Adaptive Radio Resource Management for GSM/GPRS Networks



Similar documents
Radio Resource Allocation in GSM/GPRS Networks

CURRENT wireless personal communication systems are

RESOURCE ALLOCATION FOR INTERACTIVE TRAFFIC CLASS OVER GPRS

GSM Frequency Planning with Band Segregation for the Broadcast Channel Carriers

Dynamic Reconfiguration & Efficient Resource Allocation for Indoor Broadband Wireless Networks

NOVEL PRIORITISED EGPRS MEDIUM ACCESS REGIME FOR REDUCED FILE TRANSFER DELAY DURING CONGESTED PERIODS

AN ANALYSIS OF DELAY OF SMALL IP PACKETS IN CELLULAR DATA NETWORKS

International Journal of Advanced Research in Computer Science and Software Engineering

A Novel Approach for Load Balancing In Heterogeneous Cellular Network

Packet Queueing Delay in Wireless Networks with Multiple Base Stations and Cellular Frequency Reuse

Performance of TD-CDMA systems during crossed slots

Enhancement of QoS in Mobile Network through Channel Allocation using Software Agents

How To Understand The Gsm And Mts Mobile Network Evolution

Figure 1: cellular system architecture

Mobile Tracking and Resource Reservation Scheme for Cellular Networks

PERFORMANCE AND EFFICIENCY EVALUATION OF CHANNEL ALLOCATION SCHEMES FOR HSCSD IN GSM

Using Data Mining for Mobile Communication Clustering and Characterization

Improved Channel Allocation and RLC block scheduling for Downlink traffic in GPRS

Location management Need Frequency Location updating

Voice Service Support over Cognitive Radio Networks

Indian Journal of Advances in Computer & Information Engineering Volume.1 Number.1 January-June 2013, Academic Research Journals.

Performance Evaluation of VoIP Services using Different CODECs over a UMTS Network

CHAPTER 1 1 INTRODUCTION

Guide to Wireless Communications. Digital Cellular Telephony. Learning Objectives. Digital Cellular Telephony. Chapter 8

G.Vijaya kumar et al, Int. J. Comp. Tech. Appl., Vol 2 (5),

Analysis of QoS parameters of VOIP calls over Wireless Local Area Networks

Performance advantages of resource sharing in polymorphic optical networks

Analysis and Enhancement of QoS in Cognitive Radio Network for Efficient VoIP Performance

Implementation of Mobile Measurement-based Frequency Planning in GSM

A Novel Pathway for Portability of Networks and Handing-on between Networks

Smart Mobility Management for D2D Communications in 5G Networks

Improving the Performance of Handoff Calls using Frequency Sharing

Chapter 6 Wireless and Mobile Networks

A Framework for supporting VoIP Services over the Downlink of an OFDMA Network

Dynamic Load Balance Algorithm (DLBA) for IEEE Wireless LAN

DeuceScan: Deuce-Based Fast Handoff Scheme in IEEE Wireless Networks

Drop Call Probability in Established Cellular Networks: from data Analysis to Modelling

QoS of Internet Access with GPRS

A Power Efficient QoS Provisioning Architecture for Wireless Ad Hoc Networks

ADHOC RELAY NETWORK PLANNING FOR IMPROVING CELLULAR DATA COVERAGE

Algorithms for Interference Sensing in Optical CDMA Networks

White Paper: Microcells A Solution to the Data Traffic Growth in 3G Networks?

Channel assignment for GSM half-rate and full-rate traffic

A NOVEL OVERLAY IDS FOR WIRELESS SENSOR NETWORKS

Multiobjective Multicast Routing Algorithm

Memory Allocation Technique for Segregated Free List Based on Genetic Algorithm

WBAN Beaconing for Efficient Resource Sharing. in Wireless Wearable Computer Networks

Architecture Overview NCHU CSE LTE - 1

Multi-service Load Balancing in a Heterogeneous Network with Vertical Handover

Packet Synchronization in Cellular Backhaul Networks By Patrick Diamond, PhD, Semtech Corporation

Applying Active Queue Management to Link Layer Buffers for Real-time Traffic over Third Generation Wireless Networks

A Policy-Based Admission Control Scheme for Voice over IP Networks

Weighted Thinned Arrays by Almost Difference Sets and Convex Programming

Joint Radio Resource Management and QoS Implications of Software Downloading for SDR Terminals

Technical and economical assessment of selected LTE-A schemes.

PERFORMANCE OF THE GPRS RLC/MAC PROTOCOLS WITH VOIP TRAFFIC

Introduction to Natural Computation. Lecture 15. Fruitflies for Frequency Assignment. Alberto Moraglio

Inter-cell Interference Mitigation Reduction in LTE Using Frequency Reuse Scheme

3GPP Wireless Standard

2002 IEEE. Reprinted with permission.

A Slow-sTart Exponential and Linear Algorithm for Energy Saving in Wireless Networks

A Robust Method for Solving Transcendental Equations

Antenna Based Self Optimizing Networks for Coverage and Capacity Optimization

Wireless System Design Experience: Case Study of a Manufacturing Factory

Detecting Multiple Selfish Attack Nodes Using Replica Allocation in Cognitive Radio Ad-Hoc Networks

Customer Training Catalog Training Programs WCDMA RNP&RNO Technical Training

EE6390. Fall Research Report. Mobile IP in General Packet Radio System

A Statistical Estimation of Average IP Packet Delay in Cellular Data Networks

A Novel Decentralized Time Slot Allocation Algorithm in Dynamic TDD System

2G/3G Mobile Communication Systems

From reconfigurable transceivers to reconfigurable networks, part II: Cognitive radio networks. Loreto Pescosolido

TABLE OF CONTENTS. Dedication. Table of Contents. Preface. Overview of Wireless Networks. vii xvii

Planning of UMTS Cellular Networks for Data Services Based on HSDPA

Multiuser Communications in Wireless Networks

Channel Allocation in Cellular Telephone. Systems. Lab. for Info. and Decision Sciences. Cambridge, MA

Synthesis Of Polarization Agile Interleaved Arrays Based On Linear And Planar ADS And DS.

SELECTIVE ACTIVE SCANNING FOR FAST HANDOFF IN WLAN USING SENSOR NETWORKS

Comparison of Major Domination Schemes for Diploid Binary Genetic Algorithms in Dynamic Environments

DATA SECURITY 1/12. Copyright Nokia Corporation All rights reserved. Ver. 1.0

Load balancing in a heterogeneous computer system by self-organizing Kohonen network

How To Understand The Performance Of A Cell Phone Network

An Efficient QoS Routing Protocol for Mobile Ad-Hoc Networks *

QoS-Aware Load Balancing in 3GPP Long Term Evolution Multi-Cell Networks

Lecture overview. History of cellular systems (1G) GSM introduction. Basic architecture of GSM system. Basic radio transmission parameters of GSM

ISSN: ISO 9001:2008 Certified International Journal of Engineering Science and Innovative Technology (IJESIT) Volume 2, Issue 3, May 2013

8. Cellular Systems. 1. Bell System Technical Journal, Vol. 58, no. 1, Jan R. Steele, Mobile Communications, Pentech House, 1992.

DYNAMIC RADIO RESOURCE MANAGEMENT IN GSM/GPRS USING SCALABLE RESOURCE ALLOCATION TECHNIQUE

VOICE OVER WI-FI CAPACITY PLANNING

Transcription:

Adaptive Radio Resource Management for GSM/GPRS Networks Ken Murray & Dirk Pesch Adaptive Wireless Systems Group Department of Electronic Engineering Cork Institute of Technology, Cork, Ireland Tel +353 2 4326 Fa: +353 2 4326625 E-Mail: {kmurray, dpesch}@citie Abstract Recent years have seen a dramatic increase in demand for mobile communication services and with the introduction of 25G services such as GPRS, this trend is epected to increase further The highly dynamic and bursty nature of packet switched services such as GPRS will require a much more fleible method of radio resource management so as to maimise system resources compared to currently employed fied channel allocation (FCA) schemes In this paper we propose a new pro-active method for increasing network capacity by introducing an adaptive radio resource management system into a typical GSM/GPRS network The adaptive radio resource management system predicts future radio resource requirements for both circuit switched GSM calls and packet switched GPRS sessions using neural networks (NNs) assignment is then performed using a genetic algorithm (GA) Results are presented which ehibit less resource requirements than eisting fied channel allocation (FCA) networks and performance that is comparable to recently proposed dynamic resource allocation (DRA) schemes, but with the advantage of significantly less compleity and no additional network signalling load INTRODUCTION With the evolution toward 25G services such as GPRS, the increase in demand for mobile communication services is epected to grow at an eponential rate Such systems eperience highly dynamic tele-traffic variations and the demand for managing the system and its resources in a fleible manner is increasing [,2] A plethora of concepts attempting to introduce adaptation in the form of DRA have been proposed in the past [3] All these reactive schemes operate in real-time and thus introduce high signalling loads in the network or require distributed control, which results in considerable change to the configuration and operation to both terminals and base station equipment In this work we propose a pro-active approach to radio resource management based on resource demand prediction using neural networks (NNs) Resource predictions are made for both new and handover GSM calls and GPRS sessions based on previous load characteristics Once resource predictions have been made, a GA is used to update the frequency allocation plan The GA is designed to minimise the impact on frequency changes in consecutive frequency assignment plans The updated frequency assignment plan can then be deployed to the network using mechanisms such as those proposed in [4] Neural networks have been proposed in the past for dynamic resource allocation [5,6,7,8,9] These schemes operate in real-time, with each cell requiring information regarding channel usage in neighbouring cells They also adopt a decentralised structure, which introduces high signalling loads during the collection period of performance data and the reassignment of resources throughout the network It is these attributes that prevent their integration into current GSM networks The proposed scheme does not operate in real-time but at a time granularity of between 3 minutes to one hour The granularity chosen here is one hour, as it is equal to the typical performance parameter reporting cycle of eisting GSM networks The adaptive radio resource management system is implemented in the operation and maintenance centre (OMC) of a typical GSM network This centralised location is chosen as the required performance management data for the whole cellular network is available at this location, therefore, no additional signalling load is generated for the operation of the predictive scheme 2 RESOURCE PREDICTORS As the performance of the system will depend strongly on the accuracy of the resource predictions, the resource predictors must be robust enough to track the inherent hourly changes in call traffic It has been shown that resource predictors based on multi-layered feed forward neural networks (MFNNs) can make accurate predictions when trained with sufficient amount of historical data [5] The system proposed here considers a MFNN for each type of traffic at each cell Each MFNN contains three layers with a total of 49 neurons The back-propagation learning algorithm and non-linear sigmoid

activation function are used in the learning process [] The training and prediction of the resource predictors proceeds as follows: Collect hourly radio resource demand statistics for GSM calls (new and handover) and GPRS sessions 2 Record whether the demand occurs on a weekday or weekend (day statistic) Record the time (time statis tic) These statistics constitute the initial training data set 3 The MFNNs are trained using the data arising from step 2 4 Once the MFNNs are trained, the channel demand for the net hour in each cell is predicted using the demand statistics from the previous hours, the day and time statistics 5 The predicted number of frequencies for each traffic type is assigned to each base station 6 The training set of 8 weeks is updated to contain the statistics for the current hour (assuming the network gathers statistics at least every 6 minutes) 7 Each MFNN is retrained every 24 hours to maintain accurate predictions Fig shows the performance for one of the resource predictors for GSM traffic for a period of one week This plot demonstrates the ecellent degree of accuracy achieved by the MFNNs Neural Network Prediction 35 3 Channel Demand 25 2 5 5 Actual Prediction 9 28 37 46 55 64 73 82 9 Time (hr) 9 8 27 36 45 54 63 Fig Neural Network Resource Prediction 3 SYSTEM IMPLEMENTATION The proposed adaptive radio resource management system is integrated in the OMC between the performance management (PM) and the configuration management (CM) tools The architecture is depicted in Fig 2 The centralised implementation has a number of advantages Firstly all the performance data required for training and prediction is available at the OMC and need not be communicated specially for the purpose of resource management Secondly, the centralised location within the OMC has the advantage that it does not require software or hardware updates to terminals or base station equipment The non-invasive nature of the proposed concept is one of its major advantages in that it can be implemented and improved without interference with eisting equipment Each MFNN is retrained every 24 hours from statistics pegged from the PM tool The resource predictions are sent to the CM tool where a GA is used to update the frequency assignment plan Network Statistics PM OMC OMC Assignment CM Predictive RRM Fig 2 Adaptive Radio Resource Management Architecture

4 FREQUENCY ASSIGNMENT USING A GA Many schemes incorporating GAs to solve the frequency assignment problem have been proposed in the past [,2] Although these schemes achieve ecellent degrees of optimisation (8-9%), they suffer from large variations in successive frequency assignment plans Using such GA optimisation techniques would require many base stations to change frequency carriers each hour, thus introducing a large amount of inter-cell handovers into the system [3] As an alternative to this approach, a GA based frequency assignment algorithm was developed which ensures that frequencies assigned to a cell include most of those frequencies assigned to that cell in the previous hour, thus minimising the number of frequency changes required from hour to hour The GA based frequency assignment scheme is presented in the following In the proposed adaptive radio resource management system, the resource requirements for GSM and GPRS calls arise from the resource predictors, while the interference constraints are represented by an nn compatibility matri C, where n is the number of cells in the system C = Cells n 2 2 22 n2 Cells ij n 2n nn Elements ij (i,j =,,n) represent the frequency separation required between frequencies assigned to cells i and j, respectively, necessary to maintain interference below a certain threshold Using this matri, it is possible to represent co-channel interference by choosing values for ij such that, ij = : if cell i and j cannot use the same frequency : otherwise The traffic demands can be represented by the demand vector D, with elements d i ( i =,2n) representing the number of required frequencies at cell i, the resource predictors fill out this vector at the end of each hour for both GSM and GPRS calls The frequency assignment problem is then defined as, given F frequencies and N cells each requiring d i frequencies, find an NF frequency assignment matri A given by, such that, A = Cells Frequencies a ik a ik = : if cell i is assigned frequency k : otherwise

A frequency assignment is admissible if both traffic and interference constraints are fulfilled This implies that: F k= = aik di for all i 2 Valid frequencies are assigned to cells according to the compatibility matri, C A number of binary groups of length n are created from the demand vector, D A binary within a cell group denotes a cell that requires a frequency The first group represents those cells requiring at least one frequency, the second group for those requiring at least two frequencies and so on Each group is then passed to the GA, which finds the minimum number of frequencies for the demand represented by the binary group Since the GA finds optimal solutions for each group separately, the overall solution may be sub-optimal, however, it does ensure that cells can maintain the majority of frequencies from hour to hour, as such changes will only be reflected in the last one or two binary groups, thus minimizing inter-cell handovers The optimal solutions from each group are augmented to create the final frequency assignment plan The GA works with an initial population of size 4, each individual in the population is represented as follows: Cell group Cell group N (,,,,,,,,,,,,,,,,,,,,,) Each cell group has an initial length of 7, as this is the maimum number of frequencies required for the first binary group (assuming a cluster size of 7) If the GA finds a valid assignment for seven frequencies, a solution is sought for si and so on until no better solution can be found The roulette wheel selection algorithm is used to generate the parents for the new population [4] The new population is created using the standard multi-point crossover and a special mutation operator with probabilities 6 and 3 respectively [3] 5 SIMULATION PLATFORM Two simulation models have been developed in this work an FCA model which is currently used in GSM/GPRS networks and a model based on the proposed adaptive radio resource management system with the embedded GA frequency assignment algorithm Both network models contain 49 cells with wraparound in the and y planes so that each cell has a total of 8 neighbours The structure of the GSM and GPRS simulation platforms are discussed in the following 5 GSM SIMULATION PLATFORM` In the GSM simulation platform, the load is non-uniformly distributed across the network The call arrival rate, λ, has a Poisson distribution while the call holding time, /µ, has a mean of 8 seconds The handover arrival rate in each cell is obtained by taking 5% of the sum of the call arrival rates in the si surrounding cells 52 GPRS SIMULATION PLATFORM In the GPRS simulation platform, the users are normally distributed, while the number of packets to be transmitted by each mobile follow a Poisson distribution Each mobile transmits, 2 and 3 slot packets with probabilities of 7, 2 and respectively If a mobile requires 3 slots to transmit a packet and 3 Packet Data Channels (PDCHs) are not available at the serving base station, the mobile will attempt to transmit the packet over 2 PDCHs and so on until a request for PDCH is refused The packet will then be queued at the mobile for retransmission when the required number of PDCHs becomes available

6 SIMULATION RESULTS The performance of both the FCA and adaptive resource management model is measured by the number of frequencies required to maintain the average GSM call blocking below 2% throughout the network The performance results will now be presented for both models 6 FCA NETWORK MODEL In this model each cell in the network is assigned the required number of frequencies for both GSM and GPRS calls so as to maintain the GSM call blocking below 2% at the busy hour The real-time simulation was run for the duration of 2 weeks and the call blocking statistics monitored at each cell for each traffic type Fig 4 shows the average GSM call blocking for cell site 2 The call dropping rate for handover GSM calls was found to be zero for each cell in the network A total of 29 frequencies were required for new GSM call arrivals, while 4 guard frequencies achieved the recorded GSM call dropping performance To achieve a packet blocking rate of zero, 24 frequencies were required for GPRS traffic 62 ADAPTIVE RADIO RESOURCE MANAGEMENT NETWORK MODEL The same traffic scenario was used in this model Unlike the FCA concept, cells were assigned frequencies for both GSM and GPRS calls based on resource predictions using the GA frequency assignment tool A total of 23 frequencies were required for new GSM call arrivals, producing a gain of 27% when compared with the equivalent FCA network This result is comparable to current DRA schemes [3], but with the advantage of significantly less compleity and no additional signalling load The average call blocking is shown in Fig 5 It was found that no benefits could be obtained from adaptive guard channel allocation, as the handover call arrival rate tends to be more uniformly distributed than new call arrivals Simulation results show that 2 frequencies are required for GPRS traffic, giving a resource gain of 25% when compared with the equivalent FCA network The average packet blocking is shown in Fig 6 The additional blocking in the adaptive network arises because each cell is allocated just the required number of frequencies for the net hour, thus maimising the systems resources The results are summarized in Table Average Call Blocking Average Call Blocking blocking % 4 2 8 6 4 2 9 37 55 73 9 9 27 45 63 8 99 time (hr) 27 235 253 27 289 37 325 343 blocking % 8 6 4 2 8 6 4 2 9 37 55 73 9 9 27 45 63 8 99 27 time (hr) 235 253 27 289 37 325 343 36 Fig 4 Average Call Blocking in FCA Fig 5 Average Call Blocking in Adaptive Network Network blocking % 8 7 6 5 4 3 2 Fig 6 2 39 58 77 Average Packet Blocking 96 5 34 53 72 9 time (hr) 2 229 248 267 286 Average Packet Blocking in Adaptive Network 35 324 Network Model requirements for new GSM call arrivals requirements for handover GSM calls requirements for GPRS sessions Table SUMMARY OF RESULTS FCA Adaptive resource management Resource Gain % 29 23 27 4 4-24 2 25

5 CONCLUSION An adaptive radio resource management system for current GSM/GPRS networks has been proposed Its non-invasive implementation within the OMC makes it a viable proposal for a more fleible management of resources for 25G networks Simulation results have shown resource gains of up to 27% and 25% for GSM and GPRS traffic respectively Using frequency deployment mechanisms such as those discussed in [4], this approach can achieve self-configuring cellular networks without the need of additional signalling loads and changes to both terminals and base station equipment ACKNOWLEDGMENTS The authors acknowledge the financial support of Enterprise Ireland and Motorola s European Cellular Infrastructure Division under grant AR/2/36 in the funding of the work reported in this paper REFERENCES [] VGarg, D Ness-Cohn, T Powers, L Schenkel, Direction for Element Managers and Network Managers, IEEE Communications Magazine, Oct 998 [2] J Zander, Radio Resource Management in Future Wireless Networks: Requirements and Limintations, IEEE Communications Magazine, Aug 997 [3] I Katzela, M Naghshineh, Channel Assignment Schemes: A Comprehensive Survey, IEEE Personal Communications, June 996 [4] M Perez-Carbonell, D Pesch, P Stephens Optimum Deployment in Cellular Mobile Networks using Genetic Algorithms, Irish Signals and Systems Conference, Maynooth, Ireland, June 2 [5] Peter T H Chan, Marimuthu Palaniswarni, and David Everitt, Neural Network-Based Dynamic Channel Assignment for Cellular Mobile Communication Systems, IEEE Trans Veh Technol, vol 43, pp 279-288, May 994 [6] Dietmar Kunz, Channel Assignment for Cellular Radio Using Neural Networks, IEEE Trans Veh Technol, vol 4, pp 88-93, Feb 99 [7] Enrico Del Re, Romano Fantacci, and Luca Ronga, A Dynamic Channel Allocation Technique Based on Hopfield Neural Networks, IEEE Trans Veh Technol, vol 45, pp 26-32, Feb 996 [8] Harilaos G Sandalidis, Peter P Stavroulakis, and Joe Rodriguez-Tellez, Borrowing Channel Assignment Strategies Based on Heuristic Techniques for Cellular Systems, IEEE Trans Neural Networks, vol, pp 76-8, Jan 999 [9] Nobuo Funabiki, and Yoshiyasu Takefuji, A Neural Network Parallel Algorithm for Channel Assignment Problems in Cellular Radio Networks, IEEE Trans Veh Technol, vol 4, pp 43-437, Nov 992 [] S Haykin, Neural Networks: A Comprehensive Foundation, Prentice-Hall, 994 [] Chiu Y Ngo, and Victor O K Li, Fied Channel Assignment in Cellular Radio Networks Using a Modified Genetic Algorithm, IEEE Trans Veh Technol, vol 47, pp 63-72, Feb 998 [2] Dirk Beckmann and Ulrich Killat, A New Strategy for the Application of Genetic Algorithms to the Channel Assignment Problem, IEEE Trans Veh Technol, vol 48, pp 26-269, July 999 [3] Ken Murray and Dirk Pesch, Adaptive Radio Resource Management for GSM using Neural Networks and Genetic Algorithms, IT & T Conference, Athlone, Ireland, Sep 2 [4] Goldberg, Genetic Algorithm in Search, Optimization and Machine Learning, Addison Wesley, 999