By Frank Rayal, Xona Partners, and Joe Madden, Mobile Experts
Frank Rayal, left, and Joe Madden
When LTE was first conceived in the middle of the last decade, the focus was on providing a flat IP architecture optimized for broadband connectivity. This included placing the air interface protocol functions at the base station and using wide frequency bandwidth with multiple-antenna technology that OFDM technology enables in silicon at relatively low cost. It is an approach that reversed centralization of high-layer functions, such as scheduling, in 3G networks.
Such decisions were taken just prior to the explosion in demand mobile data services initiated by the iPhone. In those days, heterogeneous networks (HetNets) were not at the forefront of the agenda. As a result, the performance of early small cells has not been consistent, and varied from stellar success to downright damaging. Improving the performance of small cells through interference coordination progressed through latter releases of LTE, but the gain in performance remains closely tied to centralization of baseband processing. Consequently, centralization is back into consideration, driven not only by the need to improve performance, but also by pragmatic and strategic motives that are embodied in Cloud RAN.
Cloud RAN is based on two tenets: centralization and virtualization of base station baseband processing. While centralization is a mean to improving performance, the driver for centralization is very pragmatic: operational cost reduction. About two thirds of a network's cost of ownership is in operational costs including site rental, energy, and support and maintenance expenses. Operators in Korea, Japan and China have demonstrated between 30% - 50% opex reduction, which is a strong motive for centralizing the radio access network. Strictly speaking, only the antenna is needed at the cell site, so why distribute equipment to the edge if it is cheaper to centralize the radio access network?
On the other hand, the second tenet of Cloud RAN, virtualization, aims at reducing capital expenditures by applying network function virtualization (NFV) to the radio access network. Using commercial servers for base station hardware allows operators to leverage the economies of scale of the IT industry. Wireless operators know this well as it can be witnessed by programs like Domain 2.0 at AT&T and UNICA at Telefonica. Also, operators realize further cost reduction from pooling and virtualizing baseband processing because we no longer need to provision for peak capacity on a per-site basis: it is possible to reduce processing requirements by as much as 75%. But virtualization has another extremely attractive feature: imagine that you can switch between base station vendors at the click of a button! While this may be a utopian view of virtualization, it helps to imagine the possibilities and power virtualization provide. We now have a glimpse of the strategic reason for advancing Cloud RAN, so why use dedicated hardware if we can use commercial servers running what software the operator desires?
Given the great advantages that Cloud RAN brings, one wonders what's the holdup? In fact, there are two issues to resolve. The first issue is fronthaul which is an impediment towards centralization. Fronthaul is the link between the baseband units and the remote radios. Fronthaul capacity, delay and synchronization requirements are stringent. A single 20 MHz 2x2 MIMO LTE channel requires about 2.5 Gbps. This capacity adds up quickly when tens even hundreds of baseband modules are collocated in a data center. A fiber network is required to connect the sites to the data centers which not all operators possess. The cost of fiber breaks the business case for operators without their own fiber assets. But operators recognize the need for fiber and there has been a strong drive by many operators to acquire and build up their fiber assets.
The second challenge centers on virtualization of the physical layer (PHY) which involves real-time processes and high computational load functions. General purpose processors are less efficient in running these functions for commercially scalable networks than dedicated processors resulting in high power consumption. Dedicated processors can have as much at 10x the performance per Watt of general purpose processors for PHY functions. A possible solution is to offload real-time and computationally expensive functions to accelerators. But there are other solutions.
The fronthaul and virtualization challenges are coupled in a manner that a solution to both can be arrived at by judiciously choosing the functional split between centralized and distributed functions. For example, it is possible to place the physical layer at the remote radio while placing higher layers in the data center. This significantly reduces fronthaul capacity and timing requirements, and consequently the cost of fronthaul, but also reduces the performance gain over traditional distributed architecture. Since it is possible to draw the line between distributed and centralized functions along different points in the protocol stack, we are set to witness a proliferation of Cloud RAN implementations (expect to hear more about 'What is Cloud RAN'!).
While the industry works at resolving the above challenges in Cloud RAN, small Cloud RAN systems are set to emerge. Such systems would be targeted at venues, for example, a stadium or a convention center, and deployed in a similar way to distributed antenna systems (DAS). More than eight different solutions are in development with various vendors, so the hype cycle is beginning. Virtualization could spread at the 'micro' level, while the technology and economic challenges are resolved for the 'macro' level (wide-area Cloud RAN networks). Small Cloud RAN systems could eventually disrupt the wireless industry with virtualization moving up into ever larger deployments – a familiar path taken by disruptive technologies in other fields ranging from steel to excavation equipment.
Another indicator of the promise of Cloud RAN is manifested in the current thinking around 5G networks. While capacity was a major requirement for 4G, 5G complements this with emphasis on scalability and energy efficiency (as much as 90% of power consumption of mobile operators is due to the radio access network). HetNets are a central feature of 5G networks which imposes requirements for simplification of all aspects of network operation and management. These facts combined point to the importance of virtualizing the radio access network and formulating a Cloud RAN architecture and deployment scenarios that leverage the benefits of Cloud RAN.
The trends in wireless network evolutionindicate that we are currently at the cusp of bifurcation in network architecture and technologies, and even commercial practices. Operational processes remain rigidly entrenched. Operators will have to look for new means to meet the demands placed upon them by customers and investors more efficiently and effectively. Cloud RAN has the elements to provide flexibility necessary in future wireless networks, and in the indoor market, emerging solutions may break new ground toward low-cost, high density mobile infrastructure.
Frank Rayal is Partner at Xona Partners. He advises investment firms, vendors, and operators on wireless technology and business strategy, spectrum, competitive positioning, and market analytics.
Joe Madden is Principal Analyst at Mobile Experts LLC. Mobile Experts is a network of market and technology experts that provide market analysis on the mobile infrastructure and mobile handset markets. He provides market forecasts for handset, DAS, small cell, and base station markets, with in-depth research down to the nitty gritty details of frequency bands and power levels. Mr. Madden graduated, cum laude, from UCLA in 1989 and is a Silicon Valley veteran. He has survived IPOs, LBOs, divestitures, acquistions, and mergers during his 24 years in mobile communications.