Content uploaded by Yu Peng
Author content
All content in this area was uploaded by Yu Peng on Mar 22, 2018
Content may be subject to copyright.
User Association Based Cooperative
Energy-Saving Mechanism in Heterogeneous
5G Access Networks
Fei Zheng, Wenjing Li *, Peng Yu, Luoming Meng
State Key Laboratory of Networking and Switching Technology
Beijing University of Posts and Telecommunications, Beijing, China
Email: {zhengfei, wjli, yupeng, lmmeng}@bupt.edu.cn
Abstract—To reduce the energy consumption in
heterogeneous 5G access networks, we propose a
cooperative energy-saving mechanism based on user
association (UA-CESM). Joint processing (JP) and
centralized resource scheduling are introduced into the
mechanism to solve inadequate coverage problem caused by
base-station (BS) dormancy by means of cooperative BS
selection and BS state control. First, the number of
candidate dormant BSs is obtained on the basis of statistical
traffic valley. Second, for heterogeneous cellular networks
with relays, a BS selection model is designed to find
cooperative BS set and a candidate dormant BS group.
Finally, an energy-saving optimization strategy based on
bipartite graph is proposed to realize user association from
candidate dormant BSs to appropriate cooperative BSs.
The simulation results show that the average rate of edge
users is increased by 92.5%, and the coverage can be
compensated without extra transmission power in the
proposed mechanism.
Key words—5G access networks;energy-consumption
mechanism ; joint processing; bipartite graph; user
association
I. INTRODUCTION
The energy consumption management for 5G system
is currently a significant issue. Especially, the challenge
is how new network structure and new communication
technology can be applied to base-station (BS) dormancy
[1]. 5G targets excellent performance, not only in terms
of higher data rates and lower latency, but also in terms
of network intelligence and capillarity. 5G networks
resort to solutions as small cell deployment, coordinated
multiple point (CoMP, ICIC) and centralized radio
access network (C-RAN) with baseband units (BBUs)
pool [2].
Appropriately handing users over to active BSs is the
basis of BS dormancy strategy. Association between BSs
and users is described with graph theory, and the
association problem is a binary integer problem (BIP) [3].
Coverage capacity of a BS, however, actually constrains
the association. In [4], though a cell can cover neighbors
with extra transmission power, this method would
introduce new interferences into networks. The resource
centralized scheduling is hardly applied in energy-saving
mechanism although it can improve energy efficiency [5].
For macro BS networks, CoMP between BSs can expand
the coverage of cells [6,7]. For heterogeneous networks
with relays, CoMP between BSs and relays can improve
the performance of users in the dormant cells [8,9].
For heterogeneous 5G networks with relays, we
propose a user association based cooperative
energy-saving mechanism (UA-CESM). The main
contributions of this paper are presented as follows.
zEstimating the number of candidate dormant BS on
the basis of valley value of statistic traffic load.
zDesigning a BS selection model to determine the
cooperative BSs and candidate dormant BSs.
zProposing a bipartite-graph-based energy-saving
optimization strategy for user association from
candidate dormant BSs to appropriate cooperative
BSs. The problem is formulated to a BIP.
II. SYSTEM MODEL
A. Network Topology and User Traffic
BSs with MIMO antennas are located in the centers of
cells and relays are deployed at the corners of cells [9].
Relays always connect with the closest BSs [10]. The
transmission powers of BS and relay are set as constants.
The path loss budget is L(d) = 34 + 40lg(d). A typical
heterogeneous 5G network with relays is shown as Fig.1.
Due to centralized resource scheduling,
time-frequency resource is enough for BSs and the
number of served users indicates traffic load of the
service BS. Suppose one user only occupies one unit
resource, the capacity of a BS is the maximum channel
number Mmax.
One BS can directly serve users and can indirectly
serve users via relays. Traffic load is the total number of
users, which the BS directly and indirectly serves.
BBU POOL
BS
BS
BS
BS
BS
BS
BS
BS
Relay
Fig.1: Network topology
B. Affinity between BS and User
The work presented in this paper has been supported by
National High Technology Research and Development Program of
China Project (No. 2015AA01A705) and National Natural Science
Foundation of China (61271187).
2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom)
and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData)
978-1-5090-5880-8/16 $31.00 © 2016 IEEE
DOI 10.1109/iThings-GreenCom-CPSCom-SmartData.2016.161
764
2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom)
and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData)
978-1-5090-5880-8/16 $31.00 © 2016 IEEE
DOI 10.1109/iThings-GreenCom-CPSCom-SmartData.2016.161
765
2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom)
and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData)
978-1-5090-5880-8/16 $31.00 © 2016 IEEE
DOI 10.1109/iThings-GreenCom-CPSCom-SmartData.2016.161
765
2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom)
and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData)
978-1-5090-5880-8/16 $31.00 © 2016 IEEE
DOI 10.1109/iThings-GreenCom-CPSCom-SmartData.2016.161
765
2016 IEEE International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom)
and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData)
978-1-5090-5880-8/16 $31.00 © 2016 IEEE
DOI 10.1109/iThings-GreenCom-CPSCom-SmartData.2016.161
765
In Fig.2, if one user can directly or indirectly
communicate with several BSs, we define the
relationship as affinity. If one user directly or indirectly
accesses one BS, we define the relationship as
association.
Fig.2: Affinity and association
Let
Af
GSU
be the bipartite, undirected affinity
graph between user set Uand BS set S. The affinity
graph can be transform into a group of sub-graph
between each S and u.
Af
:,uUG gSu
(1)
Let
As GSU
be the bipartite, undirected
association graph, which meets the following constraints.
As Af
max
:de g ree 1
:degree
d
GUS GUS
uU u
sS s M
(2)
C. Compensation Technology
We introduce two JP modes in the scene. One mode is
BS-Relay between a relay and its host BS. Another is
BS-BS between BSs, which is the main JP mode for
compensation. Because a user selects NCcooperative BSs
with favorable links in BS-BS mode, the user generates
NCtimes traffic load.
III. USER ASSOCIATION BASED
COOPERATIVE ENERGY-SAVING
MECHANISM
A. Number of Candidate Dormant BS
We estimate the number of candidate dormant BS on
the basis of value of statistic traffic load. Let the load
normalized, the number of candidate dormant BS s meets
the following inequation.
BS off on idle C off idle
BS on off
Con
..
°
®
°
¯
t
d
NN NTNNT
NNN
st NN
(3)
Where NBS is the number of BSs, Non and Noff is the
number of active BSs and the number of candidate
dormant BSs respectively. Tidle is the valley value of
traffic load at idle time. NonTidle and NCNoffTidle is the load
of active BSs and the cooperative load derived from the
load of candidate dormant BSs respectively. We
formulate Noff as follows.
idle
off
Cidle
of
BS
BfCS
1,
11
T
NN NT
NNN
«»
«»
«»
¬¼
d
(4)
B. Selecting Dormant BS
Let the network cover an area Z,
,
U
xy
and
,
O
xy
are the load density and the value of SINR at
the location (x,y). We define the regional weighted
SINR with load density as follows.
,
,,
UO
/ ;
³³
,
U
,
³³
xy Z
xy xydxdy
Z
(5)
Where
;Z
indicates the area of Z.
We need to find the optimal cooperative BS set with
the greatest
/
among several potential combinations.
Let
^`
BS
1,,
`
BS
,,
iN
Ssss
and
^`
1
,,
`
,,
kK
Rrrr
be the BS
set and the relay set respectively. We select Noff
candidate dormant BSs to be switched off at idle time.
Meanwhile, the other Non BSs and K relays compensate
for network coverage. A new BS set is defined as.
^`
^`
on
1
C
1
,,
where ,,
°
®
°
¯
^
`
1
C
,,
C
`
on
,
lL
l
llll
jN
QQQ
QS
Qsss
(6)
Where Ql is a combination of Non BSs in S, and
on
§·
¨¸
©¹
n
LN
is the total number of combinations. All
combinations compose a set
C
C
.
O
ll
QSQ
is the
corresponding combination of candidate dormant BSs.
At (x, y),
^`
on
1
,,
`
on
,
lll
jN
ppp
is the descending set of
received power from Ql and
^`
1
,,
`
,
kK
pp p
is the set
of received power from R. Users need to select the
cooperation service of BS-BS mode or that of BS-Relay
mode.
If BS-BS, then SINR is
C
on
C
1
BB
2
1
,
,
,
O
V
¦
¦
N
l
j
j
N
l
j
jN
pxy
xy
pxy
(7)
Where
2
2
is power spectral density of white noise. The
useful signals power is the sum of top NC received power,
and the interference power is the sum of others.
If BS-Relay, then SINR is
^`
on
on
R
BR
2R
1
R
1
,max
,
,,
O
V
¦
`
on
,
ll
kk
N
kll
kj
j
llll
kjN
pp
xy
ppxy
pppp
(8)
Rl
k
p
is the received power from the host BS of rk. The
useful signals come from a relay and its host BS.
BB BR
,max ,, ,
OOO
xy xy xy
(9)
Then the
/l
is
,
,,
,
1,
UO
/ ;
³³
,
U
,
³³
xy Z
xy xydxdy
lZ
lL
(10)
Finally, we research the greatest
/l
as follows.
765766766766766
*
*
max , 1,
arg max , 1,
/ /
/
l
l
ll L
lllL
(11)
The combination Ql* is just the optimal cooperative set
SCof BSs.
C. Bipartite-Graph-Based Energy-Saving Optimization
Strategy
1) Modeling problem
Users are classified into a set of to-be-compensated
users (CU)
^`
1
,
`
m
Vvv
and another set of normal
users U-V. Let
Af C
GS V
and
As C
GS V
are
affinity graph and association graph between V and SC
respectively. With (1), GAf is represented as follow.
Af C
,
jj
j
GgSv
(12)
Then we re-sorted SC in accordance with the
descending sequence of received power of vj.
'''
,
jjj
j
GgSv
(13)
The edge set in sub-graph
'
j
g
is
^`
'''
on
''
,1,,
`
on
,
jjkjkj
jk j
Eg e s v k N
sS
(14)
The energy-saving problem (ESP) becomes to find a
cooperative association graph GSE.
Fig.3: Super user associated with candidate dormant BS
Suppose a super user uO can only be associated with
candidate dormant BSs. Its traffic is enough to make any
BS full load, and it can be associated with several BSs at
the same time. As shown in Fig.3, v should be associated
with NC cooperative BSs which have been re-sorted in
accordance with the descending sequence of received
power of v.Then uO should be associated with the empty
candidate dormant BS.
We set the weight of edges as follows.
z
As
weight 1, , eeGSV
(15)
z
'
1
weight 1
.. 1
W
WW
W
!
°
®
°
¯
¦
1
jk jk
jk jk
jk
jk
e
st
(16)
W
jk
lets an user always select cooperative BSs with
greatest power, but it does not increase the load.
z
As C O
max
weight , , MGSeue
(17)
Let
SE
FG
is the total weight of GSE, and
maximizing
SE
FG
just is the solution of ESP.
^`
SE As C O
max
SE
C
max weight
,,
.. degree
ESP : ma
1,
deg e
x
re
°
°
®
°
°
¯
d
¦
j
i
e
eG SV G Su
st v
FG
N
sM
(18)
For any GSE,
SE
FG
can be formulated as follows.
As CO
''
SE
,,
'
weight weight
weight
¦¦
¦
jk
eG SV eSu
jk
eG
FG e e
e
(19)
With the weight value of edges, it is represented
As CO
''
SE
max
,,
1
1
W
¦¦
¦
jk
eG SV eSu
jk
eG
FG M
(20)
Maximizing
CO
max
,
¦
eSu
M
means the maximum
number of candidate dormant BS associated with uO.
Let
u
ªº
¬¼
ij nm
Aa
be an association matrix between S
and V.If
As '
,
ij
esv G G
,then
^`
0,1
ij
a
,
otherwise
0
ij
a
. CUs are only associated with host
BSs or cooperative BSs.
[]
ij n m
Ww
u
is a weight matrix
of edges between S and V. If
Af '
,
ij
esv G G
, then
0
ij
w
, otherwise
ij
w
subjects to the above weight set.
OO
ªº
¬¼
in
Aa
is an association vector between S and uO. If
OAfCO
,,
i
esu G S u
, then
^`
O
0,1
i
a
,otherwise
O
0
i
a
.
OO
[]
in
Ww
is a weight vector between Sand
uO. If
CO
,eSu
, then
O
0
i
w
, otherwise
O
max
i
wM
. We transform ESP into BIP.
T
OO T
1
O11
max
T2 2
T
O33
BIP : max
..
H
H
H
d
°
°
®
°d
°
¯
¦
m
jj
j
WA WA
MA AI
st A I
AI
(21)
Where
j
A
and
j
W
j
W
are j-th vector of Aand W
respectively.
>@
11m
I
,
>@
23
1n
II
,
>@
1
max
H
in
MM
,Mi is the number of normal users
associated with i-th BS.
^`
22
C
1,
HH
ªº
¬¼
jn
N
ˈ
3
off
H
N
.
IV. SIMULATION AND ANALYSIS
We simulate UA-CESM with parameters in Table I.
In Figs. 4(a), 4(b) and 4(c), we use the selection model
to get the optimal set of cooperative BSs under the
corresponding value of traffic load. With the normal-
766767767767767
Fig. 4(a): 0.45<Tidle<0.67 and Noff=1 Fig. 4(b): 0.31<Tidle<0.45 and Noff=2 Fig. 4(c): Tidle<0.31and Noff=3
-10 010 20 30 40 50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR(dB)
CDF
Norm
JP-1
JP-2
JP-3
Fig.5: Network performance under Noff=1,
2, 3
-10 010 20 30 40 50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR(dB)
CDF
Norm
Max
3dB
JP
Fig.6: Network performance under Noff=3
while using three approaches respectively
1 2 3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Number of dormant BS
Average rate of CU
JP
3dB
Max
Fig.7: Performance of CUs under Noff=1, 2,
3 while using three approaches respectively
TABLE I: Network parameters
Parameter
Value
Parameter
Value
Transmission
power
of BS
20W
Transmission
power of Relay
2W
Shadowing
8dB
C
O
3dB
NC3
2
V
-104dBm
network(all BS are active) taken as a reference, the
overall performance reduces with increase of the number
of candidate dormant BSs in Fig.5. However, JP can
improve the performance of cell edge, the performance is
better than that of the normal network at the bottom of
curves.
As shown in Fig.6, other two compensation
approaches are comparable to JP in the scene of Fig.4(c).
Max approach permits a user to select the greatest SINR
link. And Max is the common access mode in current
networks. In fact, Max cannot expand the coverage of
BSs, so it has little compensation capacity. 3dB approach
permits all active BSs to increase 3dB transmi ssion
power. Though 3dB can expand coverage of BSs, it
introduces new interference into the network. The two
highly similar curves of Max and 3dB indicate that
interferences of adjacent cells are the major cause of
performance reduction.
For CUs, Fig.7 shows the average rate of CUs after
compensation. As a reference, the average rate of edge
users is 1 in the normal network. Average rate of JP
increases by 92.5% under Noff=1. And average rate of
3dB only increases by 4.1% because of the absence of
interference suppression. Average rate of 3dB decreases
close to that of Max with increase of the number of
candidate dormant BSs. Thus, we claim that JP is the
optimal compensation approach.
V. CONCLUSION
We propose the UA-CESM to reduce energy
consumption in heterogeneous 5G access networks. The
adequate coverage problem is solved by means of
cooperative BS selection and BS state control.
Simulation results show that the proposed mechanism is
efficient to compensate for the coverage without extra
transmission power.
REFERENCE
[1] Yapeng Wang, Xu Yang, and Liang Yang, “Dynamic CoMP
configuration for OFDMA networks under different user traffic
scenarios”, Software, Telecommunications and
Computer Networks (SoftCOM), pp.274–279, Sept. 2015.
[2] Shijie Cai, Yueling Che, Lingjie Duan, et al., “Green 5G
heterogeneous networks through dynamic small-cell operation”,
IEEE Journal on Selected Areas in Communications,
vol.34, no.5, pp.1103–1115, May 2016.
[3] Kateˇrina Dufkov´a, Milan Bjelicaz, and Byongkwon Moony,
“Energy sa vings for cellular network with evaluation of impaction
data traffic performance”, 2010 European Wireless Conference
(EW), Lucca, Italy, pp.916–923, April 2010.
[4] Chunyi Peng, Songwu Lu, and Haiyun Luo, “Green BSN:
enabling energy-proportional cellular base station networks”,
IEEE Transactions on Mobile Computing, vol.13, no.11,
pp.2537–2551, 2014.
[5] Francesco Musumeci, Camilla Bellanzon, Nicola Carapellese, et
al., “Optimal BBU placement for 5G C-RAN deployment over
WDM aggregation networks”, Journal of Lightwave Technology,
vol.34, no.8, pp.1963–1970, April, 2016.
[6] Feng Han, Zoltan Safar and K. J. Ray Liu, “Energy-efficient
base-station cooperative operation with guaranteed QoS”,IEEE
Transactions on Communications, vol.61, no.8, pp.3505–3517,
2013.
[7] Samir Kolawolé Akanni Landou and André Noll Barreto, “Use of
CoMP in 4G cellular networks for increased network energy
efficiency”, International Workshop on Telecommunications
(IWT), Santa Rita do Sapucai, Brazil, pp.1–6, June. 2015.
[8] Kazi Mohammed Saidul Huq, Shahid Mumtaz, Joanna
Bachmatiuk, et al., “Green HetNet CoMP: energy ef ficien cy
analysis and optimization”, IEEE Transactions on Vehicular
Technology, vol.64 , no.10, pp. 4670–4683, Oct. 2015.
[9] Hung Chin Jang, Po Yen Huang, “Adaptive energy sa ving
strategy for LTE-advanced networks”, 2015 Seventh International
Conference on Ubiquitous and Future Networks, Sapporo, Japan,
pp.306–310, July 2015.
[10] Zhiwei Zhang, Yunzhou Li, Kaizhi Huang, et al., “Energy
efficiency analysis of cellular networks with cooperative relays
via stochastic geometry”, China Communications, vol.12, no.9,
pp.112–121, Sept. 2015.
767768768768768