Capacity Optimization for UMTS: Bounds on Expected...
Transcript of Capacity Optimization for UMTS: Bounds on Expected...
Capacity Optimization for UMTS:Bounds on Expected Interference
joint work with:Hans-Florian Geerdes
Zuse Institute Berlin (ZIB)
DFG Research Center MATHEON:Mathematics for Key Technologies
CO@Work, ZIB, Berlin, 05/10/2009
Andreas Eisenblättereisenblaetter@{atesio,zib}.de
atesio GmbH, BerlinZuse Institute Berlin (ZIB)
DFG Research Center MATHEON:Mathematics for Key Technologies
2
Radio Network Planning
antenna
cell
schematic view of a radio network, Berlin
base station position
equipment
antenna configuration
Parameters
cost
coverage
capacity
Objectives
1km
Andreas Eisenblätter / UMTS Radio Network Optimization
3
cost
coverage
capacity
Objectives
Radio Network Planning
antenna
cell
schematic view of a radio network, Berlin
base station position
equipment
antenna configuration
Parameters1km
tilt=12°azimuth=90°
tilt=4°azimuth=90°
tilt=2°azimuth=120°
Andreas Eisenblätter / UMTS Radio Network Optimization
4
Radio Network Planning
antenna
cell
schematic view of a radio network, Berlin
base station position
equipment
antenna configuration
Parameters
cost
coverage
capacity
Objectives
1km
Andreas Eisenblätter / UMTS Radio Network Optimization
5
Radio Network Planning
antenna
cell
base station position
equipment
antenna configuration
Parameters
cost
coverage
capacity
Objectives4.8
0
normalizedtraffic
intensity[km-2]
1km
Andreas Eisenblätter / UMTS Radio Network Optimization
6
Capacity, Power Control, and Interference in UMTS
inter-ference
antenna
user
establishedconnection
1km
0 0,5
0
5
Time [s]
Rec
eive
d Po
wer
[dB
m]
0 0,5 1
-10
0
10
Time [s]
Chan
nel g
ain
[dB]
Power Control
transmit power received power
on each link, the power is dynamically regulated
limiting resource is cell power
Andreas Eisenblätter / UMTS Radio Network Optimization
7
Capacity, Power Control, and Interference in UMTS
inter-ference
antenna
user
establishedconnection
1km
0 0,5
0
5
Time [s]
Rec
eive
d Po
wer
[dB
m]
0 0,5 1
-10
0
10
Time [s]
Chan
nel g
ain
[dB]
Power Control
transmit power received power
on each link, the power is dynamically regulated
limiting resource is cell power
Andreas Eisenblätter / UMTS Radio Network Optimization
8
Capacity, Power Control, and Interference in UMTS
inter-ference
antenna
user
establishedconnection
1km
0 0,5
0
5
Time [s]
Rec
eive
d Po
wer
[dB
m]
0 0,5 1
-10
0
10
Time [s]
Chan
nel g
ain
[dB]
Power Control
transmit power received power user positions
user demand
interference generated in neighboring cells
the network design
Cell capacity depends on
on each link, the power is dynamically regulated
limiting resource is cell power
Andreas Eisenblätter / UMTS Radio Network Optimization
9
Overview
System Model: Interference Coupling Systems
Capacity Maximization: Optimization Model and Methods
Assessing the Quality of Optimization Results: Bounds
Andreas Eisenblätter / UMTS Radio Network Optimization
10
State-of-the-art: Detailed View on Network
inter-ference
antenna
user
establishedconnection
rejecteduser
admission decisionper user (algorithm)
solving large equation system (user x user)
Network evaluation requires
static evaluation of a demand snapshot
CIR equation
on each link
1km
calculate cell powers
Andreas Eisenblätter / UMTS Radio Network Optimization
11
Top-Level View on Network Performance
interference coupling matrix
20
0
[W]
cell power fraction of served demand
100
95
[%]
How much power is needed to serve a fraction of the user demandvirtually equivalent to classical model(nonlinear equation system, cells x cells)
Revised Interference Coupling Equation
0.2
0.0
off-diagonalelements
Andreas Eisenblätter / UMTS Radio Network Optimization
12
New System Model: Interference Coupling Complementarity System
Properties
iterative method(modified Gauß-Seidel)
effort equivalent to solving an equation system
Solution Algorithm Analysis
Generalized pole equations: isolate interference from other cells
power control
load factor
load factor
load control
perfect
Closed-formperformancemodel: capacityfunction of matrix
Solutions to classical model: spectral radius of C and bounds
DL: unique solution UL: 0,1,…,n,…,1
solutionsAndreas Eisenblätter / UMTS Radio Network Optimization
13
Overview
System Model: Interference Coupling Systems
Capacity Maximization: Optimization Model and Methods
Assessing the Quality of Optimization Results: Bounds
Andreas Eisenblätter / UMTS Radio Network Optimization
14
Optimization Model
From System Model to Optimization Model
minimize expected interference
nonconvex in matrix
critical: # configuration options, size/resolution of scenario
System Model Expected Couplingperformance is a function of the coupling matrix
expected matrix is a good representative
Optimization Idea
Network design is characterized by a single matrix
Optimization is matrix design
What is a good coupling matrix?
constrain coverage
Neumann series
Andreas Eisenblätter / UMTS Radio Network Optimization
15
Optimization Methods and Experiments (Berlin)
Start Configuration Local Search MIP 4-opt Heuristic
uniform settings, coverage maximization, greedy site reduction
choose azimuth within ± 30°
choose downtiltwithin 2-12°
0
20
TXpower[W]
try all alternatives for a single antenna
adopt improving one
try modifications of 4 sectors at once
find improvements with approximate MIP
polish with local search
Andreas Eisenblätter / UMTS Radio Network Optimization
16
Optimization Methods and Experiments (Berlin)
Start Configuration Local Search MIP 4-opt Heuristic
uniform settings, coverage maximization, greedy site reduction
choose azimuth within ± 30°
choose downtiltwithin 2-12°
0
20
TXpower[W]
try all alternatives for a single antenna
adopt improving one
try modifications of 4 sectors at once
find improvements with approximate MIP
polish with local search
Andreas Eisenblätter / UMTS Radio Network Optimization
17
Optimization Methods and Experiments (Berlin)
Start Configuration Local Search MIP 4-opt Heuristic
uniform settings, coverage maximization, greedy site reduction
choose azimuth within ± 30°
choose downtiltwithin 2-12°
try all alternatives for a single antenna
adopt improving one
try modifications of 4 sectors at once
find improvements with approximate MIP
polish with local search
0
20
TXpower[W]
Andreas Eisenblätter / UMTS Radio Network Optimization
18
Optimization Methods and Experiments (Berlin)
Start Configuration Local Search MIP 4-opt Heuristic
uniform settings, coverage maximization, greedy site reduction
choose azimuth within ± 30°
choose downtiltwithin 2-12°
0
20
TXpower[W]
try all alternatives for a single antenna
adopt improving one
try modifications of 4 sectors at once
find improvements with approximate MIP
polish with local search
Andreas Eisenblätter / UMTS Radio Network Optimization
19
Optimization Methods and Experiments (Berlin)
Start Configuration Local Search MIP 4-opt Heuristic
uniform settings, coverage maximization, greedy site reduction
choose azimuth within ± 30°
choose downtiltwithin 2-12°
0
20
TXpower[W]
try all alternatives for a single antenna
adopt improving one
try modifications of 4 sectors at once
find improvements with approximate MIP
polish with local search
Andreas Eisenblätter / UMTS Radio Network Optimization
20
Overview
System Model: Interference Coupling Systems
Capacity Maximization: Optimization Model and Methods
Assessing the Quality of Optimization Results: Bounds
Andreas Eisenblätter / UMTS Radio Network Optimization
21
Derivation of Pole Equation
Other-to-own-cellinterference
Detailed description of interaction between cells
Single out coupling between cells
Alternative description
Regularity assumptions
Simplified view of single cell
Average other-to-own-cell interference ratio and orthogonality
Andreas Eisenblätter / UMTS Radio Network Optimization
22
Bounds for Capacity Optimization
0.1 %
13.0 %
8.4 %
12.0 %
gap to reference
20.8 %Vienna
47.2 %Turin
23.4 %Lisbon
34.7 %Berlin
gap to lower bound
avg. load factor
avg.cellload
50%
0
optimized
perfect load balance
startno overlap
lower bound
reference
Andreas Eisenblätter / UMTS Radio Network Optimization
23
Conclusion
UMTS System
Refined Model
ApproximateEvaluation
Planning
Optimization
models
methods
Static model/evaluation
Validation
experiments
analysis
Capacity planning is difficult due to interference limitation
Understand capacity through coupling matrix (alone)
Expected coupling matrix is a sensible representative
Simple methods are successful in many cases
Interference minimization with objective featuring coupling matrix
More details: PhD thesis by Hans-Florian Geerdes“UMTS Radio Network Planning: Mastering Cell Coupling for Capacity Optimization”
Andreas Eisenblätter / UMTS Radio Network Optimization
24
Thank You!
Andreas Eisenblätter / UMTS Radio Network Optimization
25
Intuition of Interference Coupling:Neumann Series
Problem
Series may not converge
Cell power constraints may be violated
If too many users request service
Andreas Eisenblätter / UMTS Radio Network Optimization
26
Local Search Scheme
??
?
start configuration
improvedconfiguration
repeatedly change configuration, evaluate
Optimization
Tabu Search Genetic Algorithms Simulated Annealing …
Refined Search Algorithms
Andreas Eisenblätter / UMTS Radio Network Optimization
27
Mixed Integer Programming Formulation for Matrix Design
traffic
high
low
Optimization Model
Convex approximation of objective function
realistic setting
Design coupling matrix
MIP model for calculating expected coupling matrix and approximated objective function
Use in k-opt heuristic (k=4,5,6)
Accept/reject MIP incumbents according to accurate nonlinear evaluation
ideal network
Matrix DesignMIP Heuristic
Andreas Eisenblätter / UMTS Radio Network Optimization
28
aggregate user parameters, calculate cell powers directly
Calculating DL Cell Powers: Interference-Coupling Equation System
power-controllink
inter-cellinter-ference
Signal Quality Equations (per user)
Cell Coupling Equation System (per cell)
Describes power balance between cells
Takes into account all relevant information on users
user parameters: orthogonality, activity, path loss, CIR target
link powers
(ignore noise)
cell powers
Andreas Eisenblätter / UMTS Radio Network Optimization
29
The Revised Pole Equation: Overcoming Restrictive AssumptionsClassical Definitions Revised Definitions
Classical Pole Equation Revised Pole Equation
Include orthogonality loss factor
Weighted average according to user specific parameters
Same simple structure
Precise parameters per cell
Other-to-own-cell interference ratio, straightforward average
Andreas Eisenblätter / UMTS Radio Network Optimization
30
Understanding the Behavior of a Single Cell with the Pole Equation
loadfactor
cellpower
Asymptote:
Derivative:
Pole equation describes dependency between load and cell powerand the impact of orthogonality and interference for a single cell
load factor
Andreas Eisenblätter / UMTS Radio Network Optimization