R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                    Applications (IJE...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                    Applications (IJE...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                                     ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                       Applications (...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                    Applications (IJE...
R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and                     Applications (IJ...
Upcoming SlideShare
Loading in …5
×

Ni2421862197

210 views

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
210
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
2
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Ni2421862197

  1. 1. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197 Evaluating the Probability of Channel Availability for Spectrum Sharing using Cognitive Radio R. Kaniezhil*, Dr. C. Chandrasekar** *(Department of Computer Science, Periyar University, Salem-636011) ** (Department of Computer Science, Periyar University, Salem - 636011) ABSTRACT discussed. When users of different service providers Spectrum sharing between service share the licensed spectrum of the primary user, providers improves the spectral efficiency, primary user has the highest priority to access that probability efficiency of sensing and reduces the spectrum. In such a case, the secondary user has to call blockage. If the number of service providers vacate the channel and to move on to another is increased to share the spectrum, this may available channel. So that, transmit power will be reduce the high traffic patterns of the calls. In reduced frequently or communication will be this paper, we present an approach for channel dropped. In order to avoid the temporal connection availability and call arrival rate. The results are loss or interference with the primary user, the used to evaluate the probability of the channel secondary user has to evaluate the channel availability of a frequency band within a time availability before using the channels of the primary period. In this work, we propose an algorithm user and predicting the traffic pattern of the primary for predicting the call arrival rate which used to user. This would increase the channel utilization and predict the traffic process. reduces the call blockage and interference. The paper structure follows: In Section II & III,Keywords - Cognitive radio, Traffic forecasting, Related work and various traffic models are brieflyTraffic model, SARIMA, Spectrum sharing, discussed. Traffic pattern prediction is discussed inSpectrum utilization section IV. Probabilities of channel availability, Prediction of call arrival rate are discussed in sectionI. INTRODUCTION V and VI. In section VII and VIII, the performance To promote licensed parties to share their measures of Spectrum sharing and simulation resultsnon-utilized resources in 2002 FCC issued 98-153 are presented. Finally, we draw our conclusions indockets, permitting many users to transmit on single Section IXchannel, using low power communication based onUltra Wide Band (UWB) communication. Recently II. RELATED WORKreleased FCC docket 03-122 revisited rule 15, In [2], presented a deterministic fluid modelallowing wireless data users to share channels with and two stochastic traffic models for wirelessradar systems on a LBT basis. Finally FCC realized models which provide each cell an infinite numberthat Cognitive Radio (CR) techniques are the future of channels such that no call blocking occurs. Itsubstrate that stimulate full growth of “open discusses about the Markovian traffic model withoutspectrum (see FCC docket 03-108 on CR Poisson arrivals and made connection thetechniques and FCC docket 04-186 on CR in TV deterministic fluid model, nonhomogeneous Poissonspectrum). process, etc. In addition to this, the paper [3], discussStatic spectrum assignment, applied to radio about different traffic models in Broadbandfrequencies for almost a century, leads to a so called Networks. Similarly, the traffic model built onquasi-scarcity of the spectrum. It would be thus historical data for the secondary users to predict thelogical to allow unlicensed users to exploit primary user‟s traffic pattern for the future use [4].dynamically (opportunistically) licensed frequencies In paper [5], classification method and predictivewhen they are free (to minimize interference) at a channel selection method outperforms opportunisticspecific place and time. Theoretically, such approach random channel selection both with stochastic andwould increase overall frequency reuse and would deterministic patterns.boost the throughput for applications thatopportunistically use the empty frequencies. Thisway of spectrum access will be called throughout III. TRAFFIC MODELS Additional spectrum requirement dependsthis paper Opportunistic Spectrum Access (OSA). on various parameters like number of subscribers,In this paper, different traffic prediction techniques the density of subscribers, terrain, pattern of trafficand the process of evaluating the channel availability (voice, data, etc.), deployment of variousfor primary users in cognitive radio systems is technological means to improve the efficient 2186 | P a g e
  2. 2. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197utilization of spectrum, the technology itself, etc. traffic models include artificial neural network, AutoWhen using other multiple-access techniques, such Regressive (AR), Auto Regressive Moving Averageas OFDM, co-channel interference must be avoided. (ARMA) with time varying coefficients, AutoTo satisfy this requirement, the dynamic-spectrum Regressive Integrated Moving Average (ARIMA),management algorithm must include a traffic model Seasonal ARIMA (SARIMA), Transform Expandof the primary user occupying a black space. The Sample (TES), and Discrete Auto Regressive (DAR)traffic model, built on historical data, provides the models. These models are used to predict traffic datameans for predicting the future traffic patterns in that in Ethernet, Internet, etc.space. This in turn, makes it possible to predict theduration for which the spectrum hole vacated by the Forecasting usually works in the following way.incumbent primary user is likely to be available for First, the historical data are analyzed in order touse by a cognitive radio operator. In a wireless identify a pattern that can be used to describe timeenvironment, two classes of traffic data patterns are series. Then, this pattern is extrapolated, ordistinguished, as summarized here [1]. extended, into the future in order to prepare a1) Deterministic patterns. In this class of traffic data, forecast. The validity of forecasting rests on thethe primary user (e.g., TV transmitter, radar assumption that the pattern that has been identifiedtransmitter) is assigned a fixed time slot for will continue in the future. A forecasting techniquetransmission. When it is switched OFF, the cannot be expected to give good predictions unlessfrequency band is vacated and can, therefore, be this assumption is valid. If the data pattern that hasused by a cognitive radio operator. been identified does not persist in the future, this indicates that the forecasting technique being used is2) Stochastic patterns. In this second class, the likely to produce inaccurate predictions. Then,traffic data can only be described in statistical terms. changes in pattern of data should be monitored soTypically, the arrival times of data packets are that appropriate changes in the forecasting systemmodeled as a Poisson process; while the service can be made before the prediction becoming tootimes are modeled as exponentially distributed, inaccurate.depending on whether the data are of packet-switched or circuit-switched kind, respectively. In the following subsections some of the traffic prediction techniques are discussed. In any event, the model parameters of stochastictraffic data vary slowly and, therefore, lend A. ARIMA Based Traffic Forecastingthemselves to on-line estimation using historical ARIMA and SARIMA models aredata. Moreover, by building a tracking strategy into extensions of ARMA class in order to include moredesign of the predictive model, the accuracy of the realistic dynamics, in particular, respectively,model can be further improved. nonstationary in mean and seasonal behaviors. In practice, many economic time series areIV. TRAFFIC PATTERN PREDICTION nonstationary in mean and they can be modeled only Different methods and models are proposed by removing the nonstationary source of variation.inorder to forecast the traffic on different The general ARIMA model contains autoregressiveheterogeneous networks. An accurate traffic (AR) and moving average (MA) parts and explicitlyprediction model should have the ability to capture includes differencing in the formulation of thethe prominent traffic characteristics, e.g. short and model. The model parameters are: the autoregressivelong dependence, self similarity in large-time scale parameter (p), the number of differencing passes (d),and multifractal in small-time scale. For these and the moving average parameter (q). ARIMAreasons time series models are introduced in network models are classified as ARIMA (p, d, q). Thetraffic simulation and prediction. Traffic models can parameter d is restricted to integer values. Inbe stationary or nonstationary. Stationary traffic FARIMA, this differencing parameter is consideredmodels can be classified into two main classes as a fraction.namely [3]: ARIMA models can be used with some kinds of Short-range Dependent (SRD) non-stationary data which is useful for series with Long-range Dependent (LRD) stochastic trends and however, they cannot be applied to predict the network traffic whichShort-range dependent models include Markov possesses the Long Range Dependent (LRD)processes and Regression models. These traffic characteristics. Traditional AR and ARMA modelsmodels have a correlation structure that is significant are used to predict the high speed network trafficfor relatively small lags. Long range dependent data and also captures traffic of the short rangetraffic models such as Fractional Autoregressive dependent (SRD). ARIMA processes work on non-Integrated Moving Average (F-ARIMA) and stationary (chaotic) data. ARIMA models are foundFractional Brownian motion have significant suitable for prediction and stochastic simulator.correlations even for large lags [6]. Nonstationary Different variations of ARIMA models e.g. simple 2187 | P a g e
  3. 3. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197ARIMA, ATHENA, SARIMA, subset ARIMA are particular, seasonal variations of the ARIMA modelpopular in the short-term traffic forecasting literature perform better than linear regression, historical[9]. average, and simple ARIMA [9].Fractional ARIMA (FARIMA) model has the ability C. Artificial Neural Network Based Trafficto capture both SRD and LRD characteristics of Forecastingtraffic and to capture the self-similarity of network One particular class of non-linear models istraffic. It is time-consuming one to discuss a two neural networks (NNs). NNs are unlike other non-stage predictor. Detecting the trend with an ARIMA linear time series models, in that there is usually nomodel is implicit i.e., it cannot calculate the exact attempt to model the innovations. NN modeling isslope of the trend line. nonparametric in character and the whole process can be automated in a short space of time. NeuralB. Seasonal ARIMA Based Traffic Forecasting network based traffic prediction approach is A simple ARIMA model is made up of complicated to implement. The accuracy andthree parts, „AR‟ i.e. autoregressive part, „I‟ i.e. applicability of the neural network approach indifferencing part, „MA‟ i.e. moving average part. traffic prediction is limited [9]. Artificial NeuralThe model introduces into the ARIMA model a Network (ANN) can capture the non-linear nature ofseasonal period parameter (S), a seasonal network traffic and the relationship between theautoregressive parameter (P), the number of seasonal output and input theoretically and however, it mightdifferencing passes (D), and a seasonal moving suffer from over-fitting. The Machine learningaverage parameter (Q). The equation representing an technique called Support vector machine (SVM)ARIMA (p, d, q) model for a time-series sequence used to forecast the traffic in WLANs. This SVM Yt (t  1,2,... n) is applied to pattern recognition and other applications such as regression estimation. There are two ( B)(1  B)d Yt  ( B)Zt drawbacks for ANN [11]: one is that ANN is (1a) essentially a black box; the other is that its network structure and the weights between neurons are fixed and cannot be adjusted adaptively during the Where ( B)  (1   1 B   2 B 2  ..... p B p  forecasting process. (1b) D. Mean Square Based Network Traffic Forecasting 2 2  q And ( B)  (1  1 B  B  .... q B ); The mean square error is an absolute error measure; therefore, it is highly influenced by the (1c) amplitude of the predicted trace. In order to eliminate this undesirable effect without losing itsWhere Zt is the white noise sequence and B is the discriminating capability, a relative error measure isBackshift operator. highly recommended. Mean square predictor which requires matrix inversion and autocorrelationA SARIMA (p, d, q) × (P, D, Q)s model can be computation and which is based on recursive linearrepresented as [8]: regression can eliminate these time-consuming computations in the expense of decreased accuracy. s s D d s ( B )( B)(1  B ) (1  B) Xt  ( B )( B)Zt Normalized Least Mean Square (NLMS) based prediction approaches are of particular interest due (2) to its simplicity and relatively good performance. They are suitable for on-line Real-time variable bitWhere (B) and (B) represent the AR and MA rate (VBR) video traffic prediction. s sparts, ( B ) and ( B ) represent the Seasonal AR E. Other Traffic Forecasting Methodsand Seasonal MA parts respectively. B is the A new forecasting technique called theBackshift operator. extended structural model (ESM) derived from the basic structural model (BSM). The ESM model isSome SARIMA models though found not the best constructed in such a way that extra parameters arefit, perform very well compared with the best fit. estimated to minimize the mean absolute percentageSARIMA models can capture the daily repetitive error (MAPE) of the validation sequence [12]. Thenature of traffic flow and the dependence of present ESM model shows an improvement in MAPE of thetraffic conditions on the immediate past. The paper test sequence over both the BSM and seasonal[7] shows that the Bayesian inference of SARIMA autoregressive integrated moving average. Themodels provides a more rational technique towards improved prediction can significantly reduce the costshort-term traffic flow prediction compared to the for wireless service providers, who need tocommonly applied classical inference [10]. In 2188 | P a g e
  4. 4. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197accurately predict future wireless spectrum service also varies with the primary user traffic [22].requirements. The amount of service that can be squeezed in fromExponential smoothing is an intuitive forecasting the free bands in a spectrum accessed by unrestrictedmethod that weights the observed time series primary users is called the capacity of secondaryunequally. The major advantage of exponential users.smoothing methods is that they are simple, intuitive,and easily understood [11]. Generally, exponential In order to assess the communication environment,smoothing is regarded as an inexpensive technique all communication channels may be monitored atthat gives good forecast in a wide variety of least once and preferable several times per second.applications. In addition, data storage and computing Through repetitive monitoring of the communicationrequirements are minimal, which makes exponential spectrum, an accurate history of channel usage cansmoothing suitable for real-time application. The be developed. Once an accurate usage history ismajor disadvantage of exponential smoothing established, predictive methods can be used to selectmethods derives from its basic premise about the a channel likely to be available. In one aspect of themodel: the level of time series should fluctuate about present disclosure, the channel selection process isa constant level or change slowly over time. When based on evaluation of available frequencies basedthe time series takes on an obvious trend, even on three criteria:adaptive exponential smoothing methods will fail togive good forecasting [11].  The most recent spectrum activity scan--If a channel shows activity in the current scan, it isThe traffic model using a fuzzy logic approach does presumed to be busy for the next operational periodnot require huge data sets and does not make any and is excluded from selection for use.statistical assumptions [13]. It is widely used in time  Database of excluded frequencies--If aforecasting. It is relatively simple and quick to channel falls within a range of excluded frequencies;compute. Hence it is fast and secure. it is excluded from selection for use.  Availability prediction based on historicThe ARIMA/GARCH traffic prediction model can use--The channel prediction algorithm attempts tobe used to build effective congestion control reduce the chances of interference by steering theschemes, e.g. dynamic bandwidth allocation and channel selection to those frequencies which areadmission control. least likely to show activity by other stations in the next operational period.V. PROBABILITY OF CHANNELAVAILABILITY The proposed work follows the coordination in a distributed manner. The goal of coordination in A. Channel Availability Evaluation spectrum sharing is to distribute concurrent, The Channel availability talks about the conflicting links across channels to avoidchannel assigned for CDMA and GSM networks. interference and improve throughput. The controlSpectrum usage deals with the availability of the free messages carry information of traffic load, availablechannel which is not used by the primary users. The channels, and usage on each channel. The keyPrimary users are said to be a licensed owner of a challenge in this module is how to addressfrequency band and the one who uses the spectrum heterogeneity in channel availability and traffic loadopportunistically for communication of the licensed during channel selection.user are called Secondary users. Secondary userssense the presence of primary user with the help of B. Evaluating the probability of ChannelCognitive radio. It tunes the spectrum band or Availabilitychannel which is not currently in use for Traffic pattern prediction enables thecommunication. If the primary user returns to the secondary users to estimate the channel availabilitychannel in which the secondary user is active, then and channel utilization. There are two factorsthe secondary user has to vacate the channel. This considered in the traffic pattern prediction in thetype is called as a forced termination. Then the voice communications. They are call arrival rate andsecondary user has to shift to another available call holding time. In order to estimate the call arrivalchannel and continues its process. Sometimes this rate and call holding time of the primary user thattype of shifting may leads to interruption in uses the channel. According to the prediction orcommunication. Inorder to overcome this difficulty, estimation results, the secondary users are able tothe secondary user has to predict the channel evaluate the channel availability for a given periodavailability of the primary user and then they have to of time. The secondary user can decide to use thego utilizing the channel in an efficient manner. channel with the help of the threshold value with the estimated or evaluated probability.Since the availability of the channel for the In a traditional way of evaluating the call arrivalsecondary user depends upon the traffic of the process of a wired or wireless networks was doneprimary user, number of the secondary user has been with the help of Poisson process. In a Cognitive 2189 | P a g e
  5. 5. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197radio system, the number of subscribers licensed to a A(t) (for t  0 ) be the number of call arrivalsservice provider is low or the call request/ call arriving in the time interval[0,t] with A(t) = 0. Theattempt is found to be low. In such cases, the rate parameter of the call arrival process is A(t) issecondary user can easily share the available a(t) where a(t) may change over time. The expectedspectrum among them. call arrival rate between time t1 and t2 is given asWe consider the number of call arrivals of primary t2user within a period of time t is limited. Hence,considering the total number of call arrivals of At   a(t )dt [ P( A  t 2 )]primary user is NA within t, then the call arrival t1would follow the Binomial distribution i.e., (4)X  Y( N A , P) where P is the probability of theservice request of an user. The number of call arrivals At to occur during any time interval of length t, then it will be givenThe mean E of binomial distribution is NA.P. using Poisson process as [14]Hence, P will be given as e  t ( t ) n E P[ A(t )  n]  Pn (t )  P (1) n! NA (n=0,1,2…) (5) t Since At is Poisson with parameter t ; an average    t (2) NA of t arrivals occur during a time interval of length t, where t is the number of user arriving i.e., so  be the average number of call arrivals per unitcall arrival rate and t is the mean value of the call time or the call arrival rate. tarrival. From equ (2), t = . E is the mean If the number of call arrivals At to occur within the NA time interval (t  t ) which follows a Poissonnumber of call arrivals for a given time period t and  refers the mean number of call arrivals per unit distribution with parameter  t , t  t , i.e.,time. This gives the call arrival rate of the primaryuser. P[ A(t  t)  A(t)  n]  P[ A(t)  n]We have to calculate the probability of the number   t ,t  tof primary user arrival in certain time duration. For e ( t ,t  t ) nthis, we can decide that the arrival of users within a  (n=0,1,2..) n!time t and not focusing on the waiting times, but (6)concentrating on the number of the users. We canassume that the probability of users in a given time t Thus, (6) gives the chance of call arrival rate in theas [16-18] time interval [t , t  t ] is Poisson distributed with mean equal to the length of the interval. Also, the  NA  0 NA 0 Ppu    0  P (1  P)  probability of no arrival during the interval   [t , t  t ] is given as 1  t  o( t )  (1  P) N A and the probability of more than one arrival t NA occurring between [t , t  t ] is  (1   ) NA (3) o( t ) Where N A is large. (3) is an approximated one Here, Network user‟s behavior may be measured asgot from the Poisson distribution. the time of calls, the average length of the call or the number of calls made in a certain period of time. ForApply non-homogeneous Poisson process to Telecommunication companies, they often use callconsider the call arrivals of the primary user. Let inter- arrival time and call holding time to calculate 2190 | P a g e
  6. 6. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197the call blocking rat, Interference and to determine  n 0the spectrum utilization efficiency. In the analysis of t    e td  nt  nnetwork traffic, the call inter-arrival time are t  t  d exponentially distributed, while the call holding time P  e td (10)fits a lognormal distribution. 0!Generally, a traffic period will be divided into 24 It is difficult to predict the ongoing call holding timetime intervals. The number of user calls is of a secondary user. This can be taken in a randomconsidered as the important calling behavior pattern manner. Hence in (10), the call holding time isin the voice network [15]. A general metric followed replaced with the average call holding time of thein the telecommunication industry is the hourly secondary user and calculate t based on the totalnumber of calls. Hence, the 24 time intervals will be number of calls within a time duration andconsidered as cumulative total call holding time. (t n , t n  1 ) (n = 0,1,2,…23) In the following discussion, we can consider the total number of arrivals of primary user NA for theThen the time duration will be t d i.e., secondary user. For this, we have to consider three approaches inorder to predict the average call t  t n 1  t n (7) holding time t of a secondary user. This t is ` d calculated using the historical calls and it does not depend upon the last call.for one time interval i.e. one hour. Thus, the call rate n Approach 1: Primary user ends its transmission atparameter (t ) will take the constant value in td one time t1 and secondary user starts its transmission at the same time t2 i.e.,each interval of time (t n , t n  1 ) i.e., t1= t2 n ( t )  [t  t n , t n  1 ] (8) Here, the secondary user has to estimate the td probability of the arrival of the primary user within the time t . Thus, the probability P0 of no primarywhere  n is the total number of call arrivals in the user‟s arrival within the time t is given bytime interval (t n , t n  1 ) . Consider the number of call arrivals within the P0 (t)  e  t (4)time interval t n  t  t  t  t n  1 i.e., t and t are Approach 2: The arrival of users between time t1=0within the same time interval (t n , t n  1 ) . Hence, and t2 =1, not by focusing on the waiting time butthe expected call rate t is given as concentrating on the number of arrivals of the user. This approach does not tell immediately when the n users arrive, but it only provides the information that  t , t  t  t (9) in a given time interval the number of customers td should have a Poisson distribution, with a parameter which is proportional to the length of the interval If we consider, (4) and (8), we get call arrivals i.e., it must be non-decreasing. The number ofwith different time intervals and period of time. This arrivals that have occurred in the interval (t1, t2) andprovides the number of call arrivals of the primary it is given byuser within the time interval (t n , t n  1 ) . From this t1<t2we can predict or evaluate the channel for thesecondary user. From (7), we get the call arrival Here, the primary user ends its transmission at t1 andrates of the primary user. If the secondary user finds secondary user starts its transmission at t 2. Thethe channel to access, it plans to start its probability that there is no occurrence before time ttransmission over this channel. Before that, it  tpredicts the call arrival of the primary user in the is, according to the current approach, equal to e .current time interval. If no primary user occupies the Hence, the probability P0 of the no arrival of thechannel during the time interval, it has to evaluatethe probability of no primary user for call holding primary user will be given bytime. From (6), we can generate the probability as 2191 | P a g e
  7. 7. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197 VI. CALL ARRIVAL RATE PREDICTION P t t 0 1 2 To estimate the call arrival rate of the primary user within a given time, secondary users P0 (t)  P0 (t) has to predict the number of call arrivals within a given time. The call arrival rate  is 18calls/second P0 (t)  e t and it is illustrated in the Fig. 1. The licensed (5) frequency bands can be used to carry voice or data in a cognitive system. The traffic pattern of primaryWhere (t 1+t2) = P (no occurrence before t1 +t2) users may vary with application. Taking the time interval into account, within a given period and the = P (no occurrence before t1) P (no corresponding number of calls in that interval and set of number of call arrivals at various time occurrence between t1 and t2); intervals are also observed using discrete time series method.Approach 3: In this approach, the primary user startsits transmission at t0 and ends at t2. The secondaryuser intends to start its transmission at t1. This will begiven as t0<t1<t2We have to assume that the secondary user holds thecall for some time otherwise it would drop the callTh. In this case, the event that representing thecurrent primary user vacate the channel within T h isX and no primary user arrives within the time t isY. when the probability P1 will be available withinthe time duration Th + t and given as P1 = Probability{X and Y} = Probability{X}.Probability{Y} f (t)  (t) k  1 e tThus, Fig. 1. Call Arrival Rate Process Th  t 1  t 2  t By using this, secondary users can able to predict the P1   f (t )dt   (6) number of call arrivals during a given time period. 0 So that, it is possible to predict the call arrivals of primary user in the future and secondary users canWhere f(t) represents the probability density able to predict the number of call arrivals of thefunction (pdf) call holding time of the primary user. primary user also. This can be predicted using the SARIMA model as a one-step prediction of numberIt is hard to obtain practically the probability density of call arrivals at a given time period.function (pdf) for the secondary users. Hence, thesecondary user has to estimate the call holding time From Section 4.2, consider the (2), A SARIMA (p,of the current primary user and probability of no d, q) × (P, D, Q)s model can be represented asprimary user arrival within time t . Then only thesecondary user has to proceed, otherwise it has move ( Bs )( B)(1  Bs ) D (1  B)d Xt  ( Bs )( B)Ztto the other available spectrum or to drop the call.It is difficult to predict the mean number of callarrivals per unit time and to estimate the call holding Where  ,  , P, D, Q represents the seasonaltime of the primary user. This will be explained in counterparts of  ,  , p, d, q and S is the seasonalthe following section of predication of call arrivalrate. autoregressive orders. From this  t is given as 2192 | P a g e
  8. 8. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197 Yt  (1  B)d (1  Bs ) D t (7) From the simulated data field of  t , with a period of s=24, Yt corresponds to Yt  (1  B)(1  B24 ) t (8) Where d=1 and D=1. Equ (8) becomes Yt   t   t  1   t  24   t  25 (9) Thus,  t becomes  t  Yt   t  1   t  24   t  25 (10) In practice, by applying the h step prediction, the linear predictor of Yn+h, which is the best in the sense of minimizing the h-step (10) is given as [19 - Fig. 2. Call Arrival Rate Prediction 21] Fig. 2 gives the call arrival rate prediction of the  n  h  Yn  h   n  h  1   n  h  24   n  h  25 (11) original calls and the forecasting (primary) calls. The call arrival rate predication is based on the mean The best linear Prediction Pt is the calculation of the square error. It is an absolute error measure with respect to the predicted traced variance. This figure estimate of the future values Yn  h (h>0) based on visually demonstrates that the proposed predictor is the current information [21]. Pt is given as follows capable of tracking the input trace in spite of from (11) variations.Pt  n  h  Pt Yn  h  Pt  n  h  1  Pt  n  h  24  Pt  n  h  25 (12) After calculating the best linear predictor of ARMA process Yt, we can compute the prediction Ptn h of Yt. For one step prediction, taking h=1, then (12) becomes Pt  n  1  Pt Yn  1   n   n  23   n  24 (13) Fig. 3. Call Arrival Rate Prediction with Trend Fig. 3 shows the call arrival rate prediction with changing trend value. Here, the prediction error is much greater than that of the Figure 2. This is the change made during the call arrival rate with 2193 | P a g e
  9. 9. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197increased or decreased trend in a given time period. c(i) =c(i) /E (i) = α ( i ) .E P .t (i) /E (i) = α ( i ) .t (i) .ηs(sp) eThe traffic trend changes significantly if there is achange in the time or space. The traffic trendchanges are done in a long term process and this is where α(i ) is the unit price ($/second/channel) forcarried out by means of considering the trend service provider sp and c(i) is the average incomechanges in the two or more consecutive periods within the observation time.which helps to track the call arrival rate. E. Free Spectrum CalculationVII. PERFORMANCE METRICS Free Spectrum will be calculated using the In this section, we look into the allocated and limited allocated spectrum values.performance metrics which are used to evaluate the Number of n users in service providers calls in thetraffic prediction of spectrum sharing between system at time t will be negligible when comparedservice providers. The metrics includes block with overall performance [23-24].counting, service unavailability, sensing efficiencyand free spectrum calculation.A. Call Block Count VIII. SIMULATION RESULTS OF TRAFFIC The blocking probability for CR nodes PREDICTION In this section, Using NS2 simulation the RBL is defined as the ratio of total blocked calls (or performance of the overall system efficiency hasspectrum requests) over total calls processed by all been evaluated. The call arrivals are modeled usingservice providers. the Poisson distribution, while the call holding times n (total) (t) are exponentially distributed. BL R BL =lim (total) Fig. 4 predicts that in the simulation, two base n processed (t) stations are defined (Node 0 and 1). Blue color node indicates the currently accessing the service provider of the spectrum by the BS. Orange color nodes indicate that the nodes are already accessed theB. Service Unavailability (Spectral Efficiency) service in previous time periods. Higher Spectrum efficiency is anticipatedcompared to service provider systems because theblocking count of user system is lower. Thus morecalls can contribute to the maximize channelutilization. t (sp) n 1 n (t) =lim  nbusy n (sp) η s dt t 0 N ch-total (t) (sp)C. Probability Efficiency ( System Efficiency) Probability efficiency metric for serviceprovider is determined by the processed trafficintensity and the total traffic loaded to serviceprovider within the observation time. E pi ) (  (i ) sys  Eini ) ( Fig. 4. Current accessing of the spectrum Fig. 5 represents the service access between BS and mobile devices are carried out dynamically. Due toD. Revenue (Cost) Efficiency mobility, nodes are moving in the network by Within the observation time, cost is choosing random position in topography and thedetermined by the number of processed calls and the service are accessed dynamically.length of call holding time. We define the metricc (i) to reflect the cost efficiency. c (i) is the ratio of e ethe cost earned within the observation time t overtotal input traffic intensity for service provider sp, isdefined as 2194 | P a g e
  10. 10. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197Fig. 5. Service performed in a dynamic mannerFig. 6 represents the active users during the trafficpattern prediction. The active users are increased dueto the decreased call blocking. Since the call arrivalrate is predicted and assigned the channels based onthe prediction of the call arrival rate and call holdingtime, traffic of the call arrivals are reduced. So that, Fig. 7. Call Arrival Rate VS Call blockingwe can increase the active users after predicting thetraffic of the primary user. As the call blocking decreases, then the occupied spectrum will be less and there is the possibility of getting more spectrum. Fig. 8 shows that the occupied band after calculating the free spectrum utilization, as the call blocking decreases. It depicts that for maximum service provider limited portion of a spectrum is occupied and it also shows that the free spectrum available for further spectrum utilization.Fig. 6. Call Arrival Rate VS Active usersFig. 7 gives the call blocking rate during the trafficprediction. It seems that the function of the offeredtraffic is greater than the given number of channels,and the arriving calls will be blocked. During thehigh traffic loads, the proposed work significantly Fig. 8. Call Arrival Rate VS Free Spectrumreduces the connection call blocking and increasesthe number of simultaneous transmission due to the As primary users (PU) have priorities to use theproper channel assignment strategy. spectrum when secondary users (SU) co-exist with them, the interference generated by the SU 2195 | P a g e
  11. 11. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197transmission needs to be below a tolerable threshold Fig. 11 shows the channel utilization during theof the PU system. Thus, to manage the interference traffic pattern prediction against the call arrival rate.to the PU system and the mutual interference among The figure shows that as the mean call arrivalSUs, power control schemes should be carefully increases, the channel utilization also increases. Asdesigned. As the number of call arrivals increases the call blocking rate is lower; thus, more calls canwith the maximum utilization of the channels, contribute to the spectrum utilization. Using this,Interference has been reduced during the heavy higher spectral efficiency has been evaluated.traffic load as illustrated in the Fig. 9.Fig. 9. Call Arrival Rate VS Interference Fig. 10 shows the system efficiency againstthe call arrival rate. When the traffic is high, thedropped calls increases with decreasing total Fig. 11. Call Arrival Rate VS Channel Utilizationprocessed calls. During the call arrival predictiondue to the high traffic loads, the system efficiency IX. CONCLUSIONincreases. This can be reduced, when the number of The proposed algorithm predicts the callcall arrival rate decreases. arrival rate of primary users and it gives an approach to predict the call holding time of the primary users. Predicting the probability of call arrival rate provides the secondary user regarding the channel availability to determine whether to use the channel or not. This proposed algorithm leads to the optimal transmission and observation time to maximize sensing efficiency satisfying the strict interference constraint of primary networks. This prediction enhances the channel utilization of primary users and enhances the communication of primary users. The simulated results show that there is a decrease in the call blocking and reduced interference. This helps to improve the channel utilization and avoids the scarce resource of the spectrum. ACKNOWLEDGEMENTS The First Author extends her gratitude to UGC as this work was supported in part by Basic Scientist Research (BSR) Non SAP Scheme, under grant reference no.11-142/2008 (BSR) UGC XI Plan.Fig. 10. Call Arrival Rate VS System Efficiency 2196 | P a g e
  12. 12. R. Kaniezhil, Dr. C. Chandrasekar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.2186-2197REFERENCES [18] http://www.mast.queensu.ca/~stat455/lecturenotes/set[1] C. Haykin, “Cognitive radio: brain-empowered 4.pdf wireless communications”, IEEE J. Selected Areas [19] http://www.scribd.com/doc/33529988/27/Large- Commun., vol. 23, Feb. 2005, pp. 201-220. Sample-Approximation-to-the-Best-Linear-Predictor[2] Kin K. Leung, Wiiliam A. Massey and Ward Whitt, [20] Ross Ihaka, “Time Series Analysis”, Lecture Notes “Traffic Models for Wireless Communication for 475.726, April 14, 2005. Available at Networks,” IEEE Journal on selected areas in http://www.stat.auckland.ac.nz/~ihaka/726/notes.pdf Communications, vol. 12, no.8, October 1994, pp [21] Cliford M. Hurvich and Chih-Ling Tsai, “selection 1353-1364. of a Multistep Linear Predictor for Short Time Series,[3] Abdelnaser Adas, “Traffic Models in Broadband Statistica Sinica 7, 1997, pp 395-406. Networks,” IEEE Communications Magazine, Vol [22] D. Willkomm, S. Machiraju, J. Bolot, and A. Wolisz, 35, no.7, July 1997, pp 82-89. “Primary users in cellular networks: a large-scale[4] Li X. & Zekavat S. A., “Traffic Pattern Prediction measurement study,” in IEEE DySPAN, Oct. 2008. and Performance Investigation for Cognitive Radio [23] R. Kaniezhil, Dr. C. Chandrasekar, “Performance Systems,” IEEE Communications Society, WCNC Analysis of Wireless Network with Opportunistic 2008 proceedings, pp. 894-899, March 2008. Spectrum Sharing via Cognitive Radio nodes,”[5] Marko H¨oyhty¨a, Sofie Pollin, and Aarne Journal of Electronic Science and Technology – to be M¨ammel¨a, “Improving the Performance of published. Cognitive Radios through Classification, Learning, [24] R. Kaniezhil, Dr. C. Chandrasekar, “ An Optimal and Predictive Channel Selection,” Advances in Channel Selection Strategy via Cognitive Radio Electronics And Telecommunications, vol. 2, no. 4, Nodes,” Journal of Sensor Letters, Vol. 10, 1–8, 2012 Dec. 2011, pp 28-38. – to be published.[6] G. Rutka, G. Lauks, “Study on Internet Traffic Prediction Models,” ISSN 1392 – 1215, Electronics R. Kaniezhil received the B.Sc. degree and Electrical Engineering, NO. 6(78), 2007, PP 47- from the University of Madras in 1998. 50. She received her MCA and M.Phil[7] Bidisha Ghosh, Biswajit Basu and Margaret Degrees from Periyar University and O‟Mahony, “A Bayesian Time-Series model for Annamalai University, in 2001 and short-term Traffic Flow Forecasting, “ pp 1-37. 2007, respectively. She is currently[8] Yantai Shu, Minfang Yu, Oliver Yang, Jiakun Liu and Huifang Feng, “Wireless Traffic Modeling and pursuing the Ph.D. degree with the Department of Prediction Using Seasonal ARIMA Models,” IEICE Computer Science, Periyar University, Salem, India. Trans. Communication, vol.e88–b, No.10, pp 3992- Her research interests include Mobile Computing, 3999, October 2005. Spectral Estimation, and Wireless Networking.[9] Chris Chatfield, “Time Series Forecasting,” ISBN 1- 58488-063-5 (alk. paper), 2000 by Chapman &C Hall/CRC.[10] Bo Zhou, Dan He, Zhili Sun and Wee Hock Ng, “Network Traffic Modeling and Prediction with Dr. C. Chandrasekar received his ARIMA/GARCH”. Ph.D degree from Periyar University. He is working[11] LI Zhi-Peng, U Hong, LIU Yun-Cai, LIU Fu-Qiang, as an Associate Professor, Department of Computer “An Improved Adaptive Exponential Smoothing Model for Short-term Travel Time Forecasting of Science, Periyar University, Salem. His areas of Urban Arterial Street,” ACTA AUTOMATICA interest include Wireless networking, Mobile SINICA, Vol. 34, No. 11, Nov. 2008, pp 1404 – Computing, Computer Communications and 1409. Networks. He is a research guide at various[12] Kohandani, F. , McAvoy, D.W. ; Khandani, A.K. , universities in India. He has published more than 80 “Wireless airtime traffic estimation using a state technical papers at various National & International space model” , Communication Networks and conferences and 50 journals. Services Research Conference, CNSR, May 2006, pp 251 – 258.[13] Guangquan Chen, Mei Song, Yong Zhang, Junde Song, Yinglei Teng, Li Wang, “ Fuzzy logic based traffic model considering cross-layer for future cognitive network “,Joint Conferences on Pervasive Computing (JCPC), Dec 2009, pp 627 – 632.[14] Wayne L. Winston, “Operations Research : Applications and Algorithms”, Chapter 20, Fourth Edition, ISBN-13:978-81-315-0190-0, 2003.[15] Hao Chen, Ljiljana Trajkovic, “Trunked Radio Systems: Traffic Prediction Based on User Clusters”, Wireless communication systems, pp 76-80, Sep 2004.[16] http://www.nas.its.tudelft.nl/people/Piet/CUPbookCh apters/PACUP_Poisson.pdf[17] http://www.cs.vu.nl/~rmeester/onderwijs/Poisson_pro cessen/poisson.pdf 2197 | P a g e

×