Академический Документы
Профессиональный Документы
Культура Документы
RESOURCE ALLOCATION IN
HETEROGENEOUS CLOUD RADIO ACCESS
NETWORKS: ADVANCES AND CHALLENGES
HAYSSAM DAHROUJ, AHMED DOUIK, OUSSAMA DHIFALLAH, TAREQ Y. AL-NAFFOURI,
AND MOHAMED-SLIM ALOUINI
(a) (b)
Figure 1. Heterogenous radio access networks schemes: a) heterogeneous cloud radio access network with diverse cell sizes (macro-
cell, microcell, picocell); BSs from different tiers are connected to the cloud via either optical or wireless backhaul links; b) hetero-
geneous cloud radio access network with multiple clouds; each cloud connects to a set of BSs through hybrid backhauling.
To overcome the limitations of conventional systems. HetNet deployment has a direct effect
HetNets, a top-down architectural change is on decreasing the propagation distance between
expected to be at the center of development of the different transmitter-receiver pairs, thereby
5G systems via connecting the BSs to central increasing the overall system power efficiency.
clouds, and also moving the networks toward However, backhaul deployment increases power
user-centric architectures. Unlike conventional consumption, so additional power adjustments
HetNet operation, in H-CRANs, entities are are needed on the level of backhaul links con-
connected to the central cloud, where most of necting the BSs from the different tiers to the
the control signaling takes place. This configura- central cloud. Power adjustment promises to be
tion allows users and machines to communicate an interesting area of future research, as high-
via several possible sets of BSs. Such a drastic lighted later in this article.
shift in the cellular paradigm in the long run Latency Management: To support real-time
may lead to deploying more BSs than the cell mobile control and applications of device-to-
phones served. This leads to several open-ended device (D2D) communications and Internet of
questions about the future of BSs as indepen- things (IoT) technology, drastic reductions in
dent heterogeneous entities, capable of manag- access time down to 1 ms round-trip latency
ing resources in ultra-dense networks. Thanks to (compared to 15 ms in 4G systems) is expected.
virtualization, the H-CRAN moves the majority Frame restructuring and cross-layer optimization
of basic network functionalities, particularly between the network and physical layers (PHY)
resource management, from the BS side to the (e.g., joint network coding and PHY designs) are
central cloud side. Such a separation between needed to provide almost zero distance connec-
the control plane (cloud) and the data plane tivity between all connected machines. Optimiz-
(BSs that are typically geographically distant) ing the backhaul links also adds an additional
changes the way network services are provided. factor to overall round-trip latency management.
It introduces several new challenges for network A common denominator among the above
modeling, analysis, and especially for new challenges is the question of how to manage the
resource allocation and optimization schemes. network resources across the different tiers and
H-CRAN design combines both BS densifica- different dimensions (space, time, frequency) by
tion and centralized processing. It reduces the choosing suitable network functionalities (user
distance between the BS and the users, leading association, beamforming, scheduling, power and
to both higher achievable data rates and efficient spectrum optimization, etc.) so as to achieve
spectrum exploitation enabled by the higher such promising metrics (rate, power, latency).
reuse of time-frequency resources across multi-
ple BSs. Such design, however, faces significant RESEARCH ON H-CRANS
challenges, mainly due to the stringent 5G By enabling joint processing at the cloud, key
requirements, which can be summarized as fol- research features become those related to the
lows: joint provisioning of resources between the back-
Massive Capacity Increase: The amount of haul links and the heterogeneous radio access
data a 5G H-CRAN serves is expected to be network. Figure 1a depicts an example of a Het-
1000-fold more than the capacity of 4G. Key tech- Net of multiple tiers (macrocells, microcells, pic-
nologies to achieve such a data deluge include, in ocells), all connected to a central cloud via
addition to densification, exploring new frequency wireless or optical fiber backhaul links. Physical
bands through mmWave spectrum allocation and properties of wired and wireless backhauls are
spectral efficiency enhancement through massive innately different, and each fits specific design
multiple-input-multiple-output (MIMO) deploy- requirements. While fiber optical backhauls are
ment. high-capacity links suitable for cells of medium
Significant Energy Savings: In 5G networks, to large sizes (micro and macrocells) [5], their
energy (measured by Joule per bit) is expected cost and scalability remain major issues, espe-
to be 100 times less than its counterpart in 4G cially in dense HetNets of hundreds or thou-
To fully exploit the bene- sands of BSs (access points). On the other hand, under which every user can associate with one
wireless backhaul offers a cheap and scalable BS only [9, references therein].
fits of the powerful cen- solution for small cell deployment [6]; however, With the fundamental architectural changes
its design performance is coupled with the natu- in H-CRANs, each user’s data is processed in
tralized cloud, joint ral latency problem of the wireless channels, and the cloud, so every user can communicate with
precoding of user signals severely depends on the conditions of the wire- several heterogeneous nodes, also known as a
less medium. In future H-CRANs, cells of differ- user-centric architecture. Each user, device, or
are typically performed ent sizes coexist, and hence both wire and machine is expected to communicate with one or
at the cloud server, wireless links are needed. Such heterogeneous many BSs at the same time. Determining the
backhauling brings in a set of new research optimal user-to-BS association, in general,
resulting in analog sig- problems, and opens several questions on how to becomes a combinatorial optimization problem
jointly account for the quantization design of the of high complexity. Exhaustive search is infeasi-
nals at the cloud that capacity limited fiber links [7] and the design of ble for any reasonably sized network, even with
are compressed and for- the transceivers of the wireless backhaul link very powerful processors at the cloud. To address
through optimally allocating its radio resources the BS association problem, simplified proce-
warded to the corre- [6]. dures have been used recently to address the
sponding BSs over the The H-CRAN scenario in Fig. 1a represents problem in cloud radio access networks (C-
a fully centralized solution, where all processing RANs) [5, 10]. In each of these references, the
backhaul links, and and control units are located at a single central user-to-BS association problem involves solving
cloud. With mass deployments of BSs, larger a sparse beamforming problem that returns
eventually transmitted areas with data traffic may be served as well by which users should be served by which BSs. The
by the BS antennas. several non-collocated clouds, where each cloud solutions in [5, 10] are inspired by results from
serves a set of BSs, as shown in Fig. 1b. In this compressive sensing, in which the original l 0 -
situation, every user can be served by a single norm minimization problem is recast as a
cloud only, and the cloud needs to distribute the reweighted l1-norm minimization, leading to effi-
user’s data to the BSs connected to that specific cient solutions to the BS association problem.
cloud only. This model is an abstraction of the It would be interesting to see such proce-
problem considered in [8], and is subject to sev- dures generalized to the multicloud situation
eral future research problems related to multi- shown in Fig. 1b. The problem would then
cloud coordination for distributed storage and involve joint cloud assignment and BS associa-
resource management. tion, as highlighted later.
This article first sheds light on advances in
the resource allocation problem in H-CRANs, JOINT PRECODING AND BACKHAUL COMPRESSION
particularly in topics related to user-to-BSs asso- The capacity limitations of the backhaul links
ciation, quantization design, and joint resource connecting the BSs to the central cloud create a
allocation and flow control. Then it proposes big challenge in the design of H-CRANs. To
promising resource allocation schemes: coordi- fully exploit the benefits of the powerful central-
nated scheduling, hybrid backhauling, and multi- ized cloud, joint precoding of user signals are
cloud association. Simulation results show how typically performed at the cloud server, resulting
the proposed strategies provide appreciable per- in analog signals at the cloud that are com-
formance improvement compared to methods pressed and forwarded to the corresponding BSs
from recent literature. The article further pre- over the backhaul links, and eventually transmit-
sents some future research topics. ted by the BS antennas.
The above approach involves both a precod-
RESOURCE ALLOCATION IN H-CRANS ing problem at the cloud and a compression
problem at the backhaul links, which introduce
The H-CRAN amalgamates both the ability of quantization noises that are a function of the
joint signal processing of data belonging to dif- backhaul capacity constraints. Joint design of the
ferent users, and strong processors capable of precoding and correlation matrices of the quan-
computing high-complexity algorithms that con- tization noises is studied in both single-cloud [7]
ventional BS processors in HetNets are inca- and multicloud [8] scenarios. Both [7, 8] propose
pable of carrying. Further, due to the separation an iterative algorithm that converges to local
between the data and control planes in H- optima using the majorization minimization
CRANs, antennas from BSs belonging to BSs approach (MM). Several compression techniques
from different tiers essentially form a virtual are additionally proposed in an uplink single-
MIMO antenna array capable of multiplexing cloud framework [11].
multiple users across different dimensions
through joint resource allocation. This section JOINT RESOURCE ALLOCATION AND
highlights the advances in optimization in BS
association and resource allocation schemes that FLOW CONTROL
are applicable to H-CRANs. Besides backhaul quantization design, traffic
engineering within the backhaul link is a critical
BASE STATION ASSOCIATION performance factor. For example, constraints
The performance of wireless network critically such as total flow rate of each backhaul link and
depends on the optimal user association, that is, per-node flow conservation constraints should be
which user roams to which BS. As conventional taken into consideration in a C-RAN’s design.
HetNets are designed under the classical cellular However, they are often neglected under the
model, BS association and cell selection termi- assumption that the data can be routed without
nologies are used interchangeably. For such sys- any cost to the backhaul network. Due to the
tems, a cell-centric architecture is assumed huge traffic demand in next generation H-
PROPOSED
RESOURCE ALLOCATION SCHEMES Cell
resource allocation problems in H-CRANs (e.g., cloud, 21 BSs, and 5 time/frequency blocks per
coordinated scheduling and power control, user BS frame. As the figure shows, coordinated
association, channel assignment, antenna selec- scheduling brings significant performance
tion). Coordinated scheduling and power control improvement compared to conventional round-
can especially be seen as a natural generalization robin scheduling. The figure especially shows
of the scheduling approach used in [13]. Such how the performance improvement in bits per
generalization is reached by using sophisticated second per Hertz of our proposed scheme
graph theoretical techniques for constructing the increases as the number of users increases. This
equivalent higher-dimensional graph. To evalu- is due to the fact that interference becomes
ate the gain harvested from coordinated schedul- higher as the number of users increases, so the
ing in a dense H-CRAN, Fig. 3 plots the role of the cloud in mitigating interference
sum-rate in bits per second per Hertz vs. the becomes more important.
number of users in a network formed by a single
HYBRID BACKHAULING
Fiber optic backhauls are relatively expensive
700 but provide high-capacity links suitable for cells
Coordinated scheduling of medium to large size [5]. Wireless backhaul,
Round-robin scheduling
650
on the other hand, offers a cheap and scalable
solution for smaller cell deployment [6]. The
diversity of cell sizes and radio access technolo-
600 gies in H-CRANs necessitates heterogeneous
Average sum rate (b/s/Hz)
By virtue of ther research follow-up on how to solve the [6] H. Dahrouj et al., “Power Spectrum Optimization for
Interference Mitigation via Iterative Function Evalua-
distributed cloud association problem and find
virtualization, H-CRAN the correlation matrix of the quantization
tion,” Proc. 1st Wksp. Distrib. Antenna Sys. for Broad-
band Mobile Commun., IEEE GLOBECOM, Houston, TX,
noise in a joint manner.
moves the majority of Dec. 2011.
[7] S. H. Park et al., “Joint Precoding and Multivariate Back-
basic network FUTURE RESEARCH haul Compression for the Downlink of Cloud Radio
Access Networks,” IEEE Trans. Signal Processing, vol.
Coordinated scheduling, hybrid backhauls, and
functionalities, particularly multicloud resource allocation trigger a set of
61, no. 22, Nov. 2013, pp. 5645–58.
[8] —, “Inter-Cluster Design of Precoding and Fronthaul Com-
resource management, other interesting problems in the field, such as pression for Cloud Radio Access Networks,” IEEE Wireless
Commun. Letters, vol. 3, no. 4, Apr. 2014, pp. 369–72.
how to solve the resource allocation problems
from base-stations to while accounting for channel estimation error
[9] K. Shen and W. Yu, “Distributed Pricing-Based User
Association for Downlink Heterogeneous Cellular Net-
and channel feedback issues (e.g., latency, over-
the central cloud. Such head). Other topics of future research are cross-
works,” IEEE JSAC, vol. 32, no. 6, 2014, pp. 1100–13.
[10] J. Zhao, T. Q. S. Quek, and Z. Lei, “Coordinated Multi-
a separation between layer optimization in H-CRANs, and finding point Transmission with Limited Backhaul Data Trans-
fer,” IEEE Trans. Wireless Commun., vol. 12, no. 6, June
optimized ways of exploiting next generation
the control and data enabling technologies (massive MIMO,
2013, pp. 2762–75.
[11] Y. Zhou and W. Yu, “Optimized Backhaul Compression
planes changes the way mmWave, D2D, and M2M) by means of offload- for Uplink Cloud Radio Access Network,” IEEE JSAC,
vol. 32, no. 6, June 2014, pp. 1295–1307.
ing and routing data among devices and
network services are machines to account for both spatial and tempo-
[12] W. C. Liao et al., “Min Flow Rate Maximization for
Software Defined Radio Access Networks,” IEEE JSAC,
ral data fluctuations, as in multihop routing.
historically provided. Important research features that need to be
vol. 32, no. 6, June 2014, pp. 1282–94.
[13] A. Douik et al., “Coordinated Scheduling for the Downlink
of Cloud Radio-Access Networks,” accepted for IEEE ICC
addressed here are how to design algorithms 2015, Arxiv e-prints, vol. abs/1411.4144, 2015.
that are partially centralized (resource manage- [14] O. Dhifallah et al., “Joint Hybrid Backhaul and Access
ment at the cloud) and distributed (data routing Links Design in Cloud-Radio Access Networks,” accept-
at the heterogeneous nodes), as a function of ed for IEEE VTC Fall-2015, Arxiv e-prints, vol.
abs/1502.07586, 2015.
both traffic distribution and the available radio [15] H. Dahrouj, T. Y. Al-Naffouri, and M.-S. Alouini, “Dis-
resources. tributed Cloud Association for Downlink Multicloud
Radio Access Networks,” Conf. Info. Science and Sys.,
CONCLUSION Arxiv e-prints, vol. abs/1503.00267, 2015.
Georgia Institute of Technology, Atlanta, in 1998, and his EURASIP Workshop on Nolinear Signal and Image Process-
Ph.D. degree in electrical engineering from Stanford Uni- ing 2001 for his work on adaptive filtering analysis; the
versity, California, in 2004. He was a visiting scholar at the IEEE Education Society Chapter Achievement award in
California Institute of Technology, Pasadena, in 2005 and 2008; and the Al-Marai Award for innovative research in
Fulbright scholar at the University of Southern California in communications in 2009. He has also been serving as an
2008. He is currently an associate professor in the Electrical Associate Editor of Transactions on Signal Processing since
Engineering Department, King Fahd University of Petroleum August 2013.
and Minerals, and jointly with the Electrical Engineering
Department, KAUST. His research interests lie in the areas MOHAMED-SLIM ALOUINI [S’94, M’98, SM’03, F’09] (slim.aloui-
of sparse, adaptive, and statistical signal processing, and ni@kaust.edu.sa) received his Ph.D. degree in electrical
their applications in wireless communications and multius- engineering from the California Institute of Technology in
er information theory. Most recently, he has been interest- 1998. He served as a faculty member at the University of
ed in compressive sensing and random matrix theory and Minnesota, Minneapolis, and then at Texas A&M University
their applications. He has over 80 publications in journals at Qatar, Education City, Doha, before joining KAUST as a
and conference proceedings, nine standards contributions, professor of electrical engineering in 2009. His current
four issued patents, four patents issued and four pending. research interests include the modeling, design, and per-
He received the Best Student Paper Award at the IEEE- formance analysis of wireless communications systems.