Cloud RAN Nextgen Architecture
Cloud RAN Nextgen Architecture
Cloud RAN Nextgen Architecture
Cloud RAN is emerging as critical to the new wireless network architecture. This white paper
investigates state-of-the-art commercial cloud RAN technology and looks ahead to new ar-
chitectures, new deployment models and new services enabled by the next-generation RAN.
Figure 1 identifies the major service categories being used to inform the development of
5G specifications, and which are driving investment in next-generation mobile networks.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 2
have very different performance requirements and traffic profiles. To serve these new
markets and increase revenues substantially, operators need highly scalable and flexible
networks. Support for these new use cases is inherent to cloud RAN development.
Operator respondents ranked system efficiency as most important driver for cloud RAN, with
47 percent ranking the ability to improve scalability and resource utilization as "very im-
portant"; this is followed by improving the economics of RAN operation, at 43 percent.
The most popular use cases driving cloud RAN deployment are shown in Figure 2: large
public venues (a combined 67 percent of operator respondents ranked this "critical" or "im-
portant" in our survey); targeted outdoor urban areas, such as public squares and shopping
streets (61 percent); and high-density urban areas (57 percent). Suburban and rural cover-
age scored poorly, indicating that cloud RAN will be used first in hotspot areas with high-
density demand.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 3
increasing to hundreds in future). Although the implementation is RAN-specific, this architec-
ture maps well to the cloud paradigm of centralized control and distributed processing.
This concept of centralized baseband and distributed radio units is well-known. Two typical
network implementations of cloud RAN are shown in Figure 3: the classic centralized RAN
architecture on the left; and the more advanced, next-generation cloud RAN that can support
multi-layer, ultra-dense operation in many different deployment scenarios on the right.
The next-generation architecture extends the cloud RAN concept to ultra-dense networks
that incorporates macro, micro and small cells in diverse spectrum bands. By scheduling
transmissions centrally, cloud RAN can improve cell edge performance by reducing inter-site
interference and can serve users across different radio access bearers. With devices able to
take advantage of inter-site connectivity and de-coupled uplink and downlink, cloud RAN in-
troduces the concept of the "no-edge" network to ensure consistently good user experience
in high-density networks.
The central controller can be designed using "cloud-native" software to run on general pur-
pose telco cloud infrastructure. This location corresponds to an edge data center facility ca-
pable of running multi-access edge computing (MEC) services. One opportunity is to con-
sider how radio access and cloud-based MEC services can be integrated – for example, to
enable developers and content providers to optimize their service according to the desired,
or available, radio bandwidth. This is expected to be useful for ultra-low-latency and mis-
sion-critical services in 5G.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 4
from 2019 onward. The technical report "Study on New Radio Access Technology; Radio Ac-
cess Architecture and Interfaces" (TR 38.801) sets out various options for the RAN and its
interfaces to the core network.
In the next-generation cloud RAN architecture, real-time (RT) functions are deployed at the
antenna site to manage air interface resources, while non-real-time (NRT) control functions
are hosted centrally to coordinate transmissions across the coverage area. In 5G, this is be-
ing formalized with the central unit (CU) and distributed unit (DU) functional split. This func-
tional architecture is now "native" to the 3GPP specification.
The appropriate CU/DU split between RT and NRT functions is important in 5G for several
reasons. On the DU side, for example, the use of massive multiple input/multiple output
(MIMO) and associated beam-tracking and beam-switching techniques require highly accu-
rate time estimation to support mobility, which points toward distributed real-time radio
functions as optimal.
Elsewhere, centralized functionality makes more sense. The 5G RAN will consist of overlap-
ping cells with multiple connectivity to the device and the network – for example, C-Band
small cells and sub 6GHz macro cells will combine to provide the end-user service. This will
need some kind of access network selection function that hosted in the CU by the central
controller – i.e., in the NRT part of a cloud RAN architecture – that has view of the overall
network state.
The following sections of this paper have more detail on how to introduce 5G NR into existing
networks, with cloud RAN technology using dual connectivity and fast user-plane switching
to manage multiple radio access bearers per device.
Operations, administration and management (OA&M) automation is vital because of the sig-
nificant operational cost dedicated to ongoing RAN optimization. To rapidly expand capacity –
using a small cell underlay, for example – typically requires of expensive human skillsets to
solve laborious problems. Automating these procedures can impact the cost of production
and operator profitability.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 5
ARCHITECTURE PRINCIPALS: RT & NRT SPLIT
Implementation of the cloud RAN architecture – and the subsequent deployment in the net-
work – depends on the functional split between distributed radio and centralized control.
There are several options for this in 4G and 5G; the main ones are shown in Figure 4.
Excluding Common Public Radio Interface (CPRI), a Packet Data Convergence Protocol
(PDCP) layer functional split has the most support in the industry today. Several additional
splits are proposed in the technical study on RAN architecture for 5G (TR 38.801), with con-
sensus again going toward a PDCP or PDCP/upper radio link control (RLC) split as the opti-
mal trade-off between flexibility and performance.
To the left of Figure 4, only the radio frequency (RF) module – which includes filters, power
amplifiers, digital-to-analog converters (DACs), etc., is distributed, while all the digital pro-
cessing is centralized. This model enables coordination at Layer 1. Moving to the right, pro-
gressively more functions are distributed, to the point where only the PDCP layer is central-
ized, which allows for Layer 3 coordination.
In principle, the split to the left of the chart, where the entire protocol stack is centrally con-
trolled, offers the greatest performance gains from coordination across the coverage area.
This requires very low-latency, high-bandwidth connection between the distributed unit and
centralized unit. Today, the CPRI protocol is used to transport digital RF over fiber optic,
with a dedicated channel for each radio.
Moving progressively to the right, the lower-layer functions migrate to the distributed unit.
Because an increasing number of scheduling decisions are taken without central coordination,
the performance gain declines. The advantage is that the transport link is more "forgiving"
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 6
in terms of performance, and multiple radios can be multiplexed onto a carrier Ethernet or
IP service.
Essentially, there is a trade-off between system performance and the performance to the
front-haul link between CU and DU. This also impacts hardware choices, and cost, at the CU
and DU. The optimal balance is difficult to determine because it depends on technical factors
and on commercial factors that are generally market- and operator-specific. If an operator
owns lots of dark fiber and can easily connect them to radio units, a full centralized model
by be appropriate. If the operator uses a lot of microwave to the cell site, a less centralized
cloud RAN may be better. In practice, multiple models are likely to prevail globally.
There is potential to introduce some very interesting products – for example, using active
antenna systems with integrated radio unit – with this functional split. As has been shown in
the handset market, tight integration of the radio components can reduce cost and power
consumption very significantly. Note also that these radios can still be software configurable
even though they do not use general purpose processors.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 7
the service is not time-critical, it may be more efficient to process data centrally. This is
shown in Figure 5.
To the left of the chart, the different levels of distributed data center mapped to transport
latency. The closer to the edge, the lower the latency. The access data center is also the fa-
cility where the central unit is deployed. With under 1 ms of latency between the RAN and
application, this might be suitable for very low-latency services, such as assisted driving, or
machine automation in industrial facilities.
For services that are delay-sensitive but don't quite require real-time processing, such as
virtual reality, augmented reality, fast image processing (e.g., for facial recognition) or
venue services (e.g., for video replay at a sports stadium), the local data center may be
more appropriate from the cost performance perspective. For other applications, a large
centralized data center may be more appropriate.
This decision is expected to be settled shortly in order to enable the 3GPP to achieve its
target of a first 5G standards freeze for non-standalone (NSA) mode by the end of 2017, to
enable commercial services from 2019. 5G in standalone (SA) mode, with a new core network
and without dependencies on LTE, is scheduled to freeze, just six months later in mid-2018,
to support commercial service launch from 2020 onward.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 8
commercial networks by some operators. Deploying 5G using non-standalone mode in a
cloud RAN architecture is shown in Figure 6.
In this model, the UE (the customer device) is served by both the 5G and LTE radios. Rather
than homing the 5G base station to the LTE base station, independent bearers, controlled
by the centralized unit, are used in cloud RAN. This is more efficient from a transport per-
spective and avoids the LTE base station becoming a bottleneck. Over time, this cloud RAN
architecture with centralized control enables superior performance because it can support
fast user-plane switching between radio access technologies (RATs) at the PDCP level. In a
multi-band, multi-RAT environment, this enables the cloud RAN to "orchestrate" resource
allocation across the coverage area.
© HEAVY READING | APRIL 2017 | CLOUD RAN & THE NEXT-GENERATION MOBILE NETWORK 9