Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
Location Estimation and Mobility Prediction Using Neuro-fuzzy Networks In Cellular Networks

Location Estimation and Mobility Prediction Using Neuro-fuzzy Networks In Cellular Networks

Ratings: (0)|Views: 79 |Likes:
Published by ijcsis
In this paper an approach is proposed for location estimation, tracking and mobility prediction in cellular networks in dense urban areas using neural and neuro-fuzzy networks. In urban areas with high buildings, due to the effects of multipath fading and Non-Line-of-Sight conditions, the accuracy of positioning methods based on direction finding and ranging degrades significantly. Also in these areas, due to high user traffic there's a need for network resources management. Knowing the next possible position of user would be helpful in this case. Here using fingerprint positioning concept, after choosing appropriate parameters for fingerprinting in GSM cellular networks, MLP and RBF neural networks were used for position estimation. Then by the use of neuro-fuzzy networks a tracking and post-processing method is applied to estimated locations. For mobility prediction purpose the use of ANFIS neuro-fuzzy is implemented.
In this paper an approach is proposed for location estimation, tracking and mobility prediction in cellular networks in dense urban areas using neural and neuro-fuzzy networks. In urban areas with high buildings, due to the effects of multipath fading and Non-Line-of-Sight conditions, the accuracy of positioning methods based on direction finding and ranging degrades significantly. Also in these areas, due to high user traffic there's a need for network resources management. Knowing the next possible position of user would be helpful in this case. Here using fingerprint positioning concept, after choosing appropriate parameters for fingerprinting in GSM cellular networks, MLP and RBF neural networks were used for position estimation. Then by the use of neuro-fuzzy networks a tracking and post-processing method is applied to estimated locations. For mobility prediction purpose the use of ANFIS neuro-fuzzy is implemented.

More info:

Published by: ijcsis on Jul 07, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

07/07/2011

pdf

text

original

 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011
Location Estimation and Mobility Prediction Using Neuro-fuzzy Networks
In Cellular Networks
Maryam Borna
Department of Electrical EngineeringIran University of Science and TechnologyTehran, Iranmaryam.borna@gmail.com
Mohammad Soleimani
Department of Electrical EngineeringIran University of Science and TechnologyTehran, Iransoleimani@iust.ac.ir  
 Abstract-
In this paper an approach is proposed for locationestimation, tracking and mobility prediction in cellularnetworks in dense urban areas using neural and neuro-fuzzynetworks. In urban areas with high buildings, due to the effectsof multipath fading and Non-Line-of-Sight conditions, theaccuracy of positioning methods based on direction finding andranging degrades significantly. Also in these areas, due to highuser traffic there's a need for network resources management.Knowing the next possible position of user would be helpful inthis case. Here using fingerprint positioning concept, afterchoosing appropriate parameters for fingerprinting in GSMcellular networks, MLP and RBF neural networks were usedfor position estimation. Then by the use of neuro-fuzzynetworks a tracking and post-processing method is applied toestimated locations. For mobility prediction purpose the use of ANFIS neuro-fuzzy is implemented.
 Keywords-position estimation; neuro-fuzzy; prediction;cellular networks.
I.
 
I
 NTRODUCTION
Positioning in wireless networks is estimating a node'sdistance with reference to a fixed node or locating it by itsgeographical coordinates. Positioning is based on parametersused by mobile or fixed nodes for communication such asReceived Signal Strength (RSS), Time of Arrival (TOA) andAngle of Arrival (AOA). According to the type of wirelessnetwork and transmission protocols, different parameters areused for communication.Among several types of wireless networks, cellular phonenetworks due to increasing usage of cell phones for communications are more distributed with more subscribersso it can be said that one of the most probable items found ineveryone's pocket is his cell phone. Having locationinformation in cellular phone networks, various services can be provided based on user's location ranging fromcommercial and advertising services to routing, navigationand emergency calls. In cellular phone networks theseservices are referred to as Location Based Services (LBS).Also by locating a user's exact position network's resourcescan be efficiently managed and allocated leading to proper handover between cells and reduced co-channel interference.However for managing networks resources consumptionand reducing the costs of location update and call delivery procedures, prediction of user's next probable location can behelpful. This is done by analyzing some patterns of hismobility behavior. Therefore searching for users will be donein smaller groups of cells avoiding expensive queries toHome Location Register (HLR). This is also useful in other wireless networks such as Ad-Hoc networks for efficient bandwidth allocation and uninterrupted hand over betweenaccess points. Next sections are as follows: section II describes the problem of positioning in dense urban areas and relatedstudies in the literature. In section III proposed approach of this paper for positioning and mobility prediction isexplained and contains 3 subsections: fingerprint based positioning in subsection
 A
, post processing of estimated path in subsection
 B
and path prediction in subsection
arediscussed. Section IV includes the results of evaluating the proposed approach on database collected from GSM mobile phone network in city of Tehran. Results were discussed insection V and last section concludes the paper.II.
 
P
ROBLEM DEFINITION
 With developments in cellular phone networks differentmethods were considered for facilitating user positioning,such as Cell-ID, Cell-ID+TA, A-GPS, AOA, … the moreaccuracy increases the more expensive the deploymentwould be and the need for hardware and software changes in both cell phone device and network infrastructure rises.Moreover most of these methods are sensitive to Non Lin of Sight (NLOS) communication between transmitter andreceiver and multipath fading, conditions that dense urbanareas are involved with. Although everyday there are moreand more mobile phone devices equipped with GPS receiverswith positioning accuracy up to few meters, but in urbanareas with high buildings where it's less likely to have line of sight communication with at least 3 GPS satellites, or inside buildings where the signals attenuate significantly passingthrough the walls, positioning accuracy degradesconsiderably. In such cases there's a need for an auxiliarymethod to overcome these problems.
65http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011
The difficulty in these areas is the complexity of  propagation model of electromagnetic waves caused bymultipath fading, diffraction and scattering that makes it hardfor geometrical and statistical positioning methods relying onrelations between signal parameters and Tx-Rx separation.Fingerprint based positioning methods are better for mentioned cases [1] [2]. In these methods first a database of signal parameters in certain places is collected with noknowledge of propagation model of the environment and position estimation will be done upon these information and possible mapping among them.One way to find the mapping relations in the fingerprintdatabase is to use Artificial Neural Networks. Thesenetworks are able to estimate the complex nonlinear functions like mapping relations by parallel processing of neurons. Position estimation can be considered as a functionapproximation problem in neural networks that aims to findthe nonlinear mapping between inputs (fingerprints) andoutputs or targets (mobile phone's coordinates). Incomparison with other database lookup methods like K- Nearest Neighbor (KNN) that uses fingerprint parameters tofind its nearest Euclidean neighbors, neural networks are better. On the other hand since neural networks approximatefunctions and fingerprint details are somewhat related todelay and power loss of arrived signals and in turn these aredependent on Tx-Rx separation, it seems that neuralnetworks combine both features of RSS and TOA-TDOA based systems [3] [4]. Two common models of neuralnetworks in function approximation, multi layer perceptron(MLP) and Radial Basis Function (RBF) networks, are usedmore [2] [3] [5] [6].After user localization, history of his travelled places can be considered as a time series that by recognizing hismobility pattern, his next location can be predicted. Inliterature for user's path prediction in wireless networks withdifferent standards, Recurrent Neural Networks (RNN),Bayesian Neural Networks (BNN) or neuro-fuzzy networkswere employed that some used user's behavioral pattern in along period of time and different situations to learn hismobility pattern and similar users then predict their nextlocation [8] [9] [10].In this paper after gathering enough fingerprints withappropriate parameters in GSM cellular phone network, positioning of mobile phone device is done by searching for the best architecture for MLP and RBF neural networks in adense urban area. Afterwards using tracking and predictionfeature of ANFIS neuro-fuzzy network, the estimated path is post processed and user's upcoming path is predicted.III.
 
P
ROPOSED
A
PPROACH
 
 A.
 
 Fingerprint based positioning 
In fingerprint based positioning methods first a databaseof fingerprints in a certain area is collected. A fingerprint of acertain point includes particular information likegeographical coordinates of the point which is a specificationof that point. This information includes estimated signal parameters that are different depending on wireless network standards and data collection tool. Data collection and pattern learning are done offline before online real-time positioning.Input parameters of neural networks, fingerprint of a point, must be measurable, collectable and different from place to place. Here we aim to attain intended fingerprintsfrom information provided by mobile phone device withoutsoftware or hardware changes in the device and additionalsignaling between MS and BTS. The data can be obtainedfrom mobile phone routing table. This table is used for selecting the best cell to reside and is resorted every fewseconds. In GSM900 standard for mobile phone networks,this table contains a list of 30 radio channels (ARFCN)sorted in descending order based on received power. Inaddition to received power other parameters of currentlyselected and neighboring cells are available like cell name,absolute radio frequency number for broadcasting cell'sstatus, received power level, received signal quality andtiming advance (TA). Also the attributes of BTS antennas of each cell like its height and installation coordinates areaccessible.From the mentioned parameters those were chosen for fingerprint that own following properties:
 
Being sensitive to spatial changes. Thereforefixed parameters within a cell boundary likeradio channel numbers, cell antenna height andsimilar parameters are not suitable for fingerprinting.
 
Parameters should be representative of multipath fading effects in propagationenvironment. Received signal level, receivedsignal quality and Timing Advance (TA) aresuch parameters. However TA is a discretevalue of estimated BTS-MS separation with anaccuracy about 550 meters, say if TA=1, MS isin a radius of 550 meters from BTS and TA=2means MS is in a radius of 550 to 1100 metersfrom cell antenna. Hence in a cell with radiusless than 550 meters-like most urban cells- TA'svalue isn't much helpful in positioning.
 
For less signaling between mobile phone andBTS it's better to acquire fingerprint in IDLEmode rather that ACTIVE mode. TA andReceived signal quality are determined inACTIVE mode while RSS is monitored periodically even in IDLE mode.We chose parameters that fulfill mentioned requirementsnamely Received signal strength from cell antenna beside itscoordinates for serving cell and two of neighboring cells. Sothere are 9 parameters to be recorded in a single fingerprint beside the coordinates of data gathering location. Bycollecting these fingerprints in sufficient data points of thedesignated area, we have a suitable database for further analysis by fingerprint based positioning methods. Asdiscussed in previous section, for database processing, neuralnetworks predominates other methods so is employed here.Training set tuples are mentioned 9 parameters as input and
66http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, June 2011
latitude, longitude of respective data point as target or outputfor neural networks.One of the problems in neural networks design, particularly MLP networks is the lack of certain equationsfor determining the perfect architecture of the network andnumber of neurons in hidden layers. In the training phase of a NN while evaluating its ability to learn, its response to newuntrained data should also be considered for the network togeneralize well. In order that 84% of data set were used for training and the remaining for testing the trained network.For finding the best architecture for MLP NN first anupper limit regarding members of training set is consideredfor maximum number of network parameters i.e. totalnumber of weights and biases in neural network then bymodifying the number of hidden layers and neurons in eachlayer in the defined range, the architecture yielding less positioning error for training, testing and the whole data setis chosen. For available database the best architecture for MLP was one hidden layer with 23 neurons. Input layer neurons were set to 9 and output layer neurons for estimatedlatitude and longitude of mobile phone's location are 2. Fig. 1In standard Radial Basis Function NN a ruling parameter in the design is the radial neuron's spread that determines itssensitivity to the resemblance between network's inputs andweights. Searching for spread parameter resulted in value of 196 leading to less error for testing data set. The number of neurons has been set to its maximum i.e. the same as trainingset members. Fig.2Another type of RBF networks employed here isGeneralized Regression Neural Network (GRNN) that has afixed structure with little difference to standard RBFnetworks. Spread parameter for radial neuron has beenobtained like former case and set to 2.After position estimation with designed neural networks,there were rather big errors in few points probably caused byinadequate members of training set and unavailability of RSSin some points. To lessen this error we applied a post procession on estimated path coming in next subsection.
 B.
 
 Post-processing the estimated path
In this section we use ANFIS (Adaptive Neuro-FuzzyInference Structure) for processing the previously NNestimated path of user's travelled places. Employed neuro-fuzzy network is the neural network equivalent for SugenoFIS (Fuzzy Inference Structure). In comparison with MLP,ANFIS has a fixed architecture and no searching for beststructure is needed. It responds faster with less computationalresource consumption. Fig. 3
Figure 1. Proposed architecture for MLP neural network Figure 2. Proposed architecture for RBF neural network Figure 3. Neuro-fuzzy network for Sugeno's fuzzy inference structure
Inputs of ANFIS network were estimated path with 5delays and the same path with no delays as output. For initialFIS generation fed to ANFIS, we used subtractive clusteringthat the influence radius of every cluster for all 11dimensions of data was set to 0.5.ANFIS is used here to estimate the user's movementfunction and smooth the NN estimated path so it can bespecified which road the user is moving on, useful in maprouting and navigation purpose.
C.
 
 Predicting user's next location
Subsequent locations travelled by mobile phone user can be assumed as a time series. Here we used the predictionability of ANFIS network. The structure is the same as before. For training the network, 20% of the beginning of thetravelled path with 2 delays was selected as input and thesame path with one precession as output. The remaining 80%of the path was used for testing. In this way trained network would be able to predict next location by knowing the present and one previous location of user. Here we've usedthe estimated path by neural network in section III andcalculated the error with respect to real path.IV.
 
EVALUATION OF THE PROPOSED APPROACH
 For data collection we've used TEMS® drive tester toolthat is used for optimization and troubleshooting of mobile phone network by monitoring its status. It representsnetwork's data intercepted by mobile phone in a computer interface for further processing and also able to record datacollection point coordinates via GPS receiver.We've used this tool for fingerprint database collection inGSM communication network in city of Tehran for about250 data points. For network training and simulation we usedMATLAB® neural network and fuzzy logic toolboxes.
67http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->