optimising cellular wireless networks using evolutionary computing

20
Optimising Cellular Wireless Optimising Cellular Wireless Networks using Evolutionary Networks using Evolutionary Computing Computing Centre for Adaptive Wireless Systems Martin Klepal 22nd June 2005

Upload: nirmala-last

Post on 15-Dec-2014

493 views

Category:

Technology


2 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Optimising Cellular Wireless Networks Using Evolutionary Computing

Optimising Cellular Wireless Networks Optimising Cellular Wireless Networks

using Evolutionary Computingusing Evolutionary Computing

Centre for Adaptive Wireless Systems

Martin Klepal

22nd June 2005

Page 2: Optimising Cellular Wireless Networks Using Evolutionary Computing

Adaptive Radio Resource Adaptive Radio Resource Management for GSMManagement for GSM

Large Scale Large Scale WLAN Design and WLAN Design and OptimisationOptimisation

(Ken Murray, Dirk Pesch)

(Martin Klepal, Alan Mc Gibney)

Page 3: Optimising Cellular Wireless Networks Using Evolutionary Computing

Adaptive Radio Resource Management for GSMAdaptive Radio Resource Management for GSM

ObjectiveObjective

The traffic evolution between cells is different

Busy periods occur at different times

Resources at quiet cells are wasted

With the introduction of 2.5G services such as GPRS and EDGE, a more flexible method of resource management is required to maximize system resources.

GSM networks employ fixed channel allocation model (FCA) to assign frequencies to base stations

Increase network capacity in GSM using an adaptive radio resource management system.

Page 4: Optimising Cellular Wireless Networks Using Evolutionary Computing

Proposed SolutionProposed Solution

Using Evolutionary computing techniques, we propose an Adaptive Radio Resource Management System

Adaptive Radio Resource Management for GSMAdaptive Radio Resource Management for GSM

Update the frequency assignment plan based on resource requirement predictions using a Genetic Algorithm

Next HourCurrent Hour

---1------1---------------1------1----------1------1-----1---1------1-----------1------1-----------------1------1---------1------1-----------1------1------1----1------1----------1------1------1---------1------1------1------1------1---1-----------1------1----------1------1----------1------1-------------1------1---1----1------1------1----------1------1---1-----1------1-----------1------1----------1-

Frequencies ->

Cells

->

---1------1---------------1------1----------1------1-----1---1------1-----1-----1------1-----------------1------1---------1------1-----------1------1-------1---1------1-------1---------1------1---------1------1----1--------1------1----1----------1------1----------1------1----------1------1----1--------1------1--------1------1------1----------1------1----1----1------1-----------1------1----------1-

Cells

->

Frequencies ->

Frequency Assignment in 20 cells

Prediction of future resource requirements for new and handover calls at each cell using Neural Networks based on historical data.

j

Input Layer

Hidden Layer

Output Layer

k

iWji

Wik

Neural Network

His

tori

cal D

ata

Pre

dic

tion

of

new

calls

Page 5: Optimising Cellular Wireless Networks Using Evolutionary Computing

Results and ConclusionResults and ConclusionAdaptive Radio Resource Management for GSMAdaptive Radio Resource Management for GSM

Simulation has shown resource gains of up to 21% when compared with current FCA frequency assignment schemes

The proposed approach has a non-invasive implementation within Operation Maintenance Centers of existing GSM network.

Page 6: Optimising Cellular Wireless Networks Using Evolutionary Computing

Martin Klepal, Alan Mc Gibney

Large Scale Large Scale WLAN Design and WLAN Design and

OptimisationOptimisation

Page 7: Optimising Cellular Wireless Networks Using Evolutionary Computing

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

MotivationMotivation

The design of wireless local area networks is currently still The design of wireless local area networks is currently still carried out in an ad-hoc fashion, with access point carried out in an ad-hoc fashion, with access point installation based on “rules of thumb” which leads to installation based on “rules of thumb” which leads to reduced performance from the deployed network.reduced performance from the deployed network.

The objective of this project is to address the issues related The objective of this project is to address the issues related to WLAN design, the use of Evolution Strategies for to WLAN design, the use of Evolution Strategies for optimisation of Access Point placement to overcome the ad-optimisation of Access Point placement to overcome the ad-hoc nature of WLAN design (WiFi, WiMax, …).hoc nature of WLAN design (WiFi, WiMax, …).

Page 8: Optimising Cellular Wireless Networks Using Evolutionary Computing

OutlineOutline

Site DescriptionSite Description

Signal Coverage and Channel Throughput Signal Coverage and Channel Throughput

PredictionPrediction

AP Placement Pre-processing & OptimisationAP Placement Pre-processing & Optimisation

Current ImplementationCurrent Implementation

Result & ScalabilityResult & Scalability

Future ResearchFuture Research

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 9: Optimising Cellular Wireless Networks Using Evolutionary Computing

Site DescriptionSite DescriptionLarge Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Multi-Storey Building

Part of CIT Campus

Page 10: Optimising Cellular Wireless Networks Using Evolutionary Computing

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Signal Coverage PredictionSignal Coverage Prediction

The Multi-Wall Model

+ Very Fast+ Very Fast

Less AccurateLess Accurate

Ray-Tracing Model

+ Accurate+ Accurate

Computation DemandingComputation Demanding

Page 11: Optimising Cellular Wireless Networks Using Evolutionary Computing

Throughput PredictionThroughput Prediction

Throughput Prediction

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Signal level + Site-specific Information

BER Prediction for CCK 11

Page 12: Optimising Cellular Wireless Networks Using Evolutionary Computing

Selection of Candidate APSelection of Candidate AP

Candidate Access Point positions forming an undirected graph that can be traversed during the optimisation

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 13: Optimising Cellular Wireless Networks Using Evolutionary Computing

Fitness FunctionFitness FunctionLarge Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

BwRwAwDwFF 4321

D … User Demand Satisfaction

A … Number of Access Points

R … Restricted Area

B … Solution Balance

wi … Waiting Factors

10

I

iiw

Elements of the Fitness Function:

The objective of the optimisation is to minimise the Fitness Function that evaluates if the suggested design of the network satisfies user demands by maximizing throughput with a minimum number of APs and other constraints.

Page 14: Optimising Cellular Wireless Networks Using Evolutionary Computing

Optimisation TechniqueOptimisation Technique

Survival of the fittest

Self-adaptationSelf-adaptation

Objective Function

Evolutionary Operators

Site-Specific Knowledge

FF

Population

Offspring (λ)

Parents(µ)

Terminate

Mutation(s)

Selection

Initialise

Evolution Strategy ModesEvolution Strategy Modes

(1+1)(1+1) Two – Membered StrategyTwo – Membered Strategy

((µ + µ + λλ) Selection includes ParentsSelection includes Parents

((µ , µ , λλ) Children only considered in SelectionChildren only considered in Selection

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Evolution Strategies

Page 15: Optimising Cellular Wireless Networks Using Evolutionary Computing

ImplementationImplementation

Evaluation Test-bed and Optimisation Kernel were implemented using Borland C++ providing both speed and stability during optimisation

Difference

Measurement

Features:

Drawing Tools for Environment Specification

Load/Save SVG Format

Signal Coverage Throughput Prediction

Wireless Technology Specification

Demands Specification

Environment Preprocessing Tools

Optimization Tools

Measurement Tools

Evaluation Test-bed

The optimisation is controlled through a GUI that allows the user to The optimisation is controlled through a GUI that allows the user to modify parameters and visualise the optimisation progress.modify parameters and visualise the optimisation progress.

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 16: Optimising Cellular Wireless Networks Using Evolutionary Computing

ResultsResults

Initial results of the optimisation technique implemented are stable because the same solution is suggested after each run on the same environment.

100% Coverage with a minimum number of AP

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 17: Optimising Cellular Wireless Networks Using Evolutionary Computing

ScalabilityScalabilityLarge Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

SegmentationSegmentation

Voronoi GraphVoronoi Graph

Crystals of Crystals of Variable SizeVariable Size

Backtracking Backtracking AlgorithmAlgorithm

Page 18: Optimising Cellular Wireless Networks Using Evolutionary Computing

Ongoing & Future ResearchOngoing & Future Research

Overcome the problem of scalability using Overcome the problem of scalability using Segmentation & Backtracking AlgorithmSegmentation & Backtracking Algorithm

3D Implementation3D Implementation

Large scale measurement and analysis of a Large scale measurement and analysis of a deployed solutiondeployed solution

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 19: Optimising Cellular Wireless Networks Using Evolutionary Computing

ConclusionConclusion

• Adaptive Radio Resource Management Adaptive Radio Resource Management System for GSMSystem for GSM

shows resource gains of up to 21% when compared with shows resource gains of up to 21% when compared with current FCA frequency assignment schemescurrent FCA frequency assignment schemes

• Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

aims to developed a computer aided automatic design tool aims to developed a computer aided automatic design tool that will provide an optimum WLAN design with minimum that will provide an optimum WLAN design with minimum number of APs providing required signal coverage and number of APs providing required signal coverage and network capacity.network capacity.

Large Scale Large Scale WLAN Design and OptimisationWLAN Design and Optimisation

Page 20: Optimising Cellular Wireless Networks Using Evolutionary Computing

Thank you for your attention!Thank you for your attention!