Artificial Neural Network-Based Cellular Network Predictive System for Resource Allocation
Chapter One
OBJECTIVES OF THE RESEARCH
The objective of this research is to develop a cellular network predictive system that,
- provides the required QoS parameters values to the network subscribers at relatively affordable
- maximizes the utilization of cellular network resources and thereby maximizing revenue for the network providers.
- is capable of being integrated into a typical mobile wireless cellular network with
- responds to random network resource demands instantly and with
- delivers to the network providers a resource management system that is relatively simple, efficient and effective.
CHAPTER TWO
LITERATURE REVIEW
CELLULAR NETWORK RESOURCE ALLOCATION
Mobile cellular networks require adequate resources to enable the provision of quality services to subscribers. These resources could be bandwidth, channels, frequency, or power. However, the problem has always been that these resource are limited, scarce, expensive and sometimes inadequate [43, 49, 51]. Hence the need arises for the provision of appropriate resource allocation schemes to cater for these shortcomings.
Over the years, several researchers have proposed different resource allocation techniques that seek to proffer solutions that address the need for adequate provision and allocation of network resources. Some of the works are based on allocation of channels [43, 46, 50, 51, 58, 59, 64]. Some others are based on bandwidth [44, 49, 60, 63, 75]. The rest of the others are based on frequency and power [47, 49]. Different algorithms and techniques provides solutions to the problem of resource constraints in mobile networks [52].
ALLOCATION OF CHANNEL
The channel is a physical or logical medium through which signals or information can be transmitted from a source to a destination. Therefore, it is one of the most important resources in a cellular network [63]. The determination of adequate channel capacity and its efficient allocation is very important to the overall performance of a cellular network. The channel capacity is the rate of information that can be reliably carried or transmitted over a communication channel.
Several scholars have carried out researches into the allocation of this very important resource using different approaches. A modified Hopfield neural network that implements a channel assignment scheme that reduces call blocking or call dropping probabilities was proposed [43]. The channel allocation problem was formulated as an energy-minimization problem. The weights of the neurons were varied depending on the constraints conditions. The algorithm tested satisfactory in seven different cases by varying the frequencies. The performance of the algorithm proved to be better than existing algorithms.
Search and computational capabilities of genetic algorithms were used to encode a potential solution to a specified problem on a chromosome-like data structures and recombination operators were applied to these structures to extract critical information for the design of a channel allocation scheme for wireless networks [46].
Unused television white spaces (TVWS) were used to develop a model for channel availability based on co-channel interference [50]. The problem was formulated to maximize the number of allocated channels and improvement of total network throughput.
A hybrid channel allocation model using evolutionary strategy called D-ring to optimize channel assignment in wireless mobile networks was proposed [51]. The proposed model uses an efficient problem representation as well as an appropriate fitness function. The D-ring yields a faster running time with simpler objective function to obtain better results.
A minimum span problem (MS-CAP) using trial and error method for the effective allocation of channels in a cellular network was formulated [58]. The employed method minimized interference effectively in a cellular network.
Different channel allocation schemes were presented based on channel control strategies and complex scenarios [59].
In another study, a resource management scheme that is based on traffic patterns and penalty functions was proposed [64]. The scheme maximizes system utilization and control of the resources of a base station. In the scheme each class of traffic is assigned a pool of resources which is dynamically adjusted in accordance with the offered traffic load.
ALLOCATION OF BANDWIDTH
In a digital communication system bandwidth is usually referred to as an arrangement of a band of frequencies allocated to a signal in a transmission medium [138]. It can also be referred to as the amount of data that can be carried from one point to another in a given period of time [127].
The problem of bandwidth allocation in a multicell environment using Linear Programming (LP) methods was investigated [44]. The problem was formulated as a linear objective function with constraints to minimize access delay and to address resource allocation issue and optimization of bandwidth with QoS requirements.
In order to address the optimum utilization of limited bandwidth demands for channel allocation and for servicing incoming call requests, a dynamic channel allocation scheme for mobile cellular networks using particle swarm optimization (PSO) technique was presented [49]. Dynamic varying inertia was used to modify a velocity update function to improve system efficiency and utilization. Channel selection and allocation were obtained at a faster rate as compared to other algorithms and lower values of call rejection ratio were observed.
A phased solution of priority detection, mobility scheduling and effective bandwidth estimation was proposed by considering physical parameters [60]. Neural network based service model is incorporated to accommodate new metrics to process handovers and task scheduling. This mechanism provides the advantages in terms of choosing the tasks in a priority based scenario and providing un-interrupted service at the time of handoffs as well as an effective way of utilizing bandwidth.
CHAPTER THREE
DEVELOPMENT OF ANN PREDICTIVE MODEL
INTRODUCTION
In this Chapter, the processes leading to the development of the ANN based predictive model are laid out. The processes starts from the description of the area of study, through how the data was collected, pre-processed, trained and finally, to the choice of the appropriate model based on detailed analysis.
METHODOLOGY
The development of the ANN based cellular network resource allocation predictive system involves the following methodical steps: data collection, data pre-processing, data training, model development, model validation, and model deployment.
CHAPTER FOUR
SIMULATION RESULTS AND RESULTS ANALYSIS
INTRODUCTION
In this chapter, the results obtained from processes in chapter three are presented, analysed and discussed. These include the results from pre-processing, training, and the developed model.
CHAPTER FIVE
APPLICATION TESTING AND PERFORMANCE ANALYSIS
INTRODUCTION
In this chapter, statistical analysis carried was carried out to evaluate the performance of the developed predictive model. The model was tested with different inputs and targets to determine its versatility. Channel allocation tables were obtained for different grade of service and finally the channel allocation was compared with the sector clusters earlier established in section 3.2.2.4 and shown in Figures 4.1 – 4.8.
TESTING THE APPLICATION WITH DIFFERENT INPUT- TARGET SCENARIOS
The cellular network resource allocation predictive App was developed to predict future traffic load in the sectors within a network given a set of inputs and a required target. The predicted traffic is in turn used to propose the allocation of network resources to the sectors to meet QoS requirements.
The Application is subjected to different input-target scenarios which are different from the ones used in developing it to test its ability to produce the desired result. The results to be analysed will be the relationship between the target and predicted traffic for a given input.
To carry out the statistical analysis of the performance of the developed model in predicting future traffic for different input-target pair, two statistical tools were considered: linear regression and correlation tests. Starting with the input-target pair used in developing the Application (EngrMomInputs.data and EngrMomTargets.data), five other different input- target pairs were considered. The input-target pairs are: Input1_Target1, Input2_Target1, Input3_Target2, Input4_Target4, and Input5_Target4.
CHAPTER SIX
MODEL DEPLOYMENT
INTRODUCTION
Having built the resource allocation predictive model, it is now ready to be developed into a system. The system is packaged and deployed as a network resource allocation Application module that can be installed on any network system for the purpose of predicting future mean traffic of an existing network and the number of traffic channels required to adequately satisfy given QoS requirements. This task is carried out using the Graphical User Interface Development Environment (GUIDE) in MATLAB.
CHAPTER SEVEN
CONCLUSION
INTRODUCTION
In this chapter conclusions are drawn; recommendations made; main contributions of this work and direction for future work are outlined.
CONCLUSIONS
A cellular network resource allocation predictive system based on artificial neural networks was developed and presented. The system is capable of predicting future traffic in the sectors of a cellular network and determining the amount of channels to be allocated to the sectors to meet QoS demands. The predictive system was developed from historical data collected over a period of time from existing typical cellular networks. The developed predictive system is relatively simple, efficient and effective and can be readily used by cellular network service providers.
The performance of the developed model in predicting the future mean traffic in each sector was compared with some existing techniques. The MSE and MAE values for the techniques were respectively found to be: Single Tree (43.18, 3.70), Tree Boost (45.26, 3.51), Multilayer Perceptron (44.83, 3.81), General Regression Neural Network (35.35, 3.50), Radial Basis
Function (63.01, 4.92), General Method of Data Handling polynomial network (17616, 54.11),
Support Vector Machine (40.43, 3.20), Gene Expression Programming (26.41, 3.13), while that of the developed ANN Model was (1.60, 1.31). These results show that the prediction capability of the developed model was superior to the existing techniques.
Using the predicted mean traffic and applying 1% blocking probability as a QoS parameter the ANN Model computes the traffic channel(s) to be allocated to each sector. Typical results for predicted traffic and channel to be allocated are given for the first 10 sectors as follows: (3.0547, 8); (2.7234, 8); (2.5671, 8); (5.7998, 12); (3.1889; 9); (4.4528, 10); (3.6770, 9); (3.6109, 9); (2.7586, 8); (2.6653, 8).
Finally, the model was packaged as an Application that fitted in easily into a cellular network system and it was successfully used to predict the number of channels needed to service a given sector based on required QoS requirements. The developed Application flexibly adapted to changes in input data, be it hourly, daily, weekly or monthly.
CONTRIBUTIONS TO KNOWLEDGE
This research work delivers a resource management system that is relatively simple, efficient and effective for incorporation into a cellular network to the providers. This novel system is intended to be patented, commercialised and made available to network providers.
This resource management system is unique because it takes different parameters simultaneously and evaluates each of them implicitly to detect complex nonlinear relationships between dependent and independent variables. It also detect possible interactions between predictor variables in relation to a desired response. The implementation of this system does not require the use of complicated algorithm as such it integrates easily into an existing system with less payload.
Also, this work provides a cost effective means of predicting future traffic load and channels required to service the load in a cellular network to meet QoS requirements. The cost of the Application is put at N481,400 equivalent to $3,000. This is relatively cheap.
LIST OF PUBLICATIONS
The following list of publications during the course of this work has added to the existing body of knowledge.
- Joseph M. Môm, Nathaniel S. Tarkaa and Cosmas Ani, “The effects of propagation environment on cellular network performance,” American Journal of Engineering Research, 2, Issue 9, Sept 2013, pp.31 – 36. e-ISSN 2320-0847, p-ISSN 2320- 0936. (http://www.ajer.org).
- Joseph M. Mom and Cosmas Ani,”Application of self-organizing map to intelligent analysis of cellular networks”, ARPN Journal of Engineering and Applied Sciences, Vol. 8, No. 6, June 2013, pp. 407 – 412. ISSN 1819-6608. (http://www.arpnjournals.com/jeas/research_papers/rp_2013/jeas_0613_896.pdf).
- Joseph M. Mom and Cosmas Ani, “An integrated block-oriented simulation model for estimating cell loss rate in ATM networks”, Pacific Journal of Science and technology, Vol. 13, No. 1, May 2012, pp. 287-291. (http://www.akamaiuniversity.us/PJST13_1_287.pdf).
- Tarkaa, N.S., Mom, J.M., and Ani, C.I., “Drop Call Probability Factors in Cellular Networks”, International Journal of Scientific & Engineering Research, 2, issue 10, October 2011, pp. 1 – 5. ISSN 2229-5518.
DIRECTIONS FOR FUTURE RESEARCH
The research embodied in this work can further be extended to include the prediction of more resources such as power, bandwidth, radio link, into the predictive model to make it more robust. Another possible area with be to develop a call admission control algorithm to dynamically predict and allocate the resources based on the established methodology in this work.
REFERENCES
- https://clients.txtnation.com/enteries/301118-All-Countries-List-Of-Mobile-Operators-By-Country-slow-loading
- Wikipedia, List of mobile network operators. [Online]. Accessed 8th March, 2014. http://en.m.wikipedia.org/wiki/List_of_mobile_network_operators.
- Nigerian Communications Commission, Quarterly Summary of telecoms Subscribers in Nigeria, [Online]. Accessed 6th June, 2014. http://www.ncc.gov.ng/
- National Population Commission of Nigeria, National and State Population and Housing Tables: 2006 Census Priority Tables (Vol. 1). [Online]. Accessed 8th March, http://www.population.gov.ng/
- Population Reference Bureau, 2014 world population data sheet. [Online]. Accessed 8th March, 2014. http://www.prb.org/publications/Datasheets/
- International Telecommunication Union, Recommendation E.507: Models for forecasting international [Online] Accessed 4th March, 2014. www.itu.int/rec/T-REC-E.507-198811-I/en
- G. Fraimis, “Channel allocation techniques,” Wireless Telecommunications Lab, Department of Electrical and Computer Engineering, University of Patras, Greece.
- -S. Chen and C.R. Dow, “Fixed channel allocation scheme performance enhancement for cellular mobile systems,” IEE Proceedings – Communications, Vol. 149, issue 4, pp. 232 – 236, 2002.
- Lin, J.Z. Wang, C. Liang and D. Qi, “A threshold-based dynamic resource allocation scheme for cloud computing,” Procedia Engineering, 2011, Vol. 23:695-703, doi:10.1016/j.proeng.2011.11.2568
- Ramanathan, K.M. Sivalingam, P. Agrawal and S. Kishore, “Dynamic resource allocation schemes during handoff for mobile multimedia wireless networks,” IEEE Journal on Selected Areas in Communication, Vol. 17, Issue 7, pp. 1270 – 1283, 2002.
- Xin, T. kai, S. Yan, L. jianghua and Y. Sixiang, “Performance analysis of dynamic bandwidth allocation scheme based on QoS in satellite networks,” Wireless Communications and Signal processing (WCSP), 2010
- Ingo, “Resolving SON interactions via self-learning prediction in cellular wireless networks,” International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM), Shanghai, China, 21st – 23rd Sept. 2012.
- Fazio, and S. Marano, “A new Markov-based mobility prediction scheme for wireless networks with mobile hosts,” International Symposium on Performance Evaluation of Computer and Telecommunication Systems (SPECTS), Genoa, Italy, 8th– 11th July, 2012.
- Makhtar, D.C. Neagu, and M. Ridley, “Predictive model representation and comparison: Towards data and predictive models governance,” UK Workshop on Computational Intelligence (UKCI), 2010.
