An Adaptive Decentralized Congestion
Control Algorithm for a Multi-link MIMO
Interference System with the BER Constraint
By
Mirza Tayyab Mehmood
Agenda
 Introduction
 Literature Review
 System and Simulation Model
 Results
 Conclusions
 Future Recommendations
2
Introduction
 Background

Resource allocation




Game theory




resources: power, time slots, spectrum etc.
required efficient solution in the interests of all the parties
advanced techniques, i.e., game theory, fuzzy logic, neural
networks, genetic algorithm
tool to solve the problems
efficient mapping of problems into mathematical games
solve the games
Power allocation


battery life in wireless communication systems
power allocation as a constraint for in designing algorithms
3
Introduction
 Problem Description

Adaptive decentralized congestion control algorithm for
multi-link MIMO interfering system with the constraint on
QoS
 Power allocation
 single-link MIMO: Water-filling algorithm
 multi-link MIMO: map into game and solve the game
 best response process and gradient play process (Arslan et al.,
2007)

Adaptive decentralized congestion control algorithm
 adaptively choose appropriate modulation scheme for each link
 decentralized algorithm that take decision about traffic
 decision depends upon the type of traffic (voice or video) and
QoS requirements

Quality of service (QoS)
 bit error rate
 bit rate
4
System Model
 Power Allocation


L links, each with Nt and Nr transmit and
receive antennas
For k-th link



xk: complex signal vector of dimension Nt
yk: received baseband complex signal vector of
dimension Nr
Input-output relationship is as follows:
yk  k H k ,k x k 
L

l 1,l  k
 k , l H k ,l x l  n k ,
5
System Model

Power Allocation


Map the problem into game as follows

G  L, P Nt  Nt , U k 

where
 L is the number of players
N  Nt
 P t
is the common strategy space defined as for
any r  0
Prr  F: F is r  r Hermitian positive semi-definite matrix and trace(F)  1 


U k is the utility for k-th link defined as
U k  P   log 2 det  I  k R k1/2 H k ,k Pk H †k ,k R k1/2 
where

P  P1 ,...,PL  with Pi  P Nt Nt , i 1,..., L
6
System Model
 Power Allocation



Gradient Play Process (Arslan et al.,
2007)
Best Response Process (Arslan et al.,
2007)
Water-filling Algorithm (Telatar, 1999)
7
System Model
If links do not satisfy the QoS
then kill the link with minimum
metric and recalculate the power
Power allocation using
gradient play algorithm in
(Arslan et al., 2007)
Power level at each
logical channel
Data
bits
Channel
coding
QAM
signal
mapping
Power
allocation
on logical
channels
Link metric
computatio
n
QoS
Adaptive Congestion Control Model
Figure 1: Proposed system model for adaptive congestion control
indicating the steps taken by each link in each iteration of the adaptive
power allocation and congestion control algorithm.
8
System Model

Adaptive Congestion Control Model

Design a Metric for k-th link as follow
 Pek ,req
Metk  
 Pe
 k ,obt




where
 Pek , req is the probability of bit error required to satisfy the
service

Pek ,obt is the probability of bit error obtained for current
service


Transmit this metric to everyone
If at least one link do not satisfy the QoS
requirements then the link with minimum metric kills
itself.
9
Simulation Model
Random
Channel
Matrix
Input
Covariance
Matrix
Gradient Play
Process
Link with
minimum
metric kills
itself
Adaptive
Modulation Scheme
No
QoS
Satisfie
d?
Metric
Calculation
Yes
Exit
Figure 3: Simulation flowchart.
10
Results
 Number of links satisfying the quality of
service




SNR
INR
Channel matrix
Input power
 Comparison


With congestion control
Without congestion control
11
Results
Number of Links Satisfied the QoS without Channel Coding
10
SNR = 20 dB
9
8
7
Number of links
In this graph, we can
conclude that number
of links satisfied the QoS
without channel coding
not only depends upon
SNR and INR but also
on channel matrix and
input power. This graph
is obtained by averaging
over 10 runs.
SNR = 15 dB
6
5
4
SNR = 10 dB
3
2
SNR = 5 dB
1
0
-5
SNR = 0 dB
0
5
10
INR (dB)
12
Results
SNR = 10 dB
10
Congestion Control
5
0
-5
0
5
INR (dB)
10
15
SNR = 15 dB
10
Number of Links
Our congestion control algorithm
gives better link utilization in
comparison with the case of
without congestion control.
Number of Links
Without Congestion Control
Without Congestion Control
Congestion Control
8
6
4
2
0
-5
0
5
INR (dB)
10
15
13
Conclusion
 there are some links that cannot support the
traffic as well as satisfy the QoS
requirements, killing some of these links can
reduce the amount of inter-link interference
and improve the BERs of the remaining links.
 Proposed algorithm is decentralized
 Proposed algorithm is stable and provides
better link utilization over the system without
congestion control.
14
Future Recommendations
 Implementation of queues in the current work can
decrease the blocking probability
 Finding the power allocation for the correlated
channel matrix can be a good direction of work



Design of proper utility function, i.e., ergodic capacity
Conditions for Nash equilibrium
OFDM and space time coding for MIMO interfering
system
 Implementation of this system on FPGA using VHDL
is a good prototype of this model
15
Thank You
16