cooperative base station coloring - university of...
TRANSCRIPT
© Robert W. Heath Jr. (2015)
Cooperative Base Station Coloring Achieving dynamic clustering gain from static partitions Robert W. Heath Jr. Wireless Networking and Communications Group Department of Electrical and Computer Engineering The University of Texas at Austin http://www.profheath.org
Joint work with Jeonghun Park and Namyoon Lee
Funded by Huawei and NSF-CCF-1218338
© Robert W. Heath Jr. (2015)
2
Out-of-cluster interference is a fundamental limit
A. Lozano, R. W. Heath, and J. Andrews, “Fundamental Limits of Cooperation,” IEEE Transactions on Information Theory, 2013
SNR
Spec
tral
Effi
cien
cy
Spectral efficiency ceiling at high SNR
It is impossible to reap benefits from full coordination of all the base stations
There are some benefits to coordination at moderate SNRs
© Robert W. Heath Jr. (2015)
Static clustering inspired by backhaul network
u Predefined cooperation clusters, heavily investigated in prior work* u Value of clustering depends heavily on the locations of the users
3 K. Huang and J. G. Andrews, “An Analytical Framework for Multicell Cooperation via Stochastic Geometry and Large Deviations,” IEEE Transactions on Information Theory, 2013
J. Zhang, R. Chen, J. G. Andrews, A. Ghosh, and R. W. Heath, “Networked MIMO with Clustered Linear Precoding,” IEEE Transactions on Wireless Communications, 2009
O. Simeone, O. Somekh, H. V. Poor, and S. Shamai, “Local Base Station Cooperation Via Finite-Capacity Links for the Uplink of Linear Cellular Networks,” IEEE Transactions on Information Theory, 2009.
L user at the cluster edge J user at the cluster center
© Robert W. Heath Jr. (2015)
Dynamic clustering centered around the user
u Dynamic cooperation clusters, also investigated in prior work* u Creates a challenging scheduling problem
4
J user at the cluster center
A. Papadogiannis, D. Gesbert, and E. Hardouin, “A Dynamic Clustering Approach in Wireless Networks with Multi-Cell Cooperative Processing,” in Proc. of IEEE ICC, 2008 N. Lee, R. W. Heath Jr., D. Morales, and A. Lozano, “Base station cooperation with dynamic clustering in super-dense cloud-RAN,” in Proc. of IEEE Globecom Workshop, 2013. N. Lee, D. Morales, A. Lozano, and R. W. Heath, “Spectral Efficiency of Dynamic Coordinated Beamforming,” IEEE Transactions on Wireless Communications, 2015
J user also at the cluster center
Base station is conflicted
© Robert W. Heath Jr. (2015)
Alternative is semi-static clustering
Static clustering
5
Dynamic clustering
Semi-static clustering – multiple predefined cluster patterns
Semi-static clustering gives the benefits of dynamic clustering with low complexity
© Robert W. Heath Jr. (2015)
6
Semi-static clustering uses cluster patterns
…… …
… … …
…
…
…
…
…
…
… … …
… … …
… … …
…
… … …
… …
… … …
… … …… ……
… … …… ……
… … …… …… f1 f2 f3 f4
Different carriers assigned to each cluster
base stations
subscribers
coordination clusters
© Robert W. Heath Jr. (2015)
How are cluster patterns formed in irregular topologies?
u In practice, BS topology is irregular, patterns not obvious
u Needs framework to design and analyze cluster pattern in irregular network topologies
7
−10 −8 −6 −4 −2 0 2 4 6 8 10−10
−8
−6
−4
−2
0
2
4
6
8
10
X coordinate
Y co
ordi
nate
Base station
© Robert W. Heath Jr. (2015)
8
Graph theoretic approach for interference models
−15 −10 −5 0 5 10 15 20−20
−15
−10
−5
0
5
10
15
Delaunay triangulation (Dual graph)
Characterization of cooperation set by the graph
2nd order Voronoi (Delaunay triangulation graph)
Voronoi diagrams of line segmentsFarthest-point Voronoi diagrams
RoundnessHigher-order Voronoi diagramsComputing the farthest-point Voronoi diagramRoundness
Second order Voronoi diagram
Computational Geometry Lecture 13: More on Voronoi diagrams
V2 (d0,d1)
V2 (d0,d2)
V2 (d0,d8)
V2 (d0,d9)
V2 (d0,d10)
d0
d1
d2
d3
d4
d5d6
d7
d8
d9
d10
1st order Voronoi (frequency reuse)
−15 −10 −5 0 5 10 15 20−20
−15
−10
−5
0
5
10
15
d0
d1
d2
d3
d4
d5d6
d7
d8
d9 d10
© Robert W. Heath Jr. (2015)
u Our approach is edge coloring ª : set of edges, : set of vertices, and ª Design a coloring function : such that if and then
u Other work that uses graph coloring ª Training resource allocation for cooperative networks[Chen et al.]
ª Resource allocation for dynamic clustering[Chang et al.]
9
Avoid base station conflicts
Z. Chen, X. Hou, and C. Yang, “Training Resource Allocation for User-centric Base-station Cooperation Networks,” IEEE Transactions on Vehicular Technology, 2015. Y. Chang, Z. Tao, J. Zhang, and C. Kuo, “A Graph-Based Approach to Multi-Cell OFDMA Downlink Resource Allocation,” in Proc. IEEE Globecom, 2008. Y. Chang, Z. Tao, J. Zhang, and C. Kuo, “A Graph Approach to Dynamic Fractional Frequency Reuse (FFR) in Multi-Cell OFDMA Networks,” in Proc. IEEE ICC, 2009.
Weisstein, Eric W. "Edge Coloring." From MathWorld--A Wolfram Web Resource. http://mathworld.wolfram.com/EdgeColoring.html
fC(E) E ! {1, ..., }v 6= w (u, v) , (u,w) 2 E c (u, v) 6= c (u,w)
E V v, w, u 2 V
© Robert W. Heath Jr. (2015)
10
Use coloring to build pair-wise cooperation sets 1/3
1. Tessellate a network plane with 2nd order Voronoi cells
d0
d1 d2
d3
V2 (d0,d1) V2 (d0,d2)
V2 (d1,d3) V2 (d2,d3)
V2 (d1,d2)
V2 (d0,d3)
d0
d1 d2
d3
J. Park, N. Lee, and R. W. Heath, “Cooperative Base Station Coloring for Pair-wise Multicell Coordination,” submitted to IEEE Transactions on Communications, available on ArXiv
© Robert W. Heath Jr. (2015)
11
Use coloring to build pair-wise cooperation sets 2/3
d0
d1 d2
d3
V2 (d0,d1) V2 (d0,d2)
V2 (d1,d3) V2 (d2,d3)
V2 (d1,d2)
V2 (d0,d3)
d0
d1 d2
d3
P1
P1
P2
P2P3
P3
G = (V, E)
BSs locations
Delaunay triangulation
2. Draw a graph
3. Solve edge-coloring for the drawn graph (color = pattern)
© Robert W. Heath Jr. (2015)
12
Use coloring to build pair-wise cooperation sets 3/3
d0
d1 d2
d3
V2 (d0,d1) V2 (d0,d2)
V2 (d1,d3) V2 (d2,d3)
V2 (d1,d2)
V2 (d0,d3)
P1 :
P2 :
P3 :
4. Serve a user according to designed pattern
} Different time-frequency resources for different patterns
No conflict with edge-coloring!
d1
d0
d2
d3
© Robert W. Heath Jr. (2015)
General network application
13
−20 −15 −10 −5 0 5 10 15 20−20
−15
−10
−5
0
5
10
15
20
X coordinate
Y co
ordi
nate
Base station
−2 −1 0 1 2 3
−1
0
1
2
3
4
5
X coordinateY
coor
dina
te
Base station!!!!:!Base!sta(on!!!!!:!User!
d0
d1
d2
d3
d4
d5 1. Tessellate a network plane with 2nd order Voronoi cell Follow same
procedure
2. Draw a graph
−10 −8 −6 −4 −2 0 2 4 6 8 10−10
−8
−6
−4
−2
0
2
4
6
8
10
X coordinate
Y co
ordi
nate
Base station
−5 −4 −3 −2 −1 0 1 2 3 4 5−5
−4
−3
−2
−1
0
1
2
3
4
5
X coordinate
Y co
ordi
nate
d0
d2
d3
d4
d5
d6
d7d8
P1
P2
P3
P4
P6
P5
P1
P2
P2
P4
P7
P1 P1
P5
P2
P5
3. Solve edge-coloring for the drawn graph (color = pattern)
© Robert W. Heath Jr. (2015)
How many resources are required in the network?
u Related to number of colors for edge-coloring (chromatic index)
u Vizing’s theorem ª A simple planar graph of maximum degree Δ has chromatic index Δ or Δ+1 in general
ª Δ means the maximum number of connected edges to a vertex ª The required resources are dominated by Δ
14
Need Δ+1 time-frequency resources to cover all the clusters
© Robert W. Heath Jr. (2015)
Example for Vizing’s theorem
15
Edge Coloring
Vizing’s theorem holds! Maximum degree determines the number of colors
Maximum degree =5
Number of colors =5
Problem Only one vertex with large degrees can cause color (resource) explosion!
© Robert W. Heath Jr. (2015)
Edge-cutting algorithm
u If one vertex (BS) that has large Δ can cause resource explosion u Can improve the network performance by sacrificing a few users
16 −5 −4 −3 −2 −1 0 1 2 3 4 5−5
−4
−3
−2
−1
0
1
2
3
4
5
X coordinate
Y co
ordi
nate
d0
d2
d3
d4
d5
d6
d7d8
P1
P2
P3
P4
P6
P5
P1
P2
P2
P4
P7
P1 P1
P5
P2
P5
−5 −4 −3 −2 −1 0 1 2 3 4 5−5
−4
−3
−2
−1
0
1
2
3
4
5
X coordinate
Y co
ordi
nate
d0
d2
d3
d4
d5
d6
d7d8
P1
P2
P3
P4
P6
P5
P1
P2
P2
P4
P7
P1 P1
P5
P2
P5
Save the resource of P6
Users here cannot be protected
Edge Cutting
© Robert W. Heath Jr. (2015)
Models for analysis
u Signal model
u Intra-cluster interference management method
ª Coordinated beamforming (CBF)
17
y` = kd0
k��/2 �h`0
�TV`
0
s`0| {z }
desired signal
+ kdjk��/2 �h`j
�TV`
js`j| {z }
intra�cluster interference
+X
dv2N`\C`
kdvk��/2 �h`v
�TV`
vs`v
| {z }out�of�cluster interference
+n`
Channel vector Distance from a BS to our user h`
i 2 CN , CN (0, 1)Precoding matrix
V`i =
⇥v`i,1, ...,v
`i,K
⇤,v`
i,k 2 CN ,��v`
i,k
�� = 1
maximize :
����h`0
�Tv`0,k
���2
subject to :
����h`0
�Tv`0,k0
���2= 0 for k0 6= k
����h`j
�Tv`j,k00
���2= 0 for dj 2 C` and 1 k00 K
: cluster pattern index `
© Robert W. Heath Jr. (2015)
u Lower bound on ergodic spectral efficiency (fixed geometry)
u Lower bound on ergodic spectral efficiency (random geometry, PPP)
Analyzing bounds on average achievable rates
18
E1
Llog2
�1 + SINR|`
��� 1
Llog2
1 +
exp ( (N � 2K + 1))
KP
dv2D`(kdvk / kd0k)��
+ kd0k� /SNR
!
where D` = {dv|dv 2 N`\C`}
cluster pattern cluster # of used colors (resources)
E1
Llog2
�1 + SIR|`
��� E
1
L
�log2
1 +
��2 � 4
�
8Kexp ( (N � 2K + 1))
!
# of users
J. Park, N. Lee, and R. W. Heath, “Cooperative Base Station Coloring for Pair-wise Multicell Coordination,” submitted to IEEE Transactions on Communications, available on ArXiv
© Robert W. Heath Jr. (2015)
Proof sketch
u Lower bound ª For non-negative random variable and ,
u Calculating ,
19
Elog2
✓1 +
S
I + 1
◆�� log2
✓1 +
eE[lnS]
E [I + 1]
◆.
S I
E [I]
E
2
4kd1k�X
di2�\B(0,kd2k)
kdik������h`i
�TV`
i
���2
3
5=KE
2
4kd1k�X
di2�\B(0,kd2k)
kdik��
3
5
=KEr1,r2
2
4E�\B(0,r2)
2
4r�1X
di2�\B(0,r2)
kdik��
������kd1k = r1, kd2k = r2
3
5
3
5
=KEr1,r2
r�1 2⇡�
Z 1
r2
r1��dr
�=
2K⇡�
� � 2
Z 1
r2=0
Z r2
r1=04 (�⇡)2 e��⇡r22r�+1
1 r3��2 dr1dr2
=8K
�2 � 4
User communicates with two nearest BSs = protection ball B (0, kd2k)
© Robert W. Heath Jr. (2015)
20
Other approaches for performance comparison
BS 1 BS 2
BS 3 BS 4
BS 1 BS 2
BS 3 BS 4
BS 1 BS 2
BS 3 BS 4
BS 1 BS 2
BS 3 BS 4
BS 1 BS 2
BS 3 BS 4
Single cell operation • No interference management is applied
Fractional frequency reuse • Adjacent BS uses different sub-band
Random clustering • BS cluster is made
with arbitrary rule (No coloring)
Conventional strategies - square grid model application example
© Robert W. Heath Jr. (2015)
u Ergodic spectral efficiency of edge users
Comparison results
21
−10 −5 0 5 10 15 20 250
0.2
0.4
0.6
0.8
1
1.2
1.4
SNR (dB)Er
godi
c sp
ectra
l effi
cienc
y (b
ps/H
z)
Proposed Clustering (simulationProposed Clustering (analytical lower bound)Proposed Clustering with ∆ EC=7
Proposed Clustering with ∆ EC=10
Proposed Clustering with ∆ EC=3
Random ClusteringFractional Frequency ReuseSingle Cell Operation
Performance improvement by edge-cutting
−10 −5 0 5 10 15 20 250
0.5
1
1.5
SNR (dB)
Sum
erg
odic
spec
tral e
fficie
ncy
(bps
/Hz)
Proposed Clustering (simulation)Proposed Clustering (analytical lower bound)Random ClusteringFractional Frequency ReuseSingle Cell Operation
Symmetric network case • No need edge-cutting
Asymmetric network case • More irregular • Edge-cutting might be needed
2x 1.5x
© Robert W. Heath Jr. (2015)
Concluding remarks u Main benefits
ª Any active user can communicate with two nearest BSs ª No BS is conflicted ª Can be combined with scheduling for greater gains
u Possible drawbacks and solutions ª Only useful for pairwise cooperation? – Two is enough* ª Edge-coloring demands too much complexity? – No done frequently ª Too many resources can be required? – Edge cutting
u Application ª Carrier aggregation
u Future direction ª Application to millimeter wave to reduce blockage effects
22 N. Lee, D. Morales, A. Lozano, and R. W. Heath, “Spectral Efficiency of Dynamic Coordinated Beamforming,” IEEE Transactions on Wireless Communications, 2015