISAZ planning, there still exist many weak-spots and hot-spots. For this scenario, the construction can be expanded as follows.
Fig. 4-4 Combination of HetNet and ISAZ The BBUs are centralized into ISAZ aggregation rooms. When the fiber resource is limited, WDM can be introduced to connect existing wireless equipment rooms into a ring. If WDM equipments can be deployed outdoor, it can also act as an aggregation point to connect together a few remote sites close to each other. The macro and micro cell BBUs are collocated
28
China Mobile Research Institute
in the same BBU pool which enables complex and fast collaborative radio technology to improve wireless performance. With WDM solution, the typical length of a WDM ring is less than 20km.
China Mobile Research Institute
29
5 Technology Trends and Feasibility Analysis In order to solve the technical challenges of C-RAN architecture, based on current technical conditions and future development trends, we suggest to do further research in the following areas. The purpose is to solve the low cost high bandwidth wireless signal transmission problem based on an optical network, dynamic resource allocation and collaborative radio technology. It also comprehends the large scale BBU pool and associated interconnection problem, virtualized BS based on open platforms and distributed service network solutions. The following is a detailed analysis and discussion of these challenges.
5.1 Wireless Signal Transmission on Optical Network The C-RAN architecture, which consists of the distributed RRH and BBU, means that need to transport untreated wireless signals between BBU and RRH. The BBU-RRH connectivity requirements pose challenges to the optical transmission speed and capacity. Usually, optical fiber transmission must be used to carry the BBU-RRH signal to meet the strict bandwidth and delay requirements.
BBU-RRH Bandwidth Requirement Air interface is upgrading rapidly, new technologies like multiple antenna technology (2 ~ 8 antenna in every sector), wide bandwidth (10 MHz ~ 20 MHz every carrier) has been widely adopted in LTE/LTE-A, thus the bandwidth of CPRI/Ir/OBRI (Open BBU-RRH Interface) link bandwidth is much higher than the 2G and 3G era. In general, the system bandwidth, the MIMO antenna configuration and the RRH concatenation levels are the main factors which have an impact on the OBRI bandwidth requirement. For example, the bandwidth for 200 kHz GSM systems with 2Tx/2Rx antennas and 4xsampling rate is up to 25.6Mbps. The bandwidth for 1.6MHz TD-SCDMA systems with 8Tx/8Rx antennas and 4 times sampling rate is up to 330Mbps. The transmission of this level of bandwidth on fiber link is matured and economic. However, with the introducing of multi-hop RRH and high orders MIMO supporting 8Tx/8Rx antenna configuration, the wireless baseband signal bandwidth between BBU-RRH would rise to dozens of Gbps. Therefore, exploring different transport schemes for the BBU-RRH wireless baseband signal is very important for C-RAN.
Transportation Latency, Jitter and Measurement Requirements There are also strict requirements in terms of latency, jitter and measurement. In CPRI/Ir/OBRI transmission latency, due to the strict requirements of LTE/LTE-A physical layer delay processing
also
improve
the
baseband
wireless
signal
transmission
delay
jitter
and
requirements indirectly. Not including the transmission medium between the round-trip time (i.e., regardless of delays caused by the cable length), for the user plane data (IQ data) on the CPRI/Ir/OBRI links, the overall link round-trip delay may not exceed 5μs. The OBRI interface requires periodic measurement of each link or multi-hop cable length. In terms of calibration, the accuracy of round trip latency of each link or hop should satisfy ±16.276ns [4].
30
China Mobile Research Institute
System Reliability For the reliability of the system, because the traditional optical transmission networks (SDH/PTN) in the access network links provide reliable loop protection, automatic replace and fiber optic link management function, C-RAN architecture in the access network must also provide comparative reliability and manageability. In traditional RAN architecture, each BBU on the access ring usually has access to the corresponding transmission equipment of the center transmission machine room through SDH/PTN. Through the SDH/PTN ring routing and protection function, the system can quickly switch to the safe routing mode when any point on this loop experiences optical fiber failure, ensuring that business is not interrupted. Under the C-RAN architecture, it also should offer a similar optical fiber ring network protection function. Centralized BBU should support more than 10~1000 base station sites, and then the optical fiber connected OBRI link between distributed RRH and centralized BBU is long. If only point-2point optical fiber transmission occurred between each distributed RRH and centralized BBU, then any fault on the optical fiber link will lead to the corresponding RRH loosing service. In order to ensure the normal operation of the whole system under the condition of any single point of failure in the optical fiber, the CPRI/Ir/OBRI link connecting the BBU-RRH should use fiber ring network protection technology, using the main/minor optical fiber of different channels to realize CPRI/Ir/OBRI link real-time backup.
Operation and Management At the same time, under the traditional RAN architecture, the transmission network which consists of SDH/PTN also provides the unified optical fiber network management ability for the access ring. This includes unified management of the access ring fiber optic link of the entire network, supervisory control of the access ring optical fiber breakdown, etc. BBU-RRH wireless signal transport directly on the access ring, whose CPRI/Ir/OBRI interface should also, provides similar management ability and fit into unified optical fiber network management.
Cost Requirements Finally, in terms of cost, the high speed optical module necessary for the CPRI/Ir/OBRI optical interface will be amongst the important factors affecting the C-RAN economic structure. Compared to traditional architecture, the wireless signal transmission data rate on C-RAN is more than 100-200 times higher than the bearer service data rate after demodulation. Building the fiber transportation network in developed city is very hard. This is less of an issue for operators that already deploy optical fiber and particularly for operators own their own optical network. Although the cost of the optical fiber employing CPRI/Ir/OBRI for high speed wireless signal transmission doesn't need to increase, the high speed optic module or optical transmission equipment costs must compare to traditional SDH/PTN transmission equipment in order to make C-RAN architecture more attractive on the CAPEX and OPEX fronts .Therefore, how to achieve a low cost, high bandwidth and low latency wireless signal optical fiber transmission will become a key challenge for realization of the future LTE and LTE network deployment by C-RAN.
China Mobile Research Institute
31
For the above problems and corresponding technical progress trend, we will analyze and put forward ideas for solving these problems.
5.1.1 Data Compression Techniques of CPRI/Ir/OBR Link In view of the above LTE/LTE-A BBU-RRH wireless signal transmission bandwidth problems, several data compression techniques that can reduce the burden on the OBRI interface are being investigated to deal with the inevitable bandwidth issue, including time domain schemes (e.g. reducing signal sampling, non-linear quantization, and IQ data compression) as well as frequency domain schemes (e.g. sub-carrier compression). For LTE system with 20MHz bandwidth, the BBU uses 2048 FFT / IFFT but the effective number of subcarriers is only 1,200, so if the FFT / IFFT is implemented in the RRH, then the Ir interface between BBU and the RRH only has to transmit effective data subcarriers, such that the Ir interface load can be reduced about 40%, However,
frequency domain
compression leads to an increase in IQ mapping complexity, which would increase the interface logic design and processing complexity. Meanwhile, the RRH needs to process parts of the RACH, Therefore, RRH cannot treat different RACH configurations transparently, instead RRH needs to process RACH based on configuration. Since there are hundreds of different configurations, each has to be controlled by different timing algorithms in the RRH, which could greatly increase the complexity of system design. Therefore, considering the implementation complexity and cost, such frequency domain compression is not feasible at the moment. DAGC time-domain based compression technology is a method used for IQ compression. The basic principle of DAGC is to select the average power reference based on the best baseband demodulation range, normalize the power of each symbol, and reduce the signal dynamic range. DAGC compression will adversely affect system performance. The receiver dynamic range of the uplink will be reduced, which leads to deterioration of the signal to noise ratio.
At the same time, the EVM indicators will worsen on the downlink. With
increased compression ratio, the system performance will deteriorate even more. Currently, we still need to investigate the impacts caused by different compression schemes. Table 2 lists the advantages and disadvantages of various compression schemes. As indicated, there is no ideal OBRI link data compression scheme. More studies in this area are required. Table 2. Comparison of Pros and Cons for Various Data Compression Techniques
Bandwidth Compression Schemes Reducing signal sampling
32
Pros Low complexity;
Cons Severe performance loss.
Efficient compression to 66.7%; Less impacts on protocols.
China Mobile Research Institute
Non-linear quantization
Improve the QSNR;
Some impacts on the OBRI interface
Mature algorithms available, e.g. A law
complexity.
and U law; High compression efficiency to 53%.
IQ data Compression
Potential high compression efficiency;
High complexity;
Only need extra decompression and
Difficult to set up a relativity model;
compression modules.
Real-time and compression distortion issues; No mature algorithm available.
High compression efficiency to 40%
Increase the system complexity;
~58%;
Extra processing ability on optical chips
Easy to be performed in downlink.
and the thermal design;
Sub-carrier Compression
High device cost; Difficulty for maintenance; RACH processing is a big challenge; More storage, larger FPGA processing capacity.
5.1.2 Transmission delay and jitter of CPRI/Ir/OBRI link As mentioned previously, CPRI/Ir/OBRI link have strict demands on transmission delay, jitter and measurement. However, because the link round trip delay requirements (5 us) of the user plane data (IQ data) in CPRI/Ir/OBRI link do not include the transmission medium round-trip time (i.e. delay in optical transmission), this requirement can be satisfied by the existing technical conditions. At the same time, because CPRI/Ir/OBRI optical fiber routing generally does not change with time and delay jitter caused by transmission is relatively small, it is easy to meet the corresponding requirements. On the other hand, because LTE/LTE-A has strict requirements about physical layer treatment delay, CPRI/Ir/OBRI total transmission delay on the link should not exceed a certain level. The physical layer HARQ process places the highest demand on processing delay. HARQ is an important technology to improve the performance of the physical layer, its essence is testing the physical layer on the receiving end of a sub-frame for correct or incorrect transmission, and rapid feedback ACK/NACK to the launching end physical layer, then let launching physical layer to make the decision whether or not to send again. If sent again, the receiver does combined processing for multi-launching signal in the physical layer, and then provides feedback to the upper protocol after demodulation success. According to the LTE/LTE-A standard, the ACK/NACK HARQ on uplink and downlink process should be finished in 3 ms after receiving the signals in the shortest case, which requires that sub-frame processing delay in the physical layer should be generally less than 1 ms. Because the physical layer processing itself takes 800-900 us, then CPRI/Ir/OBRI optical transmission delay may be 100-200 us at the most. According to the light speed(200,000 kilometers per hour) estimated in the fiber, CPRI/Ir/OBRI interface maximum transmission distance under the C-RAN framework is limited from 20 km to 40 km. Specific value is related to delay margin the physical layer treatment itself.
5.1.3 Optical Transmission Technology Progress and Cost Reduction China Mobile Research Institute
33
As mentioned above, BBU-RRH wireless signal connection supporting LTE and LTE-Advance creates new challenges to optical transmission network rates and cost. The rapid development of the optical transmission technology provides more economic solutions to solve the problem. A single fiber capacity of current commercial WDM system can be up to 3.2
T.10
Gpbs
optical
transmission
technology
applies
generally
and
become
fundamental , 40 G system is mature and gradually being commercialized, 100 G technology is still not mature and costs too much, there is still 2-3 years telecommunication
until the
commercial level, but along with coherent technical breakthroughs,
promoting of standardization has already become a now advantage. 10GE standardization and industrialization will greatly improve the relevant market capacity of the optical transmission module, which will help to reduce the cost of 10 Gbps optical modules. 40GE technology is still in the research process. On the other hand, at the access network level, 1.25 G,2.5 G EPON is already widely used in solving FTTX access, 10G PON technology can be commercial in one or two years, the future PON technological development have several directions like WDM-PON, Hybrid PON and 40G PON. Similar to what the Moore's Law is doing in the transformation of the semiconductor industry, the field of optical communication has a similar trend: Every year, the speed of optical transmission increases while the cost of the said module declines. Transceiver modules that are capable of supporting multi-wavelength WDM have emerged in the market place. Since commercial LTE deployment has just begun, we can safely predict that it will take about 5 years before the commercial LTE-A multi-carrier system deployment is needed. By then, if the optical module advancement and cost reduction has reached an acceptable level, then the RRH-BBU bottleneck will be effectively removed. Figure 5-1 shows the 2.5G SFP and 10G SFP / XFP / XENPAK optical modules pricing trends. We can deduce that optical modules pricing has dropped by 66% to 77% in nearly 3 years, and the trend will continue in the coming years, further reducing the cost of optical transmission network. If this price trend continues, it would greatly help to reduce CAPEX
10000
3000 2500 2000 1500 ↓66.7%
1000
↓54.2% 500 0 Aug-07
↓62.2%
Feb-08 10Km
Aug-08 40Km
Feb-09
Aug-09
80Km
Price history of 10G modules (RMB).
Price history of 2.5G modules (RMB).
of a C-RAN network.
9000 8000 ↓35.2%
7000 6000 5000
4000
↓61.5%
3000
↓60%
2000
1000 0 Aug-07
Feb-08 550m
Aug-08 10Km
Feb-09
Aug-09 40Km
Fig. 5-1 Price history of Commercial 2.5G/10G Optical Modules 34
China Mobile Research Institute
5.1.4 BBU-RRH Optical Fiber Network Protection Although BBU-RRH direct transmission under C-RAN framework does not provide a ring network protection function like traditional SDH/PTN, the CPRI/Ir/OBRI interface rate standards provide a similar ring network protection function, and are supported by manufacturers. At the same time, in order to avoid having every RRH fully occupy two optical fibers on a physically routed pair the RRH‟s can be connected to each in a cascaded manner according to the CPRI/Ir/OBRI interface specification. This permits two different routing trunk cables to form a ring and be connected to the same BBU, as shown in Figure 5-2. As long as the CPRI/Ir/OBRI interface rate is high enough, the BBU-RRH ring network protection technology can save the use of many optical fibers and ensure a short round trip delay. Taking a TD-SCDMA system for example, a 6.144 Gpbs CPRI/Ir/OBRI link can support 15 TD-SCDMA carriers of 8-antenna RRH and a typical TD-SCDMA macro station with 3 sectors, 5/5/5 configuration at most. The IQ data of a RRH with three sectors connected to the same BBU machine through two different physical routing backbone optical cables. When a trunk cable fails, three RRHs will connect to the BBU through another trunk cable under less than 40ms protection rotated time to guarantee that all business does not interrupt. For lower-rate GSM system, it is even simpler to connect six or more RRHs through such a CPRI/Ir/OBRI annular link and achieve the same functions. However, according to LTE/LTE-A system with higher wireless signal transmission rate, it is necessary to introduce WDM technology to realize a similar loop protection function.
Radio remote head
Trun kc
able
Optical switching box
2
Transmission ring Trunk c a
ble 1
Central apparatus room
Fig. 5-2 RRH Ring Protection Loop 5.1.5 Current Deployment Solutions In order to meet the high bandwidth transmission between RRH and BBU, operators can use different solutions based on their current transmission network resources. In China Mobile, the current backhaul is mainly an optical transport network with three layers of transmission network: the core transmission layer, the convergence transmission layer and the access transmission layer. All the layers are using ring topology to provide fail safe protection. The optical resources of different layers are similar to the following: at the core
China Mobile Research Institute
35
transmission layer, each optical route has 144 to 576 fibers; at the convergence transmission layer, each route has 96-144 fibers; while at the access transmission layer, each route has 24-48 fibers. If the Baseband pool is located in the transmission convergence equipment room, the optical fiber resource to and from the equipment room determines the coverage of the baseband pool. According to the resourcing of the optical transmission network, especially the fiber resource in the access transmission network, there are four different solutions to carry CPRI/Ir/OBRI over it: 1. Dark fiber; 2. WDM/OTN; 3. Unified Fixed and Mobile access like UniPON; 4. Passive WDM. These solutions have different advantages and disadvantages, and they are each suitable for different deployment scenarios. From the trials conducted, for a BBU pool with less than 10 macro BSs, it is preferred to use a dark fiber solution while other solutions still need more field tests and verification, because they may introduce new transmission devices and associated O&M issues. The first solution is Dark fiber. It is suitable when there is plenty of fiber resource. It is easy to deploy if there are a lot spare fiber resources. The benefits of this solution are: fast deployment and low cost because no additional optical transport network equipment is needed. The concerns of this solution are: it consumes significant fiber resource, thus the network extensibility will be a challenge; new protection mechanisms are required in case of fiber failure; and it is hard to implement O&M, therefore it will introduce some difficulties for optical network O&M. However, there are feasible solutions to address such challenges. For fiber resources, if there is already a channel route available, it is fairly inexpensive to add new fiber cables or upgrade existing fibers. To address fiber failure protection, there are CPRI/Ir/OBRI compliant products available now that have the 1+1 backup or ring topology protection features. If deployed with physical ring topology that provides alternative fiber route, it will be able to provide similar recoverability capability as SDH/PTN. For the O&M of the fiber in the access ring, we are considering introducing new O&M capabilities
in
the
CPRI/Ir/OBRI
standard
to
satisfy
the
fiber
transport
network
management requirement. The second solution is WDM/OTN solution. It is suitable for Macro cellular base station systems when there is limited fiber resource, especially where the fiber resource in the access ring is very limited, or adding new fiber in existing route is too difficult or cost is too high. By upgrading the optical access transmission network to WDM/OTN, the bandwidth of transporting CPRI/Ir/OBRI interface on BBU-RRH link is largely improved. Through transmitting as many as 40 or even 80 wavelength with 10Gpbs in one fiber, it can support a large number of cascading RRH on one pair of optical fiber. This technology can reduce the demand of dark fiber, however, upgrading existing access ring into WDM/OTN transmission network means higher costs. On the other hand, because the access transport network is usually within a few tens of kilometers, the WDM/OTN equipment can be much cheaper than those used in long distant backbone networks. OTN (Optical Transport
36
China Mobile Research Institute
Network) is another kind of WDM-based technology. ONT claims the advantages of openness, good interoperability and scalability as well as powerful O&M functions. The main issue for OTN solution lies on the high cost. The third solution is based on CWDM technology. It combines the fixed broadband and mobile access network transmission at the same time for indoor coverage with passive optical technology, thus named as „Unified PON‟. It can provide both PON services and CPRI/Ir/OBRI transmission on the same fiber [5]. In this solution, an optical fiber can support as many as 14 different wavelengths. In the UniPON standard, the uplink and downlink channel are transmitted on two difference wavelengths, thus other free wavelengths can be used for CPRI/Ir/OBRI data transmission between the BBU and RRH. Because of sharing the optical fiber resources, it can reduce the overall cost. It is suitable for C-RAN centralized baseband pool deployment of indoor coverage.
5.1.6 Other consideration Based on the above analysis, „fully centralized‟ C-RAN architecture requires a high bandwidth, low latency, high reliability and low cost optical solution to transmit high speed baseband signal between BBU and RRH. It‟s promising to find feasible solutions emerging in the near future. However, there are still many challenges in the current solutions. For example, current data compression schemes fail to satisfy OBRI transmission in the LTE-A phase. The rapid development of high-speed optical modules and the associated cost reduction is heading in the right direction but we still need a breakthrough in optical devices. Failure protection schemes for BBU-RRH connection are able to provide similar functions to SDH/PTN in case of fiber cut, but we still need to find solutions for unified O&M with traditional transmission networks. UniPON based on passive WDM technology is a promising solution for certain deployment scenarios but it must be designed to be competitive in cost. In conclusion, we have various directions to solve the high-speed baseband signal transmission requirement of C-RAN but we still need to explore new technology or a combination of existing technology to find a more economical and effective solution. Considering the technical challenges as well as the limitation in current optical network resources, it is clear that C-RAN can be widely applied in a short time frame. Instead, a stepped plan should be used to gradually construct the centralized network: first, centralized deployment can be applied in some green field or replacement of old network in a small scale. Dark fiber can be used as the BBU-RRH transmission solution. One access ring that connects 8~12 macro sites can be centralized together, with a maximum ring range of 40km. In the future, a larger number of macro BS in various deployment scenarios can be further tested.
5.1.7 Technology advancement
China Mobile Research Institute
37
In this and the subsequent sections in the White Paper, the transmission between the BBU and the RRU in C-RAN is defined as fronthaul transmission (compared with traditional backhaul transmission between the BBU and the core network).
The fronthaul transmission technology is of decisive significance to C-RAN large-scale deployment. As more operators are paying importance to C-RAN, more resources are committed to the issue. It is happy to see that many breakthroughs have been achieved recently.
CPRI compression. With the maturity of CPRI compression, several vendors have commercially realized 2:1 compression with lossless performance. It can help to save half usage of fiber consumption. In addition, the Single Fiber Bi-direction (SFBD) technology allows simultaneous UL and DL transmission on a single fiber, which further halves fiber consumption. Combining CPRI compression and SFBD can save the fiber consumption by 3 folds. CMCC has successfully verified the two technologies in C-RAN TD-LTE field trials. More details and information can be found in Chapter 6.
WDM solution. Since WDM technology is sufficiently mature, vendors can develop WDM equipments tailored to fronthaul transmisstion within a short period of time. Currently a few operators have adopted this solution to enable the large-scale C-RAN deployment. Some commercial products can support as many as 60 2.5Gbps CPRI links in one pair of fiber, which significantly reduce fiber consumption. 1+1 or 1:1 ring protection is also supported and several low data rate links can be multiplexed into one link of high data rate. The main issue for the solution lies on the high cost, which hinders its large-scale deployment by operators.
OTN solution. Compared with WDM solution, OTN provides more powerful O&M capability, longer reach as well as flexible routing function. In addition, open interface and standard protocol of OTN, in some sense, help to bring down the cost and drerease the development difficulty. Some vendors suggested to integrate OTN functions into optical modules rather than using active line cards, which can simplify network deployment and maintenance to a large extent.
Millimeter microwave transmission. In some scenarios, it is too expensive, or even impossible to deploy fiber. In that case, microwave transmission may come to play a role as the last 100 meter fronthaul solution. 60GHz is currently the most common frequency band for milli-meter microwave and can be implemented under loose
38
China Mobile Research Institute
regulation in many countries. The bandwidth in 60Ghz band is wide and thus it is easy to get channels with 250MHz or wider bandwidth. With simple modulation technique, it is easy to achieve over 1Gbps transmission rate within 100 ~ 400 meter range. For LTE RRU with 20MHz bandwidth and 2 antennas, the data rate after 2:1 compression is less than 1 Gbps and can be transmitted via millimeter microwave. 5GHz millimeter microwave products just came into the market and can support the fronthaul transmission of 20MHz LTE with 8 antennas.
CPRI redefinition. The basic idea of CPRI redefinition is to move a partial set of physical layer functions to the RRU side in order to reduce the required data rate between the BBU and the RRU. There can be several possibilities on the function partition. By carefully designing the partition scheme, the data rate between the BBU and the RRU can become elastic and varying with real user traffic, which is the opposite from traditional case in which the I/Q stream is constant even when there is no real traffic. This feature not only helps to reduce the capacity requirement on switching network within the BBU pool but also reduces the switching latency. In addition, the data can now be encapsulated in form of packets rather than a constant stream and therefore can be transmitted by packet switching protocol, such as Ethernet which enjoys the benefits of improved flexibility and improved switching efficiency. One of the biggest disadvantages however, is the need to change the existing CPRI specification, which increases the difficulty in realization.
WDM-PON. There has been some discussion on using WDM-PON as a C-RAN fronthaul alternative in FSAN and ITU-T Q2 working group recently. The basic idea is to make use of the rich fiber resource deployed for FTTx and design a new technology based on the combination of the low-cost PON and WDM for CPRI transmission. WDM solutions adopt colored optical modules, which raises the bar in SFP installation, maintenance and storage. In comparison, WDM-PON targets at using colorless SFP, which greatly helps to simplify the installation, maintenance and storage issues. In addition, WDMPON claims such advantages as cost reduction, saving on fiber consumption and flexible topology support. Despite being at the initial stage, in the long run WDM-PON can become one of the most efficient fronthaul solutions for C-RAN.
5.2 Dynamic Radio Resource Allocation and Cooperative Transmission/Reception One key target for C-RAN system is to significantly increase average spectrum efficiency and the cell edge user throughput efficiency. However, users at the cell boundary are known to experience large inter-cell interference (ICI) in a fully-loaded OFDM cellular environment, which
China Mobile Research Institute
39
will cause severe degradation of system performance and cannot be mitigated by increasing the transmit power of desired signals. At the same time, in view of the analysis, single cell wireless resources usage efficiency is low. To improve system spectrum efficiency, advanced multi-cell joint RRM and cooperative multi-point transmission schemes should be adopted in the C-RAN system.
Cooperative Radio Resource Management for multi-cells The multi-cell RRM problem has been addressed in various academic studies.
Many uses
various optimization techniques in trying to determine the optimal resource scheduling and the power control solutions to maximize the total throughput of all cells with some specific constraints. To reduce the complexity incurred in the C-RAN network architecture and the scheduling process, the joint processing/scheduling should be limited to a number of cells within a “cluster”. The complexity of scheduling among the eNBs clusters is determined by the velocity of mobile users and the number of UEs and RRHs in the cluster. Thus, choosing an optimal clustering approach will require balancing among the performance gain, the requirement of backhaul capacity and the complexity of scheduling. As shown in Fig. 5-3, UEs will be served by one of the available clusters which are formed in a static or semi-static way based on the feedback or measurements reports of UEs. In this scenario, a subset of cells within a cluster will cooperate in transmission to the UEs associated with the cluster. To further reduce the complexity, it is possible to limit the number of cells cooperating in joint transmission to a UE at each scheduling instant. The cells in actual transmission to a UE are called active cells for the UE. The active cells can be defined from the UE perspective based on the signal strength (normally cells with strong signal strength are chosen among cells within the supercell). The activation/de-activation of a cell can be done by a super eNB, which is the control entity in cell clustering and can adjust the sets scope based on the UE feedback.
Cell cluster 1 Cell cluster 2 Cell cluster 3
Fig. 5-3 The UE assisted network controlled cell clustering
40
China Mobile Research Institute
Cooperative Transmission / Reception Cooperative transmission / reception (CT/CR) is well accepted as a promising technique to increase cell average spectrum efficiency and cell-edge user throughput. Although CT/CR naturally increases system complexity, it has potentially significant performance benefits, making it worth a more detailed consideration. To be specific, the cooperative transmission / reception is characterized into two classes, as shown in Fig.5-4:
Joint processing/transmission (JP) The JP scheme incurs a large system overhead: UE data distribution and joint processing across multiple transmission points (TPs); and channel state information (CSI) is required for all the TP-UE pairs.
Coordinated scheduling and/or Coordinated Beam-Forming (CBF) With a “minimum” cooperation overhead, to improve the cell edge-user throughput via coordinated beam-forming: No need for UE data sharing across multiple TPs; Each TP only needs CSI between itself and the involved UEs (no need for CSI between other TPs and UEs).
Fig. 5-4 JP scheme and CBF scheme
Technical Challenges Cooperative transmission / reception (CT/CR) has great potentials in reducing interference and improving spectrum efficiency of system. However, this technology has many problems that need to be further studied before it can be applied to the practical networks. There are many challenges listed as follows:
Advanced joint processing schemes
DL channel state information (CSI) feedback mechanism
User pairing and joint scheduling algorithms for multi-cells
Coordinated Radio resource allocation and power allocation schemes for multi-cells.
China Mobile Research Institute
41
5.3 Large Scale Baseband Pool and Its Interconnection Centralized Baseband Pool There are many distributed BS products using RRH+BBU architecture in market. Some TEM‟s products have realized dynamic allocation of carrier processing within one BBU to adapt to dynamic workloads among different RRH connected to it. This architecture can be viewed as the first step of centralized baseband pool concept, but in general a single BBU has limited processing capability, typically only supporting about 10 macro BSs‟ carriers.
It‟s not yet
capable of supporting dynamic resource allocation across different BBU, thus hard to resolve the dynamic network load in a larger area. In the current RRH+BBU architecture, the RRH is usually connected to a particular BBU by a fixed link, and it can only transmits its baseband signal and O&M signaling to the BBU it‟s connected to. This makes it difficult for another BBU to obtain any uplink baseband data from that RRH. Similarly, any other BBU has difficulty sending downlink baseband data to this RRH. Because of this limitation, the processing resources of different BBUs can hardly be shared: the idle BBU‟s processing resources are wasted and it cannot be used to help the BBU with a heavy workload. The centralized baseband pool should provide a high bandwidth, low latency switch matrix with an appropriate protocol to support the high speed, low latency and low cost interconnection among multiple BBUs. In a medium sized dense urban network coverage (approximately 25 sq. km in area), with an average distance between BS of 500m, a centralized baseband pool that can cover the whole area needs to support about 100 BS. For a typical TD-SCDMA system with 3 sectors per macro BS and 3 carriers/sectors, it means that the centralized baseband pool needs to support 900 TD-SCDMA carriers.
Imagine if the centralized Baseband pool coverage
is even larger, such as 15 km X 15 km, then the baseband pool would need to support up to 1000 macro BSs‟ carriers. Because of the limitation in the high-speed differential signal transmission, the traditional BBU architecture cannot scale up to support such capacity by simply expanding the backplane dimensions. Infinite Band technology can provide significant switching bandwidth (20Gbps-40Gpbs/port) and very low switching latency. It is widely used in supercomputers. However, the cost per port is very high (20,000RMB) and as such does not meet the C-RAN cost requirement. Inspired by the data center network‟s distributed inter-connect architecture, the centralized BBU pool in CRAN can also use a distributed optic interconnection to combine multiple BBU into a scalable baseband pool. Based on that, the RRHs‟ signal can be routed to any one of BBUs in the pool. Thus load balance according to dynamic network load among BBUs can be achieved, and system power consumption can be reduced. It also makes the deployment of multi-point MIMO technology and interference mitigation algorithms easier, which can improve radio system capacity.
Dynamic carrier scheduling
42
China Mobile Research Institute
The dynamic carrier scheduling of resources within baseband pools enhances redundancy of the BBU and increases overall operational reliability of the baseband pool. When a baseband card or a carrier processing unit fails, the work load can be promptly redistributed to other available resources within the pool, and restore the normal operation. In addition, for areas that have strong dynamic network load, the operator can deploy fewer baseband resources to meet the demands of different sites that have opposite peak loads at different times. For example, operator can use the same BBU pool with multiple RRHs to cover both residential areas and office areas. Then dynamically allocates baseband resources to ensure basic coverage for both areas. Remaining baseband resources can be dynamically allocated to cover the business area during working hours and the residential area during after working hours. This will increase the overall carrier resource utilization.
Large-scale BBU Inter-connection A large scale baseband inter-connect solution should be able
to support 10-1000 macro BS,
with the following requirements:
Inter-connection between BBUs must satisfy the wireless signal‟s requirements of low latency, high speed, and high reliability. The requirements are similar to the CPRI/Ir/OBRI interface, and should support real-time transmission of 2.5/6.144/10Gbps rate.
Dynamic carrier scheduling among BBUs to achieve efficient load balance within the system and failure protection without service interruption.
Support multipoint collaboration (CoMP). It needs to consider the data flow between different BBUs to support collaboration radio.
Fault-tolerance. Fiber inter connection should support 1+1 failure protection, BBU frame and baseband processing board N +1 protection to achieve high system robustness.
High scalability: it can extend the system capability smoothly without services interruption.
5.4 Open Platform Based Base Station Virtualization Current Multi-Standard BS Solutions Nowadays, most major mobile operators in the world have to operate multiple standards simultaneously. It is a natural choice to use multi-mode base stations for low cost operation. Therefore, SDR based on a common platform to support multi-standards has become the mainstream in TEM‟s products. The following are the two types of multi-mode base stations.
Unified BBU system platform supporting multi-mode by plugging in different processing boards. The processing board which supports multi-standard (such as GSM, TD-SCDMA, TD-LTE) has a unified interface and can be plugged in the same BBU system platform. Operators can use one set of a BBU system platform to support multi-standard operation. In this case, some modules of BBU system such as control module, timing module and RRH
China Mobile Research Institute
43
I/O modules can be shared between BBU processing boards which support different standards. However, this structure can't share processing resources between different processing boards and usually need to replace or add new processing board hardware for upgrades.
Unified BBU system platform and unified processing board hardware platform to support multi-mode through the software re-configuration. Through software
upgrades or
configuration, the same processing board can support different standards (e.g. LTE or TDSCDMA). In some of the latest products, the RRH can also be SDR-enabled to support different standards in the same spectrum band. This solution allows the base station to be upgraded to a new standard without changing the hardware. However, current products usually require the BBU to restart in order to download new DSP / FPGA software for standards upgrade. This limits the sharing of hardware between different standards.
In
fact, this prevents the dynamic resources allocation according to real-time traffic load without interrupt of services. Current SDR base station products partially meets the requirements of multi–standards support, however, it does not satisfy the operator flexible operation requirement of dynamically shared resources among multiple standards, load-balancing, etc.
Evolution of Software Defined Radio Driven by Moore's law in semiconductor industry, Digital Signal Processor (DSP) and General Purpose Processors (GPP) have made a lot of progresses in the architecture, performance and power consumption in recent years. This provides more choices for SDR base stations. Multicore technology is widely used in DSP and 3 ~ 6 cores processors have been commercially available. At the same time, DSP floating-point processing capacity is also improving at a fast pace. The emergence of the DSP system based on SoC architecture combines traditional DSP core and communication accelerator together has improved the BBU processing density and improved the power efficiency. Moreover, real-time OS running on DSP pave the path to virtualization
of
DSP
processing
resources.
On
the
other
hand,
DSP
from
different
manufacturers and even a same manufacturer cannot guarantee backwards compatibility. The real-time operating systems are different from each other, and there is no de fact standard yet. Generally BBUs based on DSP platform are proprietary platforms. And it is still difficult to achieve smooth upgrading and resource virtualization. Meanwhile, General Purpose Processors have progressed rapidly, and they are now capable of efficiently processing wireless signals. Therefore, the telecom industry now has more choices for software defined radio. Technology evolution in areas such as multi-core, SIMD (singleinstruction multiple data), large on-chip caches, low latency off-chip system memory are facilitating the use of GPP in traditional signal processing applications such as baseband processing in base stations. Traditional general processors usually have lower performance than DSP in power efficiency; however, in recent years the general processor has made a lot of improvements in this respect. Fig.5-5 shows the general processor technical progress in
44
China Mobile Research Institute
processing performance and power consumption in nearly 6-7 years. It is can be seen that the floating point computing capacity per watt improves very fast. These data points prove that the evolution in GPP has made it an attractive solution for various data processing tasks in the base station. The advantage of GPP is that they have a long history of backward compatibility, ensuring that software can run on each new generation of processor without any change, and this is beneficial for smooth upgrade of the BBU. On the operating system side, there are multiple OS‟s available on GPP that have real-time capability, and also allow the virtualization of BS baseband signal processing.
Fig. 5-5: Compute performance evolution of GPP *
(CPUs in 50-65 watt power envelopes used as basis for comparison in graph)
Technical progress in DSP and GPP has provided more powerful signal processing with less power consumption. This progress has made the SDR based BS solutions more attractive. Traditional DSP has become matured solution for product, and will continue to evolve. The advanced research on wireless signal processing on GPP has provided more choices for the base station, and has the potential to become part of the future open, unified multi-mode BS platform.
Base Station Virtualization Once the large scale BBU pool with high-speed, low-latency interconnection, plus the common platform of DSP/GPP and open SDR solution could be realized, it has set the base for a a virtual BS. Virtualization is a term that refers to the abstraction of computer resources. It hides the physical characteristics of a computing platform from users, instead showing another abstract computing platform. If such a concept can be utilized in a base station system, the operator can dynamically allocate processing resources within a centralized baseband pool to different
China Mobile Research Institute
45
virtualized base stations and different air interface standards. This allows the operator to efficiently support the variety of air interfaces, and adjust to the tide effects in different areas and fluctuating demands. At the same time, the common hardware platform will provide cost effectiveness to manage, maintain, expand and upgrade the base station. Therefore, we believe real time virtualized baseband pools will be part of the next generation wireless network, as shown in Fig. 5-6. Within in given centralized baseband pool, all the physical layer processing resources would be managed and allocated by a real time virtualized operating system. So, a base station instance can be easily built up through the flexible resource combination. The real time virtualized OS would adjust, allocate and re-allocate resources based on each virtualized base station requirements, in order to meet its demands.
Physical Hardware
Processors
…
Processors
Processors
Processors
Base station Virtualization
Base station Instances
PHY Layer (Signal processing) resource pool
BS of standard 1
C C MAC/Trans. Layer (Packet processing) resource pool
A A
M M
P P
BS of standard 2
Accelerator (CODEC, cryto, etc.) resource pool
C C
M M
P P
BS of standard 3
C C
Control & Manage (O&M processing) resource pool
A A
A A
M M
P P
Fig. 5-6 Baseband Pool All the adjustments will be done by software only. With this mechanism, the base stations of different standards can be easily built up through resource reconfigure in software. Also, cooperative MIMO can get the required processing resources dynamically. In addition, the processing resources can be assigned in a global view, thus the resource utilization can be improved significantly.
Technical Challenges Since wireless base stations have stringent real-time and high performance requirements, traditional virtualization technique is challenged to solve the latency requirements of wireless signal processing. In order to implement real time virtualized base station in a centralized base band pool, the following challenges have to be solved:
46
High-performance low-power signal processing for wireless signals.
China Mobile Research Institute
General purpose processor and advanced processing algorithm for real time signal processing
The high-bandwidth, low latency, low cost BBU inter-connection topology among physical processing resources in the baseband pool. It includes the interconnection among the chips in a BBU, among the BBUs in a physical rack, and among multiple racks.
Efficient and flexible real-time virtualized operating system, to achieve virtualization of hardware
processing
resources
management,
and
dynamic
allocation
of
physical
processing resources to each virtual base station, in order to ensure processing latency and jitter control HW level support on virtualization in order to minimize latency.
5.5 Distributed Service Network DSN builds the elastic high-capacity switch system adopting P2P technology, which ensures high system reliability based on disaster tolerance and auto recovery technology in software implementation. By using self-organization and self-adapting technology, in conditions of capacity expansion, equipment failure or overload, the configuration can be completed automatically with little manual work, thus reducing OPEX. DSN can replace traditional carrier-class equipment with a general purpose server, and DSN introduces virtualization technology, the DSN nodes are encapsulated in VM(Virtual Machine), through VM live migration, when the traffic goes down, multi DSN nodes can aggregate to a few physical servers, and other servers can be turned off, thus implementing energy conservation and emission reduction.
Distributed Service Network DSN element C-RAN element
BBU pool BBU pool
Fig. 5-7 C-RAN Integrated with DSN
China Mobile Research Institute
47
In a platform layer, DSN and C-RAN both encapsulate their network elements through virtualization technology on general servers, so, it is possible to run DSN and C-RAN on the same virtualized platform. But how to implement the resource management (including the dimension of time and the dimension of physical resource )is the key issue in the research of platform unification for DSN and C-RAN.
48
China Mobile Research Institute
6 Recent Progress To accelerate the development and commercialization of C-RAN, China Mobile has been working actively with industry partners. We have made good progress in field trial, large scale BBU pool implementation, and baseband PoC based on IT platform. This chapter will first introduce the advantages and disadvantages observed in C-RAN field trial, followed by discussion of large scale BBU pool solution, up to 1000 carriers, based on current BBU device, and lastly the recent R&D result of multi-mode PoC based on IT platform.
6.1 C-RAN Field Trials 6.1.1 TD-SCDMA and GSM Field Trial China Mobile conducted the first C-RAN trial with partners in 2010. It is a C-RAN centralized deployment field trial within the commercial TD-SCDMA system in Zhuhai city, Guangdong province. After that, there has been multiple GSM field trials conducted in multiple cities throughout China, include Changsha, Baoding, Jilin, Dongguan, Zhaotong, etc. Rest of the section discusses the pros and cons of the C-RAN centralized deployment solution‟s pros and cons in different scenarios. For the ease of discussion, two typical cases, TD-SCDMA trial in Zhuhai city and GSM trial in Changsha city are shown here.
Overall situation The first trial in Zhuhai City only took 3 months to complete. The commercial trial has 18 TDSCDMA macro sites covering about 30 square km area. This trial has verified some centralized deployment technologies feasibility. The construction and operation of a commercial clearly highlighted the C-RAN‟s advantage over tradition RAN in cost, flexibility and energy savings. At the same time, it also exposed challenges on fiber resource, as well as transmission construction. After that, there have been several trials on centralized deployment solutions of GSM system. The network layout is mainly consisted of replacing and upgrading existing sites. There are total15 sites covering 15 square km in the trial, where only 2 of them are new sites. Compared with TD-SCDMA network, GSM solutions have
unique features, for example, it could support
daisy-chain of 18 RRHs with only 1 pair of fiber. This could significantly reduce the number of fiber resources needed in C-RAN centralized deployment with dark fiber solution. The following sections will describe the network status before and after C-RAN deployment, key technology introduced, field test results and challenges observed. .
Field Trial Area
China Mobile Research Institute
49
The trial area in Zhuhai city is mainly consisted of a national high tech development zone, a residential community, and a few college campuses. The data traffic in this area is growing rapidly, as the customers here are well-educated and early adapters of new services. Part of the trial areas has demonstrated tidal effect of traffic loading, with predictable traffic loading pattern associated with time, location or event. For example, the national high tech development zone has most people during working hours. The same group of customers usually returns to nearby community after work. Students in colleges tend to stay away from using wireless devices during school hours, while they tend to make a lot of calls in night. Traditionally, network planning must support the peak traffic load at each individual site, which is usually 10 times higher than the down time This results in a very low average utilization rate of the BTS devices. It also introduces difficulties in network planning, construction and optimization. It is suitable to adopt baseband pool with dynamic carrier allocation. In the trial field, there will be 9 sites co-located with existing GSM site, while another 9 sites is new. All these 9 sites have to be connected with new fiber channels and they are spread in 30 square km. This is a challenge for fiber construction. The trial area in Changsha city is consisted of a few campuses near Yuelu Mountains. The traffic load and traffic density is quite high here. In addition, there is a lot of dormitories, and local residential apartments. The propagation environment is very complex and the coverage KPI still has room to be improved. This makes it suitable to verify C-RAN‟s capacity in urban city environment. Finally, since most of the trial sites are reusing or upgrading existing ones, there is plenty of fiber resources.
Overall Solution The solution starts with planning of system capacity in centralized deployment. In the Zhuhai trial, each TD-SCDMA site‟s configuration is 4/4/4, which means that there are 3 sectors in each site, and every sector has 4 carriers. Overall, the 18 trial sites need 216 carriers. When considering the BBU pool capacity, the total BBU pool can be planned to support the maximum co-current traffic for the same area. There are two kinds of TD-SCDMA carriers, R4 carrier is mainly used for voice traffic, and HSDPA carrier is mainly used for data traffic. Based on China Mobile‟s planning requirements, every site‟s traffic load should not exceed 75%. As a result, each R4 carrier supports up to 203 voice users, and each HSDPA carrier can support up to 93 users. There are total 17,000 effective users in the trial area. When BBU pool is deployed, 160 carriers will be able to support 20,000 effective users. This means the C-RAN centralized deployment can save the BBU capacity by roughly 25%, compared with traditional deployment method. Similarly, the trial in Changsha also has used the co-current capacity to decide the total capacity of the BBU pool.
50
China Mobile Research Institute
The second part of the solution involves dynamic carrier allocation. In TD-SCDMA system, each RRH/sector can support maximum 6 R4 and HSDPA carriers. In the idle situation, each RRH/sector has only one R4 carrier and one HSDPA carrier. There are different carrier allocation decision criteria whether more R4 and HSDPA carriers should be added. Whenever the existing R4 carrier‟s loading rate is above a threshold, there should be more R4 carriers allocated in this site. For HSDPA carrier, similar rule applies. Where there is not enough load in multiple R4 or HSDPA carrier, it is also possible to reduce the number of R4 and HSDPA carriers in one sector. For GSM system similar rule also applies but the criteria is the utilization rate of each GSM carrier. The third portion of the solution involves RRH daisy chain and fiber failure protection technologies. These technologies are derived from the distributed BBU-RRH deployment method which usually uses point-to-point dark fiber connections. When BBU-RRHs are separated by significant distance, it is important to consider the saving of fiber resource and protection against unpredictable fiber failure caused by external factors. In TD-SCDMA, each fiber link can handle up to 6.144Gpbs transmission, enough to support 15 TD-SCDMA carriers. Thus, one pair of fiber is able to support one site with 3 sectors and maximum carrier of 15. In the Zhuhai trial, each access ring has 9 sites and used 9 pair of fibers to support the 9 sites connected to the ring. On the other hand, GSM has far less baseband requirement due to its narrow band nature; therefore it can support more capacity in daisy-chain configuration. There are commercial products that can support 18 to 21 RRH daisy chained on one pair of dark fiber. We can calculate the fiber resource required per access ring as following: usually, each access ring has 8~ 12 physical sites and each site has 3 sectors, and has 900M and 1800M dual bands. This means, each access ring may has up to 16~24 logical sites, which is 48 to 72 sectors/RRH. To connect all the RRH in daisy chain, we would need 4~5 pair of fibers in the ring. Lastly, the field trial has also verified key technology for outdoor deployment, like power supply for remote sites. In the Zhuhai Trial, there is no BTS equipment room in the 9 new sites. Thus the traditional DC power supply is not available. External power booth is used instead. Existing outdoor power solution met the need of network deployment: with sufficient operation temperature range, -40℃~+70℃, C-level anti-flash capacity and theft-proof solution to ensure the safety of device without on-site attendance. GSM and TD-SCDMA remote site both can apply this outdoor power solution.
Technical Performance This section will outline the technical performance data from selected test cases in the trial, starting with the dynamic carrier allocation procedure. The following figure illustrates the total number of carriers allocated to one sector in a typical day on one site in Zhuhai trial. The blue curve represents this sector‟s total carrier capacity, while the purple curve represents the actual
China Mobile Research Institute
51
network load for this sector. It is clearly shown that the dynamic carrier allocation has adapted effectively to dynamic load in network.
Fig. 6-1 Dynamic Carrier Allocation We also collected KPI of radio performance for both dynamic carrier allocation and static carrier allocation. We noted no KPI difference. In the Changsha trial, the C-RAN centralized deployment has shown better radio performance and improved user experience, due to the introduction of co-located multi-RRH per site technology. With this technology, multiple RRH transmit and receive signals for the same cell, just like fiber repeat does but provide additional receive combination gain. Multiple radio performance is improved, include uplink receive quality improved by 2%~3%, drop call rate was reduced and nearly eliminated in some sites. In addition, since inter-site handover has become an internal procedure in one BBU pool, the handover delay has been reduced. Finally, the fiber protection was in place
when the access fiber ring was cut accidently, the BBU-RRH
traffic will be automatically switched to another unaffected route in the ring. The switching delay during the failure protection is comparable to normal cross-BTS or cross-MSC. Thus the failure protection has very limited effects to network KPI. In summary, C-RAN centralized deployment does not have negative effect on radio performance. On the contrary, it may provide extra gains on radio performance. Moreover, RRH daisy chain could reduce the dark fiber resource needs, while out-door units meet the power requirement of out-door remote sites. Now dark fiber transportation solution has been well verified, and other transmission technologies are in testing.
Economic analysis The trial in Zhuhai city shows that, compared with traditional RAN deployment method, C-RAN centralized deployment can reduce the TD-SCDMA network‟s CAPEX and OPEX significantly, especially for new TD-SCDMA site which is not reusing existing GSM site. In the following figure, it is shown that OPEX and CAPEX can be reduced by 53%, and 30% respectively for new cell sites.
52
China Mobile Research Institute
Fig. 6-2 Economic analysis for centralized deployment On OPEX, the savings are mainly come from A/C power consumption, site rental fee, regular on-site maintenance visit , and reduced human resource on repair and upgrade. The key factor is C-RAN has only RRH in remote site and no BTS equipment room, the site rental fee is much lower, and O&M cost is also lower. This is an important saving, as the site rental fee is a significant portion of the Zhuhai system TCO. On CAPEX consideration, the savings are mainly from: no new BTS room, reduced transmission devices on each remote site, and eliminating of various supporting devices in remote site. In addition, the adaption of BBU pool can reduce the BTS configuration and potentially lower the CAPEX on RAN. In GSM trial, similar CAPEX/OPEX savings have been observed. However, it is very clear that the savings achieved in these two cases are different, due to the different fiber resources, different deployment scenarios in different city. All-in-all, the economic analysis has shown the benefits in different areas. It is able to reduce RAN‟s O&M. however, it may be important to take account of each individual case to better calculate the saving of CAPEX and OPEX. In addition, RRH requires much less and power, it is easier to find new site, and easier to move to different place, which largely reduces the risk of cell sites being forced to relocate due to regulations or neighborhood complaints, and the cost and service disruptions associated with these.
Construction Impact The centralized deployment of C-RAN greatly simplifies the remote site selection and construction requirements, construction time required for new base stations, which lead to faster network deployment. Table 3 shows the comparison of the construction process between traditional base station and C-RAN centralized approach in the China Mobile‟s TD-SCDMA network deployment in Zhuhai City, Guangdong Province. From figure 6-3, C-RAN showcases the advantage of deployment time. The savings are mainly from site selection/purchasing, base station equipment room construction and transmission system debug, etc.
China Mobile Research Institute
53
Table 3. Impact of C-RAN Centralization concept to construction cycle Process
Traditional Base Station Construction
Centralized Base Pool Construction
site selection
Stringent,
flexible
Equipment room
Site rental and construction
No site construction for RRH
Power supply
equal
Site Equipment
Installation needed
No requirement
transmission
Installation and verification needed
Only verification needed
Equipment install
Radio system and BBU
RRH and centralized BBU
Verification
Distributed BBUs require higher verification
centralized BBUs require less verification effort
Fig. 6-3 Construction cycle comparison
Power Consumption Analysis C-RAN RRH does not require on-site equipment room and associated air conditioner which reduce electricity cost. Comparing to traditional base station, single RRH can save up to 75.3% power consumption in the China Mobile‟s TD-SCDMA network deployment in Zhuhai City, Guangdong Province. The itemized energy saving is listed in table 4.
Table 4. Power consumption comparison RAN architecture
Base Station equipment
Air conditioning
Switching Supply
Traditional
0.65 KW
2.0 KW
0.2 KW
0.2 KW
0.2 KW
3.45KW
C-RAN
0.55KW
0
0.2KW
0.10KW
0
0.85KW
Storage Battery
Transmission System
Total
Summary C-RAN centralized commercial access network demonstrates several benefits including: 1) simplified site selection and improve the speed of location selection negotiations; 2) reduced base station construction and maintenance cost, improved network deployment efficiency; 3) reduced supporting facilities of remote cell sites, led to construction cost reduction by 1/3 per site.
54
China Mobile Research Institute
In terms of network operation, C-RAN takes advantage of low cost, energy efficiency RRH. Centralized BBU facilitates easy maintenance and flexible upgrade. The overall network utilization can be improved due to virtualization technology and resource sharing which not only increases
utilization
but
also
lowers
overall
power
consumption
thru
various
power
management schemes.
6.1.2 TD-LTE C-RAN Field Trial In previous sections many benefits that C-RAN can bring to network deployment have been demonstrated in terms of TCO cost reduction, speed-up of site construction and power consumption saving. However, just as every solution has its own pros. and cons., some disadvantages of C-RAN are also revealed through field trials. In particular, centralization of CRAN requires very high fiber consumption, thus imposing heavy burden on fiber resources. Take TD-LTE systems with 8 antennas, which is the most common scenario in CMCC‟s network for example. The CPRI data rate between BBU and RRU is as high as 9.8Gbps. To transmit one such carrier requires 2 pairs of fiber, i.e. 4 fiber cords when using 6G optical modules. As the number of carriers increase, the consumption on fiber resource will increase dramatically. Therefore, dark fiber solution will be no longer viable in the near future when centralization scales expands and more carriers are introduced into the networks. Other solutions are needed to further reduce fiber consumption and make it feasible for large-scale centralization.
CMCC has been conducting TD-LTE C-RAN field trial in the city of Chengdu, Fuzhou and Guangzhou since the 2nd half year of 2012 with the target to demonstrate efficient fronthaul solutions which can help reduce fiber consumption. More concretely, two technologies, i.e. CPRI compression and Single Fiber Bi-direction (SFBD) are tested in the trials
Lab tests have already shown that lossless CPRI compression can be achieved with 2:1 compression ratio. In other words, the number of fiber needed for the centralization could be saved by half with compression implemented.
SFBD is a technology to allow simultaneous DL and UL transmission within the same fiber cord. That is to say, half of the fiber can be further saved with SFBD technology.
When combining CPRI compression and SFB together, fiber resource could be saved by 4-folds with lossless performance.
China Mobile Research Institute
55
These field trails in Chengdu, Fuzhou and Guangzhou have centralization scale of 6~12 sites centralized, i.e. 18 ~ 36 8-antenna TD-LTE carriers. Similar configuration is adopted as shown in the table below while commercial eNBs and EPCs are used together with test UEs.
Table 5: The system configuration in the C-RAN field trials. Frequency
2.85GHz
Bandwidth
20MHz
UL/DL configuration type 1
Normal CP
Special Subframe configuration type7 (DwPTS:GP:UpPTS=10:2:2)
DwPTS for data transmission
Frame structure
CPRI
2:1 compression
Optic module
Single Fiber Bi-direction
UL
SIMO
DL
Adaptive MIMO
QCI
9
Scheduler
PF
Extensive test cases have been carried out including total system throughput, end to end delay, protection switching etc. to demonstrate comprehensively the performance with compression and SFBD. We also compared system performance with and without the usage of those two technologies. Test results verified that compression (with 2:1 compression ratio) and SFB are mature enough and the system performance is almost the same as without the adoption of the technologies.
Despite the fact that combination of compression and SFBD can save fiber resource to 1/4, it is still far from enough for future C-RAN large-scale deployment. Therefore now CMCC is actively exploiting other more efficient and cost-optimized fronthaul technologies. So far WDM-based schemes, which carry dozens of carriers on a single (pair of) fiber seem to be the most promising fronthaul solutions to C-RAN large-scale deployment. Several tests on WDM solutions are now undergoing. Initial results are quite promising. Being transparent to CPRI transmission, the WDM solution can be easily implemented in 2G, 3G and LTE networks. Moreover, using DWDM, it can transport more than 15 8-antenna TD-LTE carriers with just a pair of fiber, which greatly saves the fiber consumption. It also has various topology support including ring, tree, star etc., which makes it flexible for network deployment. In addition, it supports either 1+1 or 1:1 protection with low-latency link reversal. The whole results would be presented in the upcoming version of this WP.
56
China Mobile Research Institute
In addition to pure WDM solution, there is another WDM-based solution called WDM-PON, which, targeting at low-cost implementation by the nature of PON technology, is now under discussion and development stage within ITU-T.
6.2 Cooperative radio technologies under C-RAN C-RAN architecture can facilitate cooperation among multiple cells, including multi-cell cooperative radio resource management which is to maximize the total network throughput by jointly scheduling the wireless resource and power control over multiple cells rather than considering single cell. In this section, taking JP-CoMP for example, the performance comparison is shown between non-collaborative JP in a traditional distributed RAN and collaborative JP in C-RAN. The simulation parameters are listed in Table 6. 8-antenna eNB and 2-antenna UE are used. In cooperation, 2 cells are dynamically selected based on the radio channel quality. And eNB can obtain the DL channel information via TDD‟s channel reciprocity characteristic. The simulation results of the spectrum efficiency of the downlink non-coordination and coordination transmission are illustrated in Figure 6-4. The MU-MIMO downlink transmission is adopted by all the three scenarios: non-coordination, 3-cell coordination CoMP in a eNB and 9-cell coordination CoMP in 3 different eNBs.
Table 6 Parameters for System Simulation Parameters Duplex Mode Antenna
TDD (DL:UL=2:2) BS 8Tx,UE 2Rx.
CoMP Mode
2-cell dynamic JT
Channel
Non-ideal channel
Estimation
estimation based on SRS
Channel Mode
ITU UMi Channel
China Mobile Research Institute
57
Fig. 6-4: Simulation Results of Spectrum Efficiency of Coordination and Non-coordinaiton It can be seen that compared to non-cooperation (i.e. MU-BF in LTE-A) case, with 8 transmission antennas mode, in ITU UMi scenario the gain on average spectrum efficiency can be 11.7% and 28.5% for 3-cell and 9-cell cooperation respectively. In terms of cell edge spectrum efficiency, the gain can be 38.5% and 59.2% respectively. In order to obtain the CoMP gain with larger scale coordination clusters, the ideal hexagonal cell model is applied. Taking 3-cell coordination cluster on an eNB, 9-cell coordination cluster on 3 different eNBs and 21-cell coordination cluster on 7 different eNBs for example, the edge areas ,illustrated in Fiugre X, are sorted into three types: coorperative area of the same eNB, coorperative area among different eNBs and interference area. As the scale of the collaboration clusters grows larger, more CoMP gain on the edge areas can obtained and the total network performance can be improved too. Due to the difference between cooperation within an eNB and among different eNBs, the gain should be calculated separately.
58
China Mobile Research Institute
Interference Interference Area Area Coorperative Coorperative Area Area of of one one eNB eNB Coorperative Coorperative Area Area among among different different eNBs eNBs
(a) 3 Cells on one eNB
(b) 9 Cells on 3 eNBs
© 21 Cells on 7 eNBs
Fig. 6-5:Cooperative Area of Different Cooperative Cluster With the above model, the average CoMP gain of different clusters can be estimated. The ratio of the gain with co-site cooperation to dif-site cooperation is different when the inter-site distance varies. Generally speaking, the smaller the inter-site distance is, the larger the gain under multi-eNB cooperation is. The CoMP gain is shown in the figure below, considering different co-site/dif-site cooperation ratio. It can be seen that as the scale of cooperation clusters increases, CoMP gain grows too. When the distance between sites decreases, the effect is clearer.
*Gotten from the simulation result. Fig. 6-6: CoMP gain of different collaboration cluster scale
China Mobile Research Institute
59
6.3 PoC development on C-RAN BBU pooling 6.3.1 Large Scale Baseband Pool Equipment Development In the first half of 2011,
China Mobile Research Institute (CMRI) and its C-RAN partners
developed large scale Baseband Pool supporting more than thousands of carriers. The innovation includes the IQ data routing switch method designed by CMRI, using existing equipment. Several C-RAN partners have made breakthrough progress to expand the scale of Baseband Pool beyond thousands of carriers.
The large scale of Baseband Pool is based on
distributed multilayer switch architecture, with high serviceability, low maintenance and flexible capacity expansion. This section describes the key technology for large scale baseband pool development -- IQ data routing switch, and its adaptive improvement for telecommunication equipment. Finally, it briefly highlights the key technical characteristics of the equipment. IQ Data Routing Switch Architecture IQ data routing switch is the core unit of the large scale baseband pool. It is capable of switching any RRH data to any baseband processing unit for data processing. This data switch architecture is based on the Fat-Tree architecture of DCN technology. The advantages of this architecture include: -
Fault-tolerance and disaster-tolerance (high reliability)
-
Better switch capability
-
Less requirement to each switch node
The objective of Fat-Tree Network topology is to implement a non-blocking connecting data communication network. When a computer networks use a single root node and binary tree structure, the data communications between the computers that connect to separate trees will go through the same root node. The switch capacity of the root node becomes the bottleneck. The Fat-Tree topology introduces multiple nodes switch architecture with the load-balance capability. With the benefit of two or multilayer of the switch architecture, any one high node maintains connectivity to multiple low nodes. Then several high nodes can act as backups for each other, and have the same capability of switch and connection.
Under this structure, each
switch node has the same number of switch ports, and maintains the same required transmission bandwidth. Therefore reduces switch capability requirement for each node. There is at least one connection between any lower processing node and other processing node. If one connection is out of service, redundant connections can play a backup role, which results in a highly fault tolerant networks. As shown in the following figure:
60
China Mobile Research Institute
Fig. 6-7 Multilayer Rout/Switch Architecture Current commercial BBU equipment primarily used stack of baseband processing units, plus a backplane with switching capability. It switches the RRH baseband IQ data to a specified baseband processing unit, thereby creating a pre-planned processing capability of baseband pool.
The limitation of this approach is the amount of data flow from the interconnection
between any two equipments is limited by the capability of the backplane of single equipment. So today‟s design can only support connection between 2 sets of equipment. Consequently upgrading a single equipment capacity by adding more baseband process units will demand higher switch capability of the backplanes. To combat this limitation, China Mobile Research Institute proposed to apply the Fat-Tree structure into existing wireless BBU equipment. Without significant changes to the existing equipment, the proposal adds a set of high layer switch unit to form Fat-Tree Topology to gain higher switch and baseband pool processing capacities. Similar to how the Computer network works, at this network structure, each baseband processing Board, through the high layer network, can transfer its data to other baseband board that is in lower utilization state.
Furthermore having several redundancy
boards in the baseband pool will increase redundancy, and achieve real-time protection, thus improving the reliability of the equipment. However, contrasting to the computer network, IQ data routing switch has additional characteristics. First of all, Baseband signals require real time processing, and bound by its frame structure of GSM/TD-SCDMA/TD-LTE protocols. Each frame has strict timing requirements. IQ data routing switch cannot send a data packet belonging to a single carrier, over different connections to the receiver. Otherwise it will require the receiver to rearrange the received data packet, which will generate additional delay. The End-to-end transmission cannot be routed multiple times, which .causes delay and jitter at the received end. China Mobile Research Institute has proposed a Pre-distribution Routing technology to solve this problem. Its principle is to pre-
China Mobile Research Institute
61
allocate resource before connection is established, making each switching node setting aside adequate resources and identifying of the next routing port. Secondly, IQ data transmission requires relatively large bandwidth, it is important to consider transmission path load balancing, otherwise it could easily cause the route blockage during overload. Therefore China Mobile Research institute has proposed the Load Balanced technology. The principle is that: for a routing node receiving a data flow, the data flow with the source address of Src, the object address of Dst, the flow (each data spread sent of is 1 or multiple carriers) data numbered the Num, routing node finds the routing table based on Dst. If the routing table includes multiple suitable “next jumps”, the routing node will generate a random number according to (Src, Dst, Num), then determining the address of next Jump based on the random number. This has resulted in path selection of randomization. With the Path selection of randomization, even if the Src and Dst are same, the difference of the carrier number (Num) will generate different path/route, so as to achieve the load balancing. Distributed Architecture In addition to IQ data routing, we need to consider implementation of resource management, signal processing functions and so on, for a large scale baseband pool. China Mobile Research Institute has introduced the Distribute Architecture. Use ZTE equipment as an example, a single baseband processor BBU module can handle the Iub interface signaling and servicing processing, based on the largest capacity in a network with 108 carriers. A distributed framework can solve the problem of large scale processing, retain service processing unit for each box. At the same time, a separate Ethernet switch handles dynamic resource management. Each box has separate and independent Iub ports; it logically becomes independent network elements of NodeB. In addition, one extra master network element manages entire resource of the rack, and controls redistribution of individual physical resources. This approach is simple to implement, adding a box means gaining one more independent NoteB network element, without any impact to other network elements. Also, when a baseband processing unit fails, the failed unit, under the master redistribution mechanism, can redistribute its original signaling information to other box over the Ethernet.
6.3.2 C-RAN prototype based on General Purpose Processor China Mobile, in collaboration with IBM, ZTE, Huawei, Intel, Datang Mobile, France Telecom Beijing Research Center, Beijing University of Post and Telecom, China Science Institute, jointly developed the C-RAN prototype supporting multiple air interfaces, entirely using platform based on general purpose processor. The prototypes supporting GSM and TD-SCDMA have successfully completed interoperability with commercial end user devices, while the TD-LTE version has gone through testing with UE simulator. The prototypes have proved the feasibility of implement GSM/TD-SCDMA/TD-LTE physical layer signal processing on general purpose processor based platform, and a step closer to achieve greater software implementation and upgrade flexibility.
62
China Mobile Research Institute
The following sections will describe hardware and software architecture of the prototype. As shown below, the PCI Express interface is connected to CPRI/ir interface converter, which then carries GSM/TD-SCDMA/TD-LTE signals to commercial RRHs. IQ samples of all three standards are processed by the commercial server in real time.
Fig. 6-8: IT Server Platform Topology The C-RAN proof of concept focuses on baseband processing feasibility on IT server, therefore, the software develop does not cover any core network functions. The baseband processing software is developed on Linux, and has implemented Layer 1, 2 and 3 on GSM and TD-SCDMA, and Layer 1 processing on TD-LTE, with plan to add MAC scheduling in the near future. As a result, the system currently only supports single UE. In the future, the TD-LTE system will support MAC, L2, L3, LTE-A features like CoMP, and completes interoperability with commercial devices. Signal processing carries stringent real time requirements which pose challenges to the IT servers. GSM protocol requires each frame being processed within 40ms; TD-SCDMA frame is 5ms, while TD-LTE protocol requires every frame has to be completely processed within 1ms. Typical IT operating system is not designed to meet telecom grade real time requirements, therefore subframe scheduling delay, resource management are not typically guaranteed to complete fewer than 1ms. In addition, IT platform generally lacks the stringent timing required by base station. Lastly, traditional signal processing algorithm is typically designed to be implemented on ASIC, FPGA and DSP. Therefore, many believe that IT server is not capable of handling complex signal processing such those of LTE. However, the C-RAN trial has so far proved that IT server can meet the aforementioned challenges with technology innovations. First step is to expand the real time capability on IT server to meet the subframe processing timing and accuracy demand. In addition, by adding
China Mobile Research Institute
63
hard real time and synchronization on the CPRI/Ir interface card, we can separate the RRH „hard real time‟ CPRI/Ir functions from the IT signal processing tasks which only require soft real time. Finally, significant effort had been spent to optimize LTE algorithm on general purpose processor, fully utilizing every available instruction set and memory to the maximum advantages, therefore significantly increases the CPU processing efficiency. We were able to implement 3GPP release 8 TD-LTE physical layer entirely on software running on general purpose processor and meeting all the timing and delay benchmarks. The TD-LTE implementation parameters are: 20Mhz bandwidth, 2x2 MIMO downlink, 1x2 SIMO uplink, 64QAM/15QAM/QPSK modulation, Turbo decoder with adaptive early termination. Under peak throughput, every subframe was being processed under 1ms TTI, meeting the most stringent HARQ processing latency requirements in TD-LTE. As expected, GSM and TD-SCDMA processing met the timing requirements with flying colors. Based on trial results to date, we can conclude that CPU is capable to process baseband signal processing work load and associated real time requirements. Cycle counts of certain modules take up higher proportion of the overall processing time, such as turbo decoder, convolution decoding, FFT processing etc. By introducing co-processing of such tasks, we can expect to increase overall efficiency by 5 times or higher. In the not too distant future, general purpose CPU implementing BBU functions, combining with DSN, will be the foundation of an open platform that serves a large scale dynamic baseband pool, evolving into a virtualized, cloud computing C-RAN solution.
6.3.3 BBU pool prototype based on hardware accelerator The feasibility of baseband processing based on general purpose processor has been verified by the pure software BBU prototype described in previous section. However, processing efficiency tests have shown that some function modules, especially turbo decoder, convolution decoder, FFT etc., occupy a high proportion in total processing time. Further analysis has also shown that almost 10 CPU cores are needed to process one 8-antenna 20MHz TD-LTE carrier physical layer processing. Accordingly, the performance to power ratio of pure software BBU is pretty low. Based on the observations, we then proposed that hardware accelerator should be used to speed up physical layer processing so that more carriers can be handled with the same CPU processing resources, and the performance to power ratio will be improved significantly. Some prototypes are now under development which aims to verify not only hardware accelerator performance but also BBU pool functions. In this section the hardware platform architecture of the BBU pool prototype is briefly introduced.
Functional architecture of the hardware plat form According to our evaluation and assessment on preliminary pure-software prototype, in
64
China Mobile Research Institute
downlink, processing modules of high computation load include FFT, 8-antenna precoding and turbo encoding. The other modules, including scramble, modulation, layer mapping and so on, do not require high computation capability. They can be performed by CPU from the point of view of computing power and processing delay. However, it is more reasonable to put these parts on the hardware accelerators from interface data traffic point of view. The figure below shows the function block architecture of the Downlink Accelerated Processing Modules. The yellow box indicates the input data required by the physical-layer downlink processing function modules, including both physical-layer parameters and data carried by the channel. For example, for each user, PDSCH processing not only requires pending data block from the MAC layer of the user, but also needs the corresponding physical-layer configuration parameters of the user. The physical-layer downlink processing modules on the hardware platform will process the MAC data on PDSCH according to the configured physical-layer parameters. On the other hand, the algorithms for bit-level processing and symbol-level signal processing of PDCCH are of low complexity and can be completed by CPU. In order to call for the
physical-layer
downlink
processing
function
modules,
subcarrier
mapping
location
information is required. The green box indicates that corresponding physical-layer processing module only requires input of configuration parameters, such as PSS/SSS and a variety of reference signals. The generation of these reference signals follows the same algorithm principle and these sequences can be uniquely generated by determinate generating algorithms through a limited number of physical-layer parameters. Therefore, it is more appropriate to transmit a few physical-layer parameters than to transmit generated desired frequency-domain sequences on the interface, which can reduce the data transmission bandwidth on the interface.
China Mobile Research Institute
65
PDCCH/PCFICH/PHIC H Data after Precoding + Mapping Information
RS Control Information
PSS/SSS Control Information
RB Mapping + Precoding Matrix Information
PBCH Data after RM + Control Information
PDSCH/PMCH Data + Control Information
CRC(TB)
Effective data maximum throughput is about 300Kb per DL subframe
CB Seg/CRC Channel Coding Rate Matching CB Concate. Scrambling Bit-level RS generation
Modulation PSS/SSS generation Layer Mapping & Precoding Resource Element (RE) Mapping Effective data maximum throughput is about 4.3008Mb per DL subframe
Symbol-level IFFT Add CP Sample-level
Downlink Accelerated Processing Module
Fig. 6-9 Function Block Diagram of the Downlink Accelerated Processing Modules Similarly, in uplink high processing modules of high computation are IFFT, channel estimation, equalization and turbo decoding of PDSCH. However, channel estimation is one of the most flexible parts in uplink processing, which is difficult to achieve the compatibility among different vendors if it is realized in accelerators. The processing modules between equalization and turbo decoding are of low complexity, which can be performed by CPU from the point of view of computing power and processing delay. However, it is more reasonable to process these parts on the hardware accelerators from interface data traffic point of view. The function block architecture of the Uplink Accelerated Processing Modules is shown in the figure below. The green box indicates that the processing is completed by CPU and the shaded box indicates that the processing is completed by the hardware accelerator. It is more appropriate to send effective frequency data than to send raw IQ data back to CPU after CPRI termination, which can significantly reduce the interface data traffic. However, PRACH needs to be considered before FFT is performed since PRACH, PUCCH and PUSCH exist in the same subframe or multiple subframes in the form of time division. With respect to PRACH, it is in the oversampled state (especially the format 0 - format 3) when the sampling rate is 30.72Mhz. Digital down-sampling is usually adopted in the receiver to reduce the sampling points of RACH time-domain signal. Down-sampling of the oversampled PRACH time-domain signal inputted by the antenna interface is conducted to obtain the digital signal with a low sampling rate. After
66
China Mobile Research Institute
filtering and FFT, signal detection and peak position estimation can be done. Then the hardware accelerator separates UE data, PUCCH data and SRS data from frequency-domain data. Finally, DMRS data of PUSCH, PUCCH data and SRS data are sent to CPU which finishes the remaining processing including PUCCH receiving processing, SRS signal receiving processing, as well as channel estimation, time/frequency offset estimation and noise power estimation etc. of PUSCH. Eventually, the aforesaid PUSCH channel estimation results are output to the hardware accelerator to complete PUSCH subsequent equalization, demodulation and bit-level processing. It is worth pointing out that the equalization processing modules should be programmable, meeting the equalization implementation requirements from different vendors. Meanwhile, it supports the matrix operation with a maximum size of 8*8 and needs to support simultaneous processing of multiple smaller matrix operations (such as 2*2).
I/Q
Uplink Accelerated Processing Module
Sample-level CP Removal ½ subcarrier shift FFT Data Selection
PRACH
Data Equalization
DMRS Channel Estimation
IDFT
Time/freq Offset Estimation
Data/Ctrl De-multiplexing
Noise Estimation
Downsampling
DMRS maximum throughput is 1.3Mb per UL subframe Effective data maximum throughtput is 4.3008Mb per UL subframe
CE maximum throughput is 3.7Mb per UL subframe
De-modulation CP Removal De-scramble+interleaver FFT Turbo Decoding + CRC
PRACH Freq Data
PUCCH + SRS Freq Data
PUSCH CB Data
PUSCH
CQI/PMI/RI ACK/NACK
Fig 6-10 Function Block Diagram of the Uplink Accelerated Processing Module Up to September 2013, together with our partners we have developed a scalable BBU pool prototype based on simple accelerator. The complete commercial LTE/GSM protocols have been implemented in the prototype, which has shown the ability of communicating with commercial terminals. The prototype system architecture is shown in the figure below, only Turbo encoder/decoder and FFT/iFFT are processed on the accelerator. 18 GSM cells (supporting 108 TRXs), 9 2-antenna 10MHz FDD-LTE cells and 93 2-antenna 20MHz TDD-LTE cells are configured in the BBU pool.
China Mobile Research Institute
67
Fig. 6-11 System Architecture of BBU Pool Prototype based on Simple Accelerator It is calculated that 4 CPU cores are needed to process one 8-antenna 20MHz TD-LTE carrier. Compared to pure-software based prototype, the performance-power ratio has been improved significantly. If accelerator is enhanced based on above hardware platform, the processing resource for each 8-antenna 20MHz TD-LTE carrier can be reduced to 1~2 CPU cores and the power consumption per carrier can be 15~20w, taking into account future CPU evolution. In order to acquire more pooling benefits, especially dynamic carrier migration, we are working with several vendors to discuss the definition for functions and interfaces of hardware platform. The hardware platform can be logically divided into two parts based on preliminary ideas: digital front-end and accelerator.
Digital front-end CPRI/Ir/OBRI interface conversion and sample-level processing are completed by digital frontend. Then the processed data is sent to its inserted server by PCI Express, or switched to any other servers by a 10Ge/Infiniband switch network in BBU pool. Effective IQ data of several different networks (such as GSM, TD-SCDMA and TD-LTE) can be sent to general purpose processor by this digital front-end. As for LTE, the functions achieved by the digital front-end include CPRI processing, resource mapping, IFFT and CP addition with respect to downlink processing, as well as CP removal , FFT, UE data separation and PRACH processing with respect to uplink processing. IQ data encapsulation and extraction are implemented in CPRI processing, which is helpful to adapt RRHs from different vendors. The exchange data is proportional to user data rate with the help of resource mapping and UE data separation, i.e. RB insertion and
68
China Mobile Research Institute
selection. The data bandwidth can be reduced by PRACH pre-processing. Furthermore, if the digital front end is placed in RRU, then a new interface definition between RRU and BBU is formed.
Accelerator Partial physical layer processing and L2 encryption and decryption algorithms processing are integrated in accelerators, improving protocol processing efficiency of general purpose processor. As for LTE, the functions achieved by the accelerator include encryption and decryption processing, PDSCH/PMCH bit-level and symbol-level processing, PBCH symbol-level processing, RS/PSS/SSS generation and RE mapping of all channels with respect to downlink processing, as well as PUSCH equalization, symbol-level and bit-level processing with respect to uplink processing. The accelerator can not only effectively improve the performance per watt of the system, but also can reduce the interface traffic requirement between hardware platform and the general purpose processor.
6.4 Progress on C-RAN virtualization Cloudization is the core feature of C-RAN and virtualization is a key foundation to realize it. By introducing server virtualization technology, C-RAN system can run multiple independent isolated instances of virtual BBU on one physical server and enjoy the benefits such as effective server
integration,
addition, a BBU running
in
the
hardware resources virtual machine (VM)
saving can
and adjust
cost
reduction. In
processing
resources
dynamically according to traffic variation between busy andspare time. Moreover, BBUs with low traffic can be centralized onto fewer physical servers through VM live migration. By shutting down the idle servers, the overall system power consumption can be reduced. Teaming up with industry partners, China Mobile Research Institute associated has been extensively researching on the implementation solutions of cloud and virtualization based C-RAN and has achieved lots of achievements so far.
System architecture of C-RAN virtualization The system architecture of a virtualization based C-RAN is shown in figure 6-12.
China Mobile Research Institute
69
Fig. 6-12 C-RAN system virtualization architecture The Remote Radio Unit (RRU) located in the remote site is connects through fiber with the baseband resource pool in the data center. The baseband resource pool could be built on a x86 server clusters which are deployed with server virtualization. Each physical server runs several VMs. Each VM can be configured as the baseband unit (BBU) to run different mobile communication standards (GSM/TD-SCDMA/TD-LTE) according to the network operation plan, or as user-level application such as CDN and web cache to support service on edge.
C-RAN aims to migrate the traditional base station to general-purpose servers and implement virtualization. Toward this end, the correspondence between traditional BBU and VMs, i.e. the granularity of virtualization is worthy of study and analysis. The figure above is just an example in which one VM is assigned to process an original BBU. In practical deployment, it is possible that a virtual machine can handle only one carrier within the BBU or even smaller units. On the other hand, there is another layered approach in which a virtual machine handles
just one
layer such as L1, L2 or L3 ,or even a certain part of a layer.
Due to the difference of the communication protocols, the requirements of processing resources for GSM, TD-SCDMA and TD-LTE are quite different in terms of compute capability and realtime performance. In practice, different virtualization granularity should be considered accordingly. In the case of TD-LTE which has the highest demand for the processing resources, a virtualization granularity as figure 6-13 showed could be adopted.
70
China Mobile Research Institute
BBU1
VM
BBU2
BBU1 L3
VM BBU2 L3
VM
VM
VM
VM
VM
VM
Cell 1 L2/part L1
Cell 2 L2/part L1
Cell 3 L2/part L1
Cell 4 L2/part L1
Cell 5 L2/part L1
Cell 6 L2/part L1
Hypervisor/VMM
Hardware
Accelorator(part L1)
Physical Server
Fig. 6-13 An example of virtualization granularity for TD-LTE system Solutions for real-time constraint While C-RAN tries to migrate the traditional communication equipments to general-purpose IT platform and implement virtualization, the biggest challenge lies on the real-time constraint of wireless signal processing. The most strict real-time demand of wireless signal processing comes from the physical layer (L1). One feasible solution is to introduce the physical layer hardware accelerator so that part or all of the L1 signal processing with high real-time requirement can be dealt with by the accelerator with
dedicated chipsets, e.g. DSP, FPGA
(ASIC when commercial), SoC, GPU, etc.
Compared with L1, the real-time requirement of L2/L3 is relatively low, which makes it possible to be placed in the virtual machine and processed by the CPU. However, certain technical solutions for the hypervisor and the guest-OS of VM are still needed to ensure the real-time signal processing, i.e., the so-called real-time cloud and real-time virtualization. A real-time operating system is the basis of real-time virtualization. Taking into account openness, versatility and overall system cost, we are now mainly using Linux as the operating system for C-RAN. The research so far has shown that Linux kernel integrated with real-time patch is suitable and effective for C-RAN virtualization. By introducing Linux real-time preemption patch (PREEMPT_RT), conducting a series of parameter configuration and optimization, supporting preemption and hard interruption, improving locking mechanism, the kernel can immediately response and process correspondingly when receiving an interrupt request for signal processing, so as to ensure the real-time performance.
Hypervisor also needs to leverage a similar real-time solution. By integrating the real-time patch into the hypervisor or its host-OS and making parameter adjustment, the real-time problem could be solved. In addition, the hypervisor needs to be further optimized to minimize
China Mobile Research Institute
71
the overhead introduced by virtualization and the impact to the VM‟s real-time performance.
As for the selection of the type of hypervisor technology, in addition to the requirements of CRAN‟s openness and versatility, whether it can provide solutions to meet the real-time demands is the basic requirement. Currently, China Mobile Research Institute, together with partners, has proposed a series of real-time solutions based on KVM and ESXi, and kept continuous research on the performance optimization. The interrupt latency of an optimized RT-KVM system is shown as figure 6-14, in which the maximum interrupt latency is less than 14µs [12].
Fig. 6-14:
Interrupt latency of an optimized RT-KVM system
Virtualization management functions should also fulfill the real-time constraints. Taking VM live migration as an example, the service interruption time of traditional live migration cannot fulfill the real-time constraints of mobile networks. If utilized in C-RAN directly, it might happen that the VM could not accomplish the migration process normally. So certain technical solutions must be taken into account. China Mobile Research Institute is now working with partners to research the appropriate technology solutions.
I/O virtualization As mentioned above, C-RAN requires the use of L1 accelerator to solve part of the real-time problems. The data exchange between L1 and L2/L3 raises very high requirements for the I/O performance between the accelerator and the VM. In a traditional virtualization environment, due to the introduction of the hypervisor, the data communication between VM and the underlying hardware needs the hypervisor‟s intervention, which brings additional overhead, thus resulting in degradation of I/O performance. Therefore, I/O virtualization technique should be introduced to improve the system I/O performance.
72
China Mobile Research Institute
In the case of an accelerator serving only one VM, PCI passthrough technique is competent. But in the case of one accelerator serveing multiple VMs, SR-IOV technique should be adopted. Single Root I/O Virtualization (SR-IOV) is an I/O virtualization technique for single physical server, which is based on PCI-E specification and relatively mature. By dividing the physical I/O channel of PCI-e interface into multiple virtual I/O channels, the system can assign one or more virtual channels to a VM when creating it. Each virtual machine's virtual channel is independent and the data exchange between accelerator and its connected VMs is done through their separate virtual channels. Without the intervention of hypervisor, the system I/O performance is greatly improved. A server virtualization architecture applied with SR-IOV is shown in figure 6-15.
VM Cell 1 L2/L3
VM Cell 2 L2/L3
VM Cell 3 L2/L3
Hypervisor/VMM
Hardware Accelorator Server
Fig. 6-15 C-RAN virtualization architecture with SR-IOV The application of SR-IOV also has its limitations, i.e., it can only support I/O virtualization within one physical server. One of the objectives for C-RAN implementing virtualization is to centralize the relatively idle VMs onto fewer physical servers through VM live migration, and shut down the disengaged physical servers to reduce energy consumption. However, the accelerator located on the physical server needs to be connected with the remote RRU and cannot be shut off. So extra power supply for the accelerator is needed when servers are shut down, which means that the servers need special customization. Furthermore, the physical bonding of the accelerators and physical servers makes it very inflexible for system maintenance.
Developing on the basis of SR-IOV, Multi Root I/O Virtualization (MR-IOV) supports I/O resource sharing and virtualization among multiple physical servers. By separating the PCI-e devices from the physical server and forming a resource pool, I/O virtualization could be done in a resource pool range. Taking advantage of MR-IOV, the binding relationship between accelerators and physical servers can be removed, so that the system could allocate and orchestrate the resources uniformly and more flexibly. So far MR-IOV is still on the development stage. China Mobile Research Institute hopes to promote the MR-IOV technology and push forward maturity
China Mobile Research Institute
73
of ecosystem together with industry partners. By then its application in C-RAN would be expected.
6.5 Edge Applications on C-RAN With the large scale commercialization of LTE network, a wide variety of bandwidth-hungry and latency-sensitive mobile data applications are expected to explode exponentially. For operators, traditional telecom services (e.g. voice and SMS) with high profit margin are going down continuously. Despite operators‟s heavy investment in network construction and maintenance and mobile networks tend to become “cheap pipes” of OTT (over the top) internet companies. In addition, self-operated data services by operators are also facing fierce and homogentious competition, which makes the existing mobile business model hard to sustain in the near future. It is critical for operators not only to reduce network expansion cost, but also to provide differentiated QoE for mobile subscribers. On the other hand, mobile base stations, as operators‟ important and differentiated assets, have not been utilized all the time. The idea of combination of C-RAN with application is to exploit the unique advantage by building applications over the edge of mobile networks, especially C-RAN BBU pool. It is expected that in this way operators can take full advantage of distributed computation and storage capabilities to reduce network congestion and latency, and provide differentiated QoE to improve subscribers‟ loyalty as well.
6.5.1 Edge applications architecture over C-RAN GPP BBU pool
As shown in Fig. 6-16, edge applications and BBU pool software are deployed over the same hardware platform and isolated from each other via virtual machines. It is expected not only to reduce backbone traffic and latency, but also to provide rapid time-to-market deployment via GPP platform and virtualization technologies. The architecture has the following characteristics.
1.
Distributed computation and storage:
User plane traffic can be locally cached and
processed. If we deploy applications over C-RAN BBU pool serving 5000~1000 users, the hit rate of content retrieval can be guaranteed and deployment cost can be reduced as well.。 2.
Radio API: by providing real-time and refined radio network information, e.g. real-time loading and radio link status, applications on edge can be improved further.
3.
GPP platform and IT technologies: low-cost and faster development, release, maintenance can be achieved with mature development toolkits when using GPP platform.
74
China Mobile Research Institute
4.
Collaboration with cloud data centers: Applications in the cloud data centers and over the edge can collaborate and complement each other to build a smarter pipe. Data Center
Network optimization
Other servers
Content source
Core Network IP 接口
API
BBU Pool
S1
Data Content Online collection Cache gaming
Location based
VM
VM GPP HW
Base Staitons
Edge App Platform
Fig. 6-16 Edge applications architecture over C-RAN GPP BBU pool
6.5.2 Current progress We are working closely with C-RAN vendors and edge application providers. And the focus includes architecture and interfaces, e.g. radio API specification.
We have made rough estimation of the number of LTE users and the traffic models in the next 5 years. Based on those assumptions, CDN/Cache deployment locations and hit rate of content retrieval have also been analyzed. The results show that C-RAN architecture is a perfect match for edge CDN/Cache deployment. For example, we assume that the penetration rate of LTE users is 30% and the total number of LTE users is expected to reach 200 million with some provinces having 10 million users. We also assume that approximately10% of LTE users watch video for 5 minutes per day and the average bit rate is 1Mbps. Then the total traffic reaches up to 200Gbps and 8Gbps per province could be attained. It is clear that the expected traffic is going to bring considerable pressure over the access and backbone transmission network. In order to overcome those challenges, it would be beneficial to distribute CDN/Cache over the edge for better traffic offloading. Further analysis shows that if C-RAN consolidates more than 50 base stations and serves more than 10 thousand users, the hit rate of content retrieval could reach as high as 30% or so.
China Mobile Research Institute
75
In addition, we have specified radio APIs including dozens of radio parameters on user, cell and eNB levels. The APIs include real-time loading, ongoing and potential QoS, mobility, signal strength, transmission power etc. These open APIs enable the edge applications to adjust parameters such as UL/DL audio/video bit rates, picture and text compression rates etc. Edge applications can also customize the reporting periodicity and subscribe/notify the specific parameters. Currently our C-RAN vendors are developing the related API interface based on the specifications.
Lastly, we are working on Proof of Concept and verification. Figure 6-17 is the architecture of the prototype under development.
The offloading of user plane traffic can be achieved by
implementing data routing module over LTE BBU pool. Upon applications‟ request, radio API platform is able to provide real-time user/cell/eNB/Pool level radio link information. We are focusing on two types of edge applications. One is content caching with video acceleration. The popular and top viewed video is cached over C-RAN and the bit rates are adjusted. Alternatively the content can be pre-cached based on real-time radio network status. The other is LTE network optimization. By deploying the software module of radio data collection, processing and distribution over C-RAN BBU pool, most of radio data can be processed locally. Unlike traditional ways of collecting all traffic with higher bandwidth and processing requirements, in this way only small amount of necessary radio data is required to be collected and backhauled to the centralized OAM center.。
Core Netowork
C-RAN GPP platform LTE BBU Pool
User plane data routing
Radio API platform
Edge applications
Fig. 6-17 the architecture of edge application prototype
We have finished the first TD-LTE PoC with edge services recently. We adopted open source OAI for RAN protocol stack and implemented end-to-end communications with commercial terminals. In addition, we have integrated three different services to demonstrate the strengths of edge computing. 1. Voice recognition: the software client installed in terminals uploads collected voice clips to the analyzing server deployed close to RAN. And the server can convert speech to text quickly
76
China Mobile Research Institute
and send response in a rather real-time manner. In the future, we are going to explore the possibility of introducing image and video recognition. 2. Radio signal quality analysis: when the drop rate or interference is high in current serving cell, BBU software can start analyzing signal quality by importing IQ data. In this way the interference issues and other abnormal cases can be quickly identified and resolved. 3. Radio signal map: real-time radio signal statistics per users can be collected via an open API of BBU to form a real-time radio signal map so that operator can monitor each base station globally and visually.
We would like to work together with industrial community to accelerate the research, standardization, development and commercialization of edge applications. On one hand, we focus on interface standardization and functional extension of RAN equipments, in which the standardization of radio API is indispensible for inter-operability in a multi-vendor environment. Currently user plane traffic routing is based on relatively static configurations (e.g. APN, GW location), which tends to place restrictions on the flexibility of edge services. More flexible data traffic routing is essential, e.g. based on application types.
On the other hand, in order to
make full advantage of distributed computation and storage capabilities, we also hope to work with application providers to dig more use cases of edge services.
China Mobile Research Institute
77
7 Evolution Path C-RAN is not only a new network infrastructure, but also a combination of introducing a serial of new technologies in this new infrastructure. Thus, the deployment of C-RAN must be a wellplanned, step-by-step process. Due to the fact that there are many different areas with different deployment scenarios, and there are many different existing radio access technologies deployed, it is natural that C-RAN would have different scheme to best suite the above situations. For China Mobile, 2G and 3G is not longer the focus of future network build-out, but here are still very large assets of 2G and 3G system throughout the county. Thus for existing system, the focus is to maintain the network and increase network capacity where needed. The focus is TD-LTE network construction, which is also the focus of C-RAN deployment. Considering the TD-LTE construction progress and future timeline, C-RAN‟s overall deployment strategy can be divided into three stages, where different area may have different timeline on these stages.
First Stage: Early Stage in TD-LTE Roll-Out At this stage, China Mobile‟s TD-LTE coverage is focusing on macro cell coverage in selected major cities in China, to ensure the mobile broadband coverage in urban environment. Thus, the inter-cell interference among different macro cells and different sectors within one cell under same frequency network deployment is a key issue that hampers network performance. Mean while, because many cities use Band D (2.6GHz) for TD-LTE network, denser BTS is required, which further increase the challenges of finding new sites, upgrading on existing sites etc.
To address this deployment challenges, C-RAN‟s major deployment scenario is to combine with all-service access network deployment, which centralized BTS within 3~5 squire km into aggregation transport equipment room. In the same time, remote micro-RRU can be used for hotspot and coverage hole deployment to improve the network coverage and performance for area where it is hard to find new sites.
On the BTS main equipment side, the centralized BBU can build Baseband Pool, and introduce co-operative algorithm like CS/CB or JT/JR etc. Because TD-LTE with same frequency network deployment is more sensitive to inter-cell interference, it requires higher accuracy in network planning and optimization. BTS main equipment needs to gradually migrate to GPP based open platform to better support third party „network application‟ software that can monitor and analysis the network performance and interference situation in real-time. This will enable much faster network optimization and tuning based on effective in-field measurement data, and will be helpful to improve the overall network quality. Considering the time needed in BTS main
78
China Mobile Research Institute
equipment‟s migration from proprietary hardware to open hardware, it is possible to adopt hybrid proprietary + open hardware or co-located deployment of these two architecture in this stage to allow a graceful migration.
For the fronthaul transport, introduction of WDM solution is needed. This will resolve many deployment scenarios‟ difficulty in finding extra fiber sources by using existing dark fiber to transmit multiple route of CPRI/Ir/ORI in WDM. In places where there is no fiber resource at all or too expensive to deploy fiber, mmWave solution can be considered to address the „last hundred meter‟ fronthaul transmission.
Second Stage: TD-LTE Mid Term Deployment At this stage, the basic TD-LTE coverage network is completed in selected cities. The major challenge in network is further improve in capacity need in hotspot area and network coverage quality.
To deal with the capacity need, macro base station will introduce more carriers and carrier aggregation technology. In addition, more remote micro-RRUs will be introduced into network to improve the network capacity for the surging mobile broadband traffic. The TD-LTE HetNet will be composed by the overlay of existing macro base station and the additional micro-RRUs will be an important deployment scenario. There are many different technical solutions for HetNet, including same frequency of macro and micro cell deployment, carrier aggregation, separate of user plan and control plan between macro and micro cell.
Due to the tight connection between macro and micro cell in the HetNet deployment scenario, it is necessary for centralized BBU pool for HetNet. This will in turn named many advanced features in HetNet, including but not limited to: co-operative radio transmission/receiving between macro and micro cell, separation of user plan and control plan based on carrier aggregation between macro and micro cell etc. The BTS main equipment should be based on GPP open platform plus physical layer accelerator card. Meanwhile, network side applications like CDN/Cache can be deployed on GPP based BBU pool to cache certain data. This will greatly reduce the backhaul transmission traffic. Shared open platform will largely reduce the overall TCO operating Radio Access Network and the CDN devices, which is one key benefits of C-RAN.
Third Stage: Massive Capacity Enhancement of TD-LTE Network With the even surging mobile broadband traffic continue in the future, TD-LTE network will become a high bandwidth, high capacity network. The major challenge at this stage will be the limited spectrum resource to serve the huge capacity need. To improve the network capacity,
China Mobile Research Institute
79
more and more micro-RRU will be deployed in the network, and these micro-RRU will have overlapping coverage with neighbor micro-RRU. And due to the increasing number of microRRU and baseband pool size, the advantage of baseband pooling will be significant.
At this stage, more and more BBU pool will be using GPP-based open platform. The legacy 2G/3G network can be gradually replaced by soft BTS deployed on the same TD-LTE BBU pool and multi-band multi-mode RRU. In addition, more and more high layer „network application‟ can be deployed in the same BBU pool, which will enable better operation and optimization of the Radio Access Network, and enable faster innovation and deployment of services. This will enable the mobile operators to avoid the „dumb pipe‟ position in mobile internet ecosystem, and to become „smart pipe‟ to better compete with internet service providers in the future.
80
China Mobile Research Institute
8 Global landscape of C-RAN activities The C-RAN concept is being recognized by more operators and vendors to address the difficulties in LTE network deployment. It is also deemed to be one of the promising trend toward future wireless networks. So far there have been a few operators adopting C-RAN for the LTE network deployment. In the meantime, C-RAN is becoming a hot topic in many SDOs.
C-RAN deployment SK telecom and Korea Telecom, the two biggest carriers in south Korea which is famous for rich fiber availability, adopts C-RAN centralization method to deploy the commercial LTE networks. Their C-RAN deployment has a high centralization scale. It is said that the LTE network in Seoul is centralized into around 10 central offices with each supporting on average several hundred LTE carriers. In Japan with similarly rich fiber resource, DoCoMo just released a public release claiming their plan of using C-RAN for future LTE-A deployment.
C-RAN in SDOs In 2009 a dedicated C-RAN project P-CRAN was founded in the alliance of Next Generation Mobile Networks (NGMN) [15]. Led by China Mobile and received extensive supports from both operators and vendors including KT, SKT, Orange, Intel, ZTE, Huawei and Alcatel-Lucent, this project aimed at promoting the concept of C-RAN, collecting requirements from operators and helping build the ecosystem. The project was closed at the end of 2012, releasing four deliverables to the industry. Through the deliverables, the advantages of C-RAN in saving TCO cost and speeding up site construction are well understood. These deliverables also include the C-RAN requirements and initial study on key technologies as well as the potential SDO impact. In 2013 NGMN extended the study on C-RAN in a C-RAN work stream under the project of RAN Evolution. On the basis of previous C-RAN project, this work stream aims at further detailed study on key technologies critical to C-RAN implementation, including BBU pooling, RAN sharing, function split between the BBU and the RRU, and C-RAN virtualization. In addition, the requirements on C-RAN fronthaul are specified, which is of significance to C-RAN deployment. Another organization worth mentioning is Network Functions Virtualisation (NFV) Industry Specification Group (ISG) under the auspicious of European Telecommunications Standards Institute (ETSI). Founded in 2012, this ISG is devoted to the development on the virtualization requirements and the system architecture. The idea behind NFV is to “consolidate many network equipment types onto industry standard high volume servers, switches and storage, which could be located in Data centers, network nodes and in the end user premises” [12]. So far this ISG has attracted more than 150 members from not only telecom but also IT industry. In addition, there are several C-RAN related projects under European Commission‟s Seventh Framework Programme (FP7). For example, the iJOIN project deals with the interworking and joint design of an open access and backhaul network architecture for small cells on cloud networks [13]. Another project, Mobile Cloud Networking (MCN) aims at exploiting Cloud Computing as infrastructure for future mobile network deployment, operation and innovative value-added services [14].
China Mobile Research Institute
81
9 Conclusions With the arrival of the mobile Internet era, today‟s RAN architecture is facing more and more challenges that the mobile operators need to solve: mobile data flow increases drastically caused by the popularity of smart terminals, spectrum efficiency bottleneck, lack of multistandard flexibility on the same platform, dynamic network load because of the “tidal effect”, and expensive to provide ever increasing internet service to end users. Mobile operators must aggressively consider the evolution of the RAN to a high efficiency and low cost architecture. C-RAN is a promising solution to the challenges mentioned above. By using new technologies at various stages of C-RAN, we can improve and simplify the network construction and deployment, fundamentally change the cost structure of mobile operators, and provide more flexible and efficient services to end users. With the distributed RRH and centralized BBU architecture, advanced multipoint transmission/reception technology, SDR with multi-standard support, virtualization technology on general purpose processor, more efficient way of dealing with the tidal effect and service on the edge of the RAN, C-RAN will provide today‟s mobile operator with a much more efficient, competitive, and profitable infrastructure in the dynamic market environment. China Mobile has been developing and deploying C-RAN systems since 2009. In particular, CMCC has conducted extensive field trials in more than 10 cities across China. Our field trials in GSM and TD-SCDMA have vigorously demonstrated the benefits that C-RAN centralization can bring to operators. For example, compared to distributed TD-SCDMA networks, up to 15% CAPEX and 50% OPEX could be saved using C-RAN centralization. Moreover, system roll out time is saved by 1/3 and in view of green deployment, the saving on power consumption can be as high as 70%. In the meantime, the TD-LTE C-RAN trials in the cities of Fuzhou, Chengdu and Guangzhou have verified the maturity and effectiveness of CPRI compression and single fiber bi-direction (SFBD) technologies in the fronthaul implementation. Using SFBD and CPRI compression with 2:1 compression ratio, the fiber consumption can be saved by 3 folds while keeping system performance lossless. Moreover, WDM-based solutions are being tested currently, which promises even greater potential save on fiber resource and facilitate large-scale C-RAN deployment. On the road toward virtualization, CMCC has developed an x86-based 3-mode base station prototype in which GSM, TS-SCDMA and TD-LTE were realized in a pure software manner. Although CMCC has demo-ed an end-to-end call using commercial UE and core network. the pure software implementation is not achieved cost effectively. The power-performance ratio is low. It is therefore concluded that a dedicated hardware accelerator is needed for processing partial L1 functions that are computation-intensive, e.g. iFFT/FFT and channel coding/decoding. Under the guidance of this idea, two sets of PoCs were further developed. One PoC was developed by commercial LTE protocol stack. It was proved that with the adoption of dedicated accelerator, the performance-power ratio of the general-purpose processor platform is comparable to that of traditional proprietary platform. The other PoC demonstrated the power of edge computing for new service introduction and innovative network operation and management. The ultimate goal of C-RAN is to realize resource cloudization and one of the possible solutions to achieve that is virtualization technology. In this White Paer a system reference architecture
82
China Mobile Research Institute
is proposed and further analysis showed that the major challenges for virtualization implementation lie on granularity of virtual machine, hypervisor and operating system optimization and I/O virtualization. C-RAN is a multi-stage RAN evolution which requires joint efforts from every partner in the ecosystem including both IT and telecom industry. CMCC would like to take this WP as an opportunity to call for more action, contribution and commitment on C-RAN research and development, which is the sure trend to the future.
China Mobile Research Institute
83
Acknowledgements We would like to thank Alcatel-Lucent, IBM China Research Lab, Intel Cooperation and Institute of Computing Technology, Chinese Academy of Sciences for their valuable contribution to this white paper. We would also like to express our gratitude to all C-RAN team members in China Mobile for their hard work.
84
China Mobile Research Institute
Terms and Definitions This section provides the terms and definitions for this document. 3GPP
3rd Generation Partnership Project
AIS
Alarm Indication Signal
ASIC
Application Specific Integrated Circuit
ARPU
Average Revenue Per User
BBU
Base Band Unit
BS
Base Station
CAGR
Compound Annual Growth Rate
CAPEX
Capital Expenditure
CBF
Coordinated Beam-Forming
CDN
Content Distribution Network
CoMP
Cooperative Multi-point processing
C-RAN
Centralized, Cooperative, Cloud RAN
CSI
Channel State Information
CT/CR
Cooperative Transmission/Reception
DPI
Deep Packet Inspection
DSP
Digital Signal Processing
DSN
Distributed Service Network
eNB
Evolved Node B
FEC
Forward Error Correction
FTTX
Fiber To The X
FPGA
Field Programmable Gate Array
GGSN
Gateway GPRS Support Node
GPP
General Purpose Processors
GSM
Global System for Mobile Communications
HW/SW
Hardware/Software
ICI
Inter-cell Interference
IQ
In-phase/Quadrature-phase)
I/O
Input/Output
JP
Joint Processing
LTE
Long Term Evolution
LTE-A
Long Term Evolution - Advanced
MAC
Media Access Control
China Mobile Research Institute
85
MIMO
Multiple Input Multiple Output
MNC
Mobile Network Controller
OBRI
Open BBU RRH Interface
NFV
Network Functions Virtualisation
OFDM
Orthogonal Frequency Division Multiplexing
OPEX
Operating Expenditure
OTN
Optical Transmission Net
O&M
Operations and Maintenance
P2P
Peer to Peer
PA
Power Amplifier
PHY
Physical Layer
Pon
Passive Optical Network
QoS
Quality of Service
RAN
Radio Access Network
RF
Radio Frequency
RNC
Radio Network Controller
RRH
Remote Radio Head
RRM
Radio Resource Management
SDR
Software defined Radio
SFP
Small Form-factor Pluggable
SGSN
Serving GPRS Supporting Node
TCO
Total Cost of Ownership
TDD
Time Division Dual
TD-SCDMA
Time Division-Synchronous Code Division Multiple Access
TEM
Telecom Equipment Manufacturer
TP
Transmission Point
UE
User Equipment
UL/DL
Uplink/Downlink
UMTS
Universal Mobile Telecommunications System
UniPon
Unified Passive Optical Network
VNI
Visual Networking Index
VM
Virtual Machine
VoIP
Voice over IP
WCDMA
Wideband Code Division Multiple Access
WDM
wavelength Division Multiplexing
86
China Mobile Research Institute
XENPAK
10 Gigabit Ethernet Transceiver Package
XFP
10-Gigabit small Form-factor Pluggable
China Mobile Research Institute
87
References [1] Co-Platform Multi-Mode BTS (C-P MMBTS): Leading the Trend of Multi-Mode Network Convergence, white paper from In-Stat, 2009.Multi standard [2] Cisco Visual Networking Index, URL: www.cisco.com/web/go/vni [3] Geza Szabo,Daniel Orincsay,Balazs, Peter Gero,Sandor Gyori,Tamas Borsos, “Traffic Analysis of Mobile Broadband Networks”, Third Annual International Wireless Internet Conference October 22-24, 2007, Austin, Texas, USA [4] CPRI
Specification
V4.1,
Common
Public
Radio
Interface
(CPRI);
Interface
Specification. 2009-02-18 [5] F.-Joachim Westphal. Trends and evolution of transport networks. SL SI, IBU Telco, SSC ENPS [6] 3GPP, R1-093273, SRS feedback mechanism based CoMP schemes in TD-LTE-Advanced [7] Q. H. Spencer, A. L. Swindlehurst and M.Haardt, “Zero-forcing methods for downlink spatial multiplexing in multiuser MIMO channels,” IEEE Transactions on Signal Processing, vol. 52, pp. 461 – 471, Feb. 2004. [8] L. U. Choi and R. D. Murch, “A transmit preprocessing technique for multiuser mimo systems using a decomposition approach,” IEEE Trans. Wireless Commun., vol. 3, no. 1, pp. 20–24, Jan. 2004. [9] Jun Zhang, Runhua Chen, J. G. Andrews and R. W. Heath, “Coordinated multi-cell MIMO systems with cellular block diagonalization,” Proc.41st Asilomar Conference on Signals, Systems and Computers (ACSSC‟ 07), pp. 1669 – 1673, Nov. 2007. [10] Rajesh Gadiyar, John Mangan, “Using Intel Architecture for implementing SDR in Wireless Basesations”, SDRForum, SDR09‟. [11] White Paper of Distributed Service Network. China Mobile Research Institute. [12] ETSI NFV ISG (2012) Network Functions Virtualisation. [Online].
Available:
http://portal.etsi.org/portal/server.pt/community/NFV/367
[13] www.ict-ijoin.eu [14] https://www.mobile-cloud-networking.eu [15] www.ngmn.org
88
China Mobile Research Institute
© 2013 CMCC. All rights reserved.
Contact:
HUANG Jinri
DUAN Ran
Email:
[email protected]
[email protected]
China Mobile Research Institute
89