Joint Energy Minimization and Resource Allocation in C-RAN with Mobile Cloud

TL;DR: In this article, the joint energy minimization and resource allocation in C-RAN with mobile cloud computing under the time constraints of the given tasks is studied, and a non-convex optimization is formulated into an equivalent convex problem based on weighted minimum mean square error (WMMSE).

Abstract: Cloud radio access network (C-RAN) has emerged as a potential candidate of the next generation access network technology to address the increasing mobile traffic, while mobile cloud computing (MCC) offers a prospective solution to the resource-limited mobile user in executing computation intensive tasks. Taking full advantages of above two cloud-based techniques, C-RAN with MCC are presented in this paper to enhance both performance and energy efficiencies. In particular, this paper studies the joint energy minimization and resource allocation in C-RAN with MCC under the time constraints of the given tasks. We first review the energy and time model of the computation and communication. Then, we formulate the joint energy minimization into a non-convex optimization with the constraints of task executing time, transmitting power, computation capacity and fronthaul data rates. This non-convex optimization is then reformulated into an equivalent convex problem based on weighted minimum mean square error (WMMSE). The iterative algorithm is finally given to deal with the joint resource allocation in C-RAN with mobile cloud. Simulation results confirm that the proposed energy minimization and resource allocation solution can improve the system performance and save energy.

Summary

Introduction

• Cloud radio access network (C-RAN) has emerged as a potential candidate of the next generation access network technology to address the increasing mobile traffic, while mobile cloud computing (MCC) offers a prospective solution to the resource-limited mobile user in executing computation intensive tasks.
• DRAFT high definition video playing and gaming appear in their daily life, make the load of both the mobile phone and the network, in terms of energy and bandwidth, ncrease hugely.
• Thus, how to save the whole system’s energy is of huge importance and interest in the operators’ eyes.
• The mathematical models for the mobile cloud computation as well as the C-RAN are presented.

A. Mobile Clone and System Architecture

• The authors have noticed that when the mobile users encounter the computational intensive or high energy required tasks, they sometimes do not want to offload those tasks into the mobile cloud, as transmitting those program data to the cloud still costs some energy [5].
• Mobile clone can be implemented by the cloud-based virtualmachine which holds the same software stack, such as operating system, middleware, applications, as the mobile user.
• The mobile user only needs to cost a small amount of energy and time overhead.
• Similar to [7] and [5], the authors assume that each of UEi has the computational intensive taskUi to be accomplished in the mobile clone as follows.

C. Network Model

• After the mobile clone completes the task execution, the results will be returned to the mobile user through C-RAN.
• The received signal at the UEi under the complex baseband equivalent channel can be written as yi = ∑ j∈C hij H vijxi +.
• The time cost in sending the execution results back to UEi from the RRHs is given by T Tri = Di ri (8) where.

D. Fronthaul Constraints

• The fronthaul link can carry the task results from the mobileclone to the UE through C-RAN.
• (12) One can see that the number of non-zeros elements of the transmitti g beamforming vector|vij| 2 also indicates the number of data symbol streams, carried bythe fronthaul link from BBU to RRH j for the i-th mobile user.
• Thej-th fronthaul constraint can be modeled as the maximum data rates which can be allowed to transmitting through BBU toj-th RRH asCj ≤ Cj,max.
• The authors also use this fronthaul constraint in their paper.

E. QoS Requirement

• The qualify of service (QoS) can be given as the whole time cost for completing the required task and returning the results back to the mobile user.
• The authors also assume that the task has to be accomplished in time constraintsTi,max in order to satisfy the mobile user’s requirement, then the QoS constraint can be given as Ti ≤ Ti,max (16) III.
• The authors provide the energy minimization problemformulation.
• First, the authors formulate the energy minimization for the mobile clone and then they give the energyminimization formulation for C-RAN with the fronthaul constraints.

B. Energy Minimization for C-RAN

• The authors assume the time constraints for transmitting the task results through C-RAN to UEi as T Tri,max.
• DRAFT the equality holds for the last constraints ofP2 and then, the minimum transmission data rate can be given by ri ≥ Di T Tri,max .
• The authors are interested in solving the energy miniization and resource allocation optimization jointly between the mobile cloud and mobile network.
• W assume that the task has to be completed in the given total time constraint, including theex cuting time plus the transmitting DRAFT time.
• In the next subsections, the authors will provide the iterative algorithms based on weighted minimum mean square error solution to deal with it.

B. WMMSE-based Solution

• The WMMSE method is introduced by [21], [22] and use to address the weighted sum rate problem.
• Similar with Fig. 5, Fig. 6 shows that the whole energy consumption of mobile cloud and C-RAN decreases either with the increase of the time constrai ts or with the decrease of the CPU cycles required by each task.

Figures

Fig. 1. A cloud radio access network with mobile cloud system.

Fig. 6. Total energy consumption vs. time constraint under different CPU cyclesFi with Di = 1000.

Fig. 7. Total energy consumption vs. CPU cycles under different T Tri,max with Di = 1000.

Fig. 3. Total energy consumption vs. CPU cycles under different Ti,max with Di = 1000.

Fig. 8. Total energy consumption vs. data size under different T Tri,max with Fi = 1500.

Fig. 4. Total energy consumption vs. data size under different Ti,max with Fi = 1500.

Fig. 2. C-RAN network withL = 4 RRHs andN = 5 UEs.

Fig. 5. Total energy consumption vs. time constraint under different data sizeDi with Fi = 1500.

Full text


                 !"#$%
&  '(        )    *)      +  ,'''
-./ 0%+..+1/"*11"+,!#/$*1#/#
2(,'''
3  .44+4#"+##"54-+!"#/+!6!!705
8.44+4#"+##"54-+!"#/+!6!!7059
-  :      ;      
.44+++44.407/554
3:(:. %
3:(<.+.(=;
(: %4.(+.
;;..(.;:.(
;;.(*;*.>..
  .  .      .:          ;  .
:(.43.+-
((+?((
;;.;.(+ -;.(
:.44+++4.+
-(@;>.:;
:..+-4;
.: ; .  : .<   .
(A+%

IEEE TRANSACTIONS, VOL. X, NO. X, XXX XXXX 1
Joint Energy Minimization and Resource
Allocation in C-RAN with Mobile Cloud
Kezhi Wang, Kun Yang, Senior Member, IEEE, and Chathura Sarathchandra Magurawalage
Abstract
Cloud radio access network (C-RAN) has emerged as a potential candidate of the next generation
access network technology to address the increasing mobile traffic, while mobile cloud computing (MCC)
offers a prospective solution to the resource-limited mobile user in executing computation intensive tasks.
Taking full advantages of above two cloud-based techniques, C-RAN with MCC systems are presented
in this paper to enhance both performance and energy efficiencies. In particular, this paper studies the
joint energy minimization and resource allocation in C-RAN with MMC under the time constraints of
the given tasks. We first give the computational model and network model with the energy and time
cost. Then, we formulate the joint energy minimization into a non-convex optimization with fronthaul
rate constraints. An equivalent convex feasibility problem is reformulated and the iterative algorithm
based on weighted minimum mean square error (WMMSE) is given to deal with the non-convexity.
Simulation results confirm that the proposed energy minimization and resource allocation solution can
improve the system performance and save energy.
Intex Terms - C-RAN, Joint Energy Minimization, MCC, Resource Allocation.
I. INTRODUCTION
Nowadays, the number of smart devices and the corresponding mobile traffic have grown
rapidly, which poses an increasingly high burden on the existing cellular network. It is predicted
that the mobile device traffic will increase one thousand times and the cost is expected to
decrease one hundred times by 2020, with the help of new network and computation paradigm [1].
Recently, more and more computational resource intensive tasks, such as multimedia applications,
Kezhi Wang, Kun Yang and Chathura Sarathchandra Magurawalage are with the School of Computer Sciences and Electrical
Engineering, University of Essex, CO3 4HG, Colchester, U.K. (e-mails: {kwangj, kunyang, csarata}@essex.ac.uk).
DRAFT

2 IEEE TRANSACTIONS, VOL. X, NO. X, XXX XXXX
high definition video playing and gaming appear in our daily life, make the load of both the
mobile phone and the network, in terms of energy and bandwidth, increase hugely. Also, those
types of applications have the trend of attracting more attention from the smartphone users.
In traditional cellular networks, each base station (BS) transmits data signal separately to
the UE, so that the energy cost in the BS will be very high, in order to overcome the path
loss and the interference from other BSs. Coordinated Multi-Point (CoMP) technique has been
proposed to mitigate interference by using cooperation techniques, such as joint transmission
(JT) and coordinated beamforming (CBF), between different BSs. However, CoMP technique
sometimes cannot achieve the best performance, due to traditional X2 interface limitation, i.e.,
low bandwidth, high latency and inaccurate synchronization.
It is very fortunate that recently, a new promising network infrastructure, i.e., cloud radio
access network (C-RAN), has been presented and soon received a large amount of attention
in both academia and industry [2]. C-RAN is a cloud computing based, centralized, clean and
collaborative radio access network [3]. It divides the traditional BS into three parts, namely,
serval remote radio heads (RRHs), the baseband unit (BBU) pool, and the high-bandwidth,
high-speed, low latency fiber transport (or fronthaul) link connecting RRH to the BBU cloud
pool. In C-RAN, most of the intensive network computational tasks, such as baseband signal
processing, precoding matrix calculation, channel state information estimation are moved to
BBU pool in the cloud, which is composed of numerous software defined virtual machines with
the feature of dynamically configurable, scalable, sharable, re-allocatable per demand. On the
other hand, RRH only needs to up-covert the received baseband signal from the BBU cloud
and transmit them in the RF frequency band. In this case, RRHs with limited functions, only
including A/D, D/A conversion, amplification, frequency conversion, make them very easy to
distribute, according to the network requirement. Thanks to the separation of BBU and RRH
and the cooperation between different BBUs, significant performance gain can be achieved in
terms of efficient interference cancellation and management as well as the increase of network
capacity and decrease of the energy cost.
Another very impressive technique, i.e. mobile cloud computing (MCC) has also attracted
a huge number of interest recently [4], [5]. MCC is inspired by integrating the popular cloud
computing into mobile environment, which enables that mobile user with increasing computing
demands but limited computing resource can offload tasks to the powerful platforms in the cloud.
DRAFT

SHELL et al.: BARE DEMO OF IEEETRAN.CLS FOR JOURNALS 3
The reference [5] has investigated if the offloading operation to the cloud can save energy and
extend battery lifetimes for UEs. The reference [6] has provided a theoretical framework of
energy optimal mobile cloud computing under stochastic wireless channel while the reference
[7] has proposed a game theoretical approach for achieving efficient computation offloading
for MCC. Although the cloud computing has demonstrated the potential ability to improve the
performance, in not only the MCC, but also C-RAN, the research of integration between them
is rarely less. Fortunately, [8], [9] have shown that the combination of MCC and C-RAN is of
huge interest.
Also, pursuing computational intensive or high bandwidth tasks in the UE side increases the
operating expense and capital expenditure of the mobile operators, which drastically reduce their
profit and make them face a very hard situation. It has been shown that the energy overhead or
the electricity cost are among the most important factors in the overall operational expenditure
[10]. Thus, how to save the whole system’s energy is of huge importance and interest in the
operators’ eyes.
To address the above-mentioned questions, in this paper, we propose a novel C-RAN structure
with the mobile clone (virtual machine) co-located with the BBU in the cloud pool. The mobile
clone is responsible for the computational intensive task while the BBU is in charge of returning
the execution results to the UE via RRHs. We aim to jointly reduce the total energy cost under
the time constraints of the given task in C-RAN and mobile cloud. In particular, we model
the energy cost in executing the task in mobile cloud and the energy cost in transmitting the
results back to UE through RRHs. We also model the time spent in the mobile cloud and in
wireless transmission process. Then we formulate the joint energy minimization into a non-
convex optimization, which is NP-hard. By converting it to the equivalent weighted mean square
error (WMMSE) and using the iterative algorithm, we can successfully address the joint resource
allocation between the mobile cloud and C-RAN and deal with beamforming vector design in
RRHs.
The remainder of this paper is organized as follows. Section II introduces the system model
including the mobile cloud computational model and the network model. Section III presents
the optimization problem formulation as well as two separate energy minimization solutions in
mobile cloud and C-RAN, while Section IV introduces the joint energy minimization in mobile
cloud and mobile network. Simulation results are shown in Section V, followed by concluding
DRAFT

4 IEEE TRANSACTIONS, VOL. X, NO. X, XXX XXXX
remarks in Section VI.
II. SYSTEM MODEL
In this section, the mathematical models for the mobile cloud computation as well as the
C-RAN are presented. First, we introduce the concept of the mobile clone in MCC and the
whole system design, and then we describe the computation models, including the energy and
time consumption model in the cloud and in the network. Finally, the QoS requirement is given
through the time constraint of the given task in the last subsection.
A. Mobile Clone and System Architecture
We have noticed that when the mobile users encounter the computational intensive or high
energy required tasks, they sometimes do not want to offload those tasks into the mobile cloud,
as transmitting those program data to the cloud still costs some energy [5]. In some cases, it
is even better to execute those tasks locally if transmission overhead is too high. Therefore, it
is better to have the mobile user’s computational tasks and corresponding data in the mobile
cloud first. We can give the name of those mobile cloud with the user task and data on board as
mobile clone. Mobile clone can be implemented by the cloud-based virtual machine which holds
the same software stack, such as operating system, middleware, applications, as the mobile user.
Then, if the mobile user wants to execute some task, it only needs to send the indication signal
and the corresponding user configuration information to the mobile clone (virtual machine),
which will execute those task on mobile user’s behalf. In this case, the mobile user only needs
to cost a small amount of energy and time overhead. After the task execution completion, the
mobile clone will transmit the computation result data back to the mobile user through C-RAN.
Another advantage of having mobile clone is that each mobile clone can talk to each other in the
cloud without through the wireless link. In this case, each mobile user’s communication can be
possibly transferred into the communication between the mobile clones, thereby saving a great
number of the wireless network resources as well as the energy and time overhead.
In this paper, we consider there are N = {1, 2, ..., N} UEs, each with one antenna, deployed
in the C-RAN. Also, we consider there are L = {1, 2, ..., L} RRHs, each of which has K ≥ 1
antennas, connecting to the BBU pool through high-speed fiber fronthaul link, as shown in Fig. 1.
We consider the case that each mobile user already has one specific mobile clone, established in
DRAFT

Frequently asked questions

Q1. What are the contributions mentioned in the paper "Joint energy minimization and resource allocation in c-ran with mobile cloud" ?

Taking full advantages of above two cloud-based techniques, C-RAN with MCC systems are presented in this paper to enhance both performance and energy efficiencies. In particular, this paper studies the joint energy minimization and resource allocation in C-RAN with MMC under the time constraints of the given tasks. 

Q2. What is the energy consumption of the mobile clone and the C-RAN?

with the increase of the time constraint, the total energy decrease, as the mobile clone and the C-RAN can have more time to complete the task and return the result to the mobile user. 

Q3. What is the advantage of having a mobile clone?

Mobile clone can be implemented by the cloud-based virtual machine which holds the same software stack, such as operating system, middleware, applications, as the mobile user. 

Q4. What is the main advantage of having a mobile clone?

if the mobile user wants to execute some task, it only needs to send the indication signal and the corresponding user configuration information to the mobile clone (virtual machine), which will execute those task on mobile user’s behalf. 

Q5. What is the energy cost of the RRH to send the task to the UE?

(15)The authors also assume that the task has to be accomplished in time constraints Ti,max in order to satisfy the mobile user’s requirement, then the QoS constraint can be given asTi ≤ Ti,max (16)Also, the whole energy cost in executing this task and transiting the results back to i-th UEcan be given asEi = E C i + ηiE Tr i(17)where ηi ≥ 0 is a weight to trade off between the energy consumptions in the mobile cloud and the C-RAN, and it can be also explained as the inefficiency coefficient of the power amplifier at RRH. 

Q6. What is the power to send this task by RRHs?

the authors can assume the power to send this task by RRHs is pi, then the energy consumed by the serving RRHs isETri = pi · T Tr i = piDi ri(9)where pi can be given as pi = ∑ j∈C |vij| 2. 

Q7. What is the energy minimization problem for the mobile clone?

DRAFTThe authors assume the time constraints for completing the task in mobile clone as TCi,max, then theenergy minimization optimization problem for the mobile clone can be given asP1 : minimize fCiN ∑i=1ECisubject to TCi ≤ T C i,max, f C i ≤ f C i,max, i = 1, 2, ..., N.(18)Assume fC ∗ i as the optimum solution for problem P1. 

Q8. how can the authors solve the transmit beamforming problem?

by fixing the transmit beamforming vector vi and the MMSE receiver ui, the corresponding optimal MSE weight φi can be given byφi = ∂τ(ei)∂ei =Diκ C i (ν C i − 1) log(2)(BiFi log(ei) BiTi,max log(ei)+Di log(2))νCiBiei log 2(ei)+ Bitiei log(2) .(43)Then, by fixing the optimal MSE weight φi and MMSE receiver ui, the optimal transmitbeamforming vector vi can be calculated by solving the following SOCP problem asP9 : minimize ri,vij,CN ∑i=1φi · ei + βi(vi)subject to : Constraints of (P6).(44)DRAFTThus, the authors can deal with the overall optimization problem with WMMSE-based iterative methodas in Algorithm 2, where Z(n) = ∑N i=1 αi(r (n) i ) + t (n) i and ε is a small constant to guarantee convergence. 

Q9. What is the possible rate of interference in the UE i?

the signal-to-interference-plus-noise ratio (SINR) can be expressed bySINRi = | ∑ j∈C vij H hij|2∑N k 6=i | ∑ j∈C vkj Hhkj|2 + σ2, i = 1, 2, ..., N. (6)Then, the system capacity and the achievable rate for UE i can be given asri = Bilog (1 + SINRi) , i = 1, 2, ..., N (7)DRAFTwhere