Title: A Short Course on the Application of Game Theory to Wireless Networks
1A Short Course on the Application of Game Theory
to Wireless Networks
2Homework 4 Solution
- Apply Backwards Induction to the centipede game
shown - below.
1
1
C
2,4
4,6
3,1
0,2
5,3
S
S
1,0
3Homework 4 Solution
2. Find the strategic form equivalent to (note
this is 3 stages)
Player 1 strategies
2,4
S,(S,S), S,(S,C), S,(C,S),
S,(C,C) C,(S,S), C,(S,C), C,(C,S),
C,(C,C)
Player 2 strategies
(S,S), (S,C), (C,S), (C,C)
4Homework 4 Solution
(C,C)
(C,S)
(S,S)
(S,C)
S,(S,S)
S,(S,C)
S,(C,S)
S,(C,C)
C,(S,S)
C,(S,C)
C,(C,S)
C,(C,C)
5Homework 4 Solution
- 3. Express the traditional Prisoners Dilemma as
an extensive form game.
6Homework 4 Solution
Formulate a call admission process for a network
with a capacity of 3 and 4 possible subscribers
as an extensive form game.
M 1,2,3,4
O set of subscribers out of network
Cannot join if N 3
N set of subscribers in network
Should simultaneous calls increase N gt 3,
player with highest player number is blocked.
k is the number of stages a subscriber
has remained in the network
7Lecture 6
- Mixed Strategies
- Wireless Networks as Games
8Mixed Strategies Wireless Networks as Games
- Mixed Strategies
- NE Existence in Mixed Strategies
- Wireless Networks as Games
- Power Control
- Adaptive Interference Avoidance
- Network Formation
- Node Migration (Participation for Vivek)
Skipping today
9Wireless Networks and Games
10There are three primary types of wireless
networks.
11Adaptive wireless network considerations.
12Game theory can be used to analyze distributed
power control algorithms.
Solution of games Nash equilibriums yields
information on networks convergence, fairness,
and optimality
13Common Types of Adaptive (and Uncoordinated)
Behavior in Wireless Networks
- Physical Layer Adaptations (MAC Strategies)
- Ad-hoc Network Formation
- Node Mobility (Participation)
14Physical Layer Adaptations
Nodes are free to alter some aspect of their
physical layer behavior (power level or
waveform). Interaction occurs through
interference caused at nodes of interest.
Interference is a function of both power and the
waveforms selected by each node. Simplified
algorithms may consider one adaptation or the
other separately.
Examples Distributed Power Control (like in
reverse link in cellular or in an
ad-hoc)
Adaptive MAC strategies (alter link in response
to perceived
changes in environment channel,
noise,
interference)
15Physical Layer Model
M the set of decision making radios Ei the
set of possible energy levels available to radio
i ei the energy level chosen by i e - the
tuple of chosen energy levels of all radios in
the network ?i the set of signature waveforms
available to radio i ?i the chosen waveform
of i ? - the tuple of chosen waveforms of all
radios in the network Ni noise power at node
i ?ij - the correlation between the signature
waveform sequences of radios i and j. Note that
?ij necessarily equals ?ji.
16Radio Model (2)
vi node of interest for node i
?i,j path loss from i to j
v1
17Network Formation
Basic Idea Wireless nodes change which nodes
they are communicating with. Some cost
associated with forming a link (function of path
loss and waveform), benefit from minimal number
of hops to all other nodes in the network.
Example routing table generation
18Network Formation Model
M 1,2,3,,n the set of decision making
radios ?i,j path loss from i to j li,j lj,i a
link from i to j ci(li,j) cost of a link from i
to j L A particular network formed from links
li,j ui(L) benefit to i from network L
Actions are ultimately a yes-no decision to each
link.
May be single-sided or two-sided decision process.
19Node Mobility
Basic Idea Wireless nodes change which
network they are participating in. This can be
triggered through node mobility or through
changes in network composition. Performance seen
by a node is generally a function of the
specific network the node is participating in,
the number of nodes in each network, and channel
conditions. Examples Co-existing networks,
handoffs
20Node Mobility Model
M 1,2,3,,m set of nodes
N n1,n2,n3,,nn - set of networks
Ni n1,n2,n3,,nn - set of networks available
to node i
?i network chosen by node i
? network choice vector
? ?i Ni network choice space
- channel seen by node i in network ?i
Value that node i receives from network choice
vector given
21Quick Example
Distributed Power Control
22 Mobiles, uncoordinated power control with base
station. Each adjusts its power level to achieve
a target BER. Rationale For many applications
there is break-point where BER gt target yields
unacceptable QoS, while BER lt target provides
little additional benefit. Mobiles would like
to achieve target BER with minimum transmit
power so as to conserve battery life. Power is
updated mobile decides increase or decrease of
its power level would bring it closer to target
BER.
22Model
Single Cell, 22 devices (Player set 1,2,,22)
No sectorization
Reverse Link (No Synchronization)
DS-SS N 63
at base-station (Node 0)
AWGN
Mobile Power Range (-?, 20 dBm (Action Set)
23A target BER scheme is both convergent and fair.
Distributed algorithm
NE given by maximizer of
where
Network Behavior
Value to Nodes
24Comments
- Modeled System is a (Ordinal) Potential Game
- Network which is a (Ordinal) Potential Game
requires minimal overhead to support convergence
of adaptive behavior. - Note weve not commented on efficiency but if
measured in terms of system capacity, then
performance is optimal (equals coordinated
capacity).
25Extra Slides
26Mixed Strategies
The cliffhanger
Jody, whats mixed strategies?
Mixed Extension to Strategic Form Game
A particular probability assignment
27Mixed Strategies Stories
- Objects of Choice
- Stochastic Steady State
- Pure Strategies in an Extended Game
- Uncertainty in Other Players Payoffs
28N - player set
Ai (Finite) Set of (Pure) Strategies (Actions)
Available to Player i
?i A
Ai (Finite) Set of Actions Available to Player
i
29Mixed Strategy Games
Components
N Set of Players Ai Finite Set of
Actions Available to Player i p ui Set of
Individual Objective Functions
30Nash Equilibrium
Mozart-Mahler
a2
b2
1
a1
1,1
0, 0
b1
0, 0
3, 3
0
1
311
0
1
32Support and Dominance